SEPA
United States
Environmental Protection
Agency
EPA/600/R-14/043 | September 2014
        www.epa.gov/ord
       LINKING NUTRIENTS TO ALTERATIONS
       IN AQUATIC LIFE IN CALIFORNIA
       WADEABLE STREAMS
      Office of Research and Development
      National Health and Environmental Effects Research Laboratory, Atlantic Ecology Division
                                                 	

-------

-------
                                                             EPA/600/R-14/043
                                                              September 2014
                                                             www.epa.gov/ord
Linking Nutrients to Alterations in Aquatic Life
           in California Wadeable Streams
             A. Elizabeth Fetscher, Martha Sutula, and Ashmita Sengupta
                Southern California Coastal Water Research Project
                             Naomi Detenbeck
            National Health and Environmental Effects Research Laboratory
                     U.S. Environmental Protection Agency
                          Atlantic Ecology Division
                         Narragansett, Rhode Island
                        U.S. Environmental Protection Agency
                        Office of Research and Development
                National Health and Environmental Effects Research Laboratory
                            Atlantic Ecology Division
                         Narragansett, Rhode Island 02882

-------
This is contribution number ORD-007337 of the Atlantic Ecology Division, National Health and Environmental
Effects Research Laboratory, Office of Research and Development, U.S. Environmental Protection Agency. The
information in this document has been funded wholly by the U.S. Environmental Protection Agency under
contract EP-12-D-000146 to the Southern California Coastal Water Research Program (SCCWRP). This document
has been reviewed by the U.S. Environmental Protection  Agency, Office  of Research and Development, and
approved for publication. This report was first reviewed through ORD's internal  peer  review  process and
approved for  publication. In accordance with  guidance in the US EPA's Peer Review Handbook for Influential
Science Information, the document was subsequently sent out for an independent, external peer review to five
subject matter experts with expertise in algal and macroinvertebrate ecology, nutrient effects on freshwater
streams, and  nutrient criteria development. The document was revised based on reviewer recommendations.
Mention of trade names or commercial products does not constitute endorsement or recommendation for use.

-------
List of Tables	v
List of Figures	vii
List of Acronyms and Abbreviations	xi
Acknowledgements	xiii
Executive Summary	xiv
1. Introduction	1
    1.1 Introduction and Project Objectives	1
    1.2 Document Organization	2
    1.3 Literature Cited	3
2. Estimation of Reference and Ambient Concentrations of Stream Eutrophication Indicators	5
    2.1 Introduction	5
    2.2 Methods	5
       2.2.1 Approach	5
       2.2.2 Data Sources, Site Selection, and Stream Sampling Protocol	5
       2.2.3 Distribution of Wadeable Stream Nutrients and Primary Producer Inldicator Values	12
    2.3 Results	13
       2.3.1 Distribution of Nutrients and Primary Producer Indicators at Reference Sites	13
       2.3.2 Ambient Distribution of Nutrients and Primary Producer Abundance Indicator	14
    2.4 Discussion	19
    2.5 Literature Cited	20
3. Thresholds of Adverse Effects of Primary Producer Abundance and Nutrients on Wadeable Stream
    Aquatic Life Indicators	21
    3.1 Introduction	21
    3.2 Methods	23
       3.2.1 Conceptual Approach	23
       3.2.2 Aquatic Life and Stressor Indicators	25
       3.2.3 Detection of Ecological Thresholds	30
       3.2.4 Data Sources	31
       3.2.5 Data Analyses	33
    3.3 Results	40
       3.3.1 BMI and Diatom Responses to Biomass and Nutrient Gradients Based on Shifts in
       Community Composition	40
       3.3.2 Biotic Responses to Biomass Gradients Based on Shifts in Integrative Measures of
       Community Composition (Metrics and Indices)	50
       3.3.3 Examining Relative Influence of Biomass, Nutrients, and Other Factors on Integrative
       All Measures	63
       3.3.4 Thresholds for Biomass and Nutrient Effects on Biotic Response	75

-------
    3.4 Discussion	77
       3.4.1 Statistically Detected Thresholds of Adverse Effect in California Wadeable Streams	78
       3.4.2 Variable Response of All Types to Biomass and Nutrients: The Biological Condition
       Gradient	84
       3.4.3 Study Findings in Context of Policy Applications	85
    3.5 Literature Cited	85
4. Evaluation of Nutrient Numeric Endpoint Benthic Biomass Spreadsheet Tool	91
    4.1 Introduction	91
    4.2 Methods	92
       4.2.1 Conceptual Approach to Validation and Error Analysis	92
       4.2.2 Background on Benthic Biomass Spreadsheet Tool Development and Testing
       (Objective 1)	92
       4.2.3 Data Sources	97
       4.2.4 Validation of the NNEBBST Tool	98
       4.2.5 Bias Analysis (Objective 2)	98
       4.2.6 Exploratory Analyses Using Boosted Regression Trees (Objective 3) and Bayesian
       CART Analysis (Objective 4)	99
    4.3 Results	102
       4.3.1 Model Performance (Objective 1)	102
       4.3.2 Random Forest Regression for Bias and Variance Analysis (Objective 2)	105
       4.3.3 Results of BRT Analyses (Objective 3)	108
       4.3.4 Results of Bayesian CART Analyses (Objective 4)	115
    4.4 Discussion	122
       4.4.1 Validation Exercise Shows Considerable Room for Improvement in BBST	122
       4.4.2 Inclusion of Landscape and Site-Specific Factors Provide Avenue for Model
       Refinement	123
       4.4.3 Summary of Validation and Recommendations for Refining Wadeable Stream
       Nutrient Algal Abundance Models	126
    4.5 Literature Cited	126

Appendix A. Important Definitions	128
Appendix B. Graphics and Tables Supporting Analyses of Reference and Ambient Concentrations
    of Stream Eutrophication Indicator	131
Appendix C. Graphics and Tables Supporting Analyses of Thresholds of Adverse Effects of
    Primary Producer Biomass and Nutrient on Wadeable Stream Aquatic Life	141
Appendix D. Graphics and Tables Supporting Evaluation of Benthic Biomass Response Models.
Appendix E. Quality Assurance/Quality Control Summary	193
Literature Cited	199
                                                  IV

-------
Table 2.1. Metric descriptions and codes for stream primary producer abundance indicators	10
Table 2.2. Variables used for assigning sites to "site disturbance classes" per the state's bio-objectives
    process (adapted from Ode et al., under review)	11
Table 2.3. Sources of data for landscape, meteorological, and geological explanatory variables used
    in predictive models	12
Table 2.4. Median,  75th, and 95th percentiles of raw (unweighted) TN, TP benthic chlorophyll a, AFDM,
    and macroalgal percent cover (PCT_MAP), statewide and by region, at reference sites (both
    probability and targeted datasets included)	14
Table 2.5. Extent estimates for the site-evaluation categories based on  reconnaissance information
    across the PSA  and SMC probability surveys from 2008-2011	15
Table 2.6. Statewide estimates for distributional properties of primary producer abundance indicator
    values in California perennial, wadeable streams	15
Table 2.7. Estimated median values (with 95% confidence intervals) for selected ambient stream
    nutrient and primary producer abundance indicators statewide and by region	17
Table 2.8 Percent of perennial wadeable stream kilometers exceeding 75th and 95th percentiles of
    statewide or regional  reference values for nutrient and primary producer abundance gradients	17
Table 3.1. Description of aquatic life indicators (Alls) used in the analyses	26
Table 3.2 Descriptions of the focal stressor gradients and other explanatory variables used in the
    analyses, listed in the  table alphabetically according to the shorthand version of the name, within
    their respective categories	28
Table 3.3. Sample sizes	32
Table 3.4. Summary of analytical techniques used for threshold  estimation	34
Table 3.5. Results of CART analyses with NMS axis 1 scores for either the BMI or the diatom community
    as the response variable	43
Table 3.6. TITAN and nCPA results for BMI and diatom community composition data	45
Table 3.7. Summary of piecewise regression results for all All response types for which at least one
    version of the analysis (weighted or unweighted) fulfilled all four "strict" criteria,
    as described in the Methods	51
Table 3.8. Summary of boosted regression tree models of All variables, and relative influence (and rank)
    of biomass and nutrient predictors used in each	64
Table 3.9. Relative  influence of predictors from BRT models	65
Table 3.10. Partial Mantel coefficients (95% CIs) for correlation between biomass/nutrient predictors
    and All variables and  p-values	69
Table 3.11. Quantitatively determined thresholds of stream (or  river) All responses to
    nutrient concentrations	83
Table 4.1. Summary of types of models contained in BBST	93

-------
Table 4.2. Stream scouring cutoff values for the watersheds developed based on precipitation and
    watershed imperviousness	98
Table 4.3. Final set of classification variables used in B-CART analysis	102
Table 4.4. Distribution of residuals of >0.5 standard deviation of mean for chlorophyll a by model type	103
Table 4.5. Model performance (R2, slope and intercept) for all sites combined	103
Table 4.6. Model performance (R2) for all, Reference, Intermediate, and Stressed sites (see Chapter 2)
    for predicted mean chlorophyll a	104
Table 4.7. Variables ranked according to importance for random forest regression for Chlorophyll a
    (Chi o) and AFDM by model type	106
Table 4.8 Relative influence of nutrient species on abundance of stream biomass of six different types,
    from BRT models that included environmental co-factors	109
Table 4.9. Partial Mantel coefficients (95% CIs) for correlation between nutrient predictors and biomass
    variables; p values	Ill
Table 4.10. Results of B-CART analyses based on reduced sets of classification variables	116
Table 4.11. Variables retained in regression  analyses to predict benthic biomass (loglO chlorophyll o)
    based on Dodds-type models for nodes  in B-CART models 1 and 3, and based on QUAL2K-type
    models for nodes in B-CART models 2 and 4	120
Table A.I Definition of beneficial uses applicable to freshwater habitat	130
Table B.I. Number of sites within each Level III ecoregion (Omernik, 1987) in the South Coast,
    by site disturbance class	139
Table C.I. TITAN change point values for BMI and diatom taxa ("pure" and "reliable")	144
Table C.2. Results of piecewise regressions for all analyses in which "relaxed" criteria were met	158
Table C.3. Shown are 1) a summary of thresholds, from all analyses, for the chlorophyll a, AFDM, TN,
    and, TP gradients, and 2)  mean distributions of All values among sites that had gradient values
    below and above the indicated threshold	166
Table C.4. Summary of recommended numeric endpoints for stream NNE indicators,
    by beneficial use, from Tetra Tech (2006)	178
Table D.I. Details on the models	187
Table D.2. Results of Bayesian CART analysis of full data  set using all potential classification variables	189
Table D.3. Frequency of inclusion of classification variables in Bayesian CART TNTP models
    (Training sets 1-10)	189
Table D.4. Results of Bayesian CART analysis of full data  that includes PSA ecoregions	190
Table D.5. Results of Bayesian CART analysis of full data  set that uses  an empirical rather than
    the  PSA ecoregions	190
Table E.I. Nutrient fractions for samples from 47 USGS NAWQA stream stations in California
    sampled biweekly over the year	194
                                                  VI

-------
Figure 2.1. All algae sampling sites (probability and targeted) included in this report, shown by the
   Perennial Stream Assessment (PSA) ecoregion in which they occur	
Figure 2.2. Statewide CDFs for biomass measures and macroalgal percent cover (attached and/or
   unattached combined) by site disturbance class	16
Figure 2.3. CDFs for biomass measures and macroalgal percent cover (attached and/or unattached
   combined), broken down by PSAS ecoregion	18
Figure 3.1. Simplified conceptual model of eutrophication in wadeable streams depicting the
   relationship between nutrients, stream co-factors, ecological response and ecosystem services	22
Figure 3.2. Conceptual model depicting stages of change in biological conditions in response to an
   increasing stressor gradient. Reproduced from Davies and Jackson (2006)	24
Figure 3.3. Examples of two statistical approaches used to derive quantitative water quality goals	25
Figure 3.4. Examples of types of threshold relationships	30
Figure 3.5. All response to a stressor gradient showing the "reference envelope" along with "resistance"
   and "exhaustion" thresholds	31
Figure 3.6. Scatterplots and splines for non-metric multidimensional scaling (NMS) axis 1 values from the
   benthic macroinvertebrate (BMI) community against biomass and selected cover and nutrient
   gradients on  log scale, using the statewide data set	41
Figure 3.7. Scatterplots and splines for NMS axis 1 values from the diatom community against biomass
   and selected  cover and nutrient gradients on log scale, using the statewide data set	42
Figure 3.8. Cut points from CART analyses using NMS axis 1 scores from either the BMI or the diatom
   community as the response variable	44
Figure 3.9. Plots of "sum(z)"  scores (depicted as dots) from TITAN analysis of BMI community data along
   chlorophyll a, AFDM, and TN gradients, and the cumulative threshold frequency graphs (depicted as
   lines) for the sum(z) scores	47
Figure 3.10. Plots  of taxon-specific change from TITAN analysis of diatom community data along AFDM
   and TP gradients	48
Figure 3.11. Nonparametric change point analysis (nCPA) results	49
Figure 3.12. Summary of TITAN  and nCPA change points along biomass/nutrient gradients, based on BMI
   and diatom community composition using the statewide dataset	50
Figure 3.13. Breakpoints, with 95% confidence intervals, for the chlorophyll a  gradient, from piecewise
   regressions using all available All data types	53
Figure 3.14. Breakpoints, with 95% confidence intervals, for the AFDM gradient, from piecewise
   regressions using all available All data types	54
Figure 3.15. Breakpoints, with 95% confidence intervals, for the TN gradient, from piecewise regressions	55

                                                vii

-------
Figure 3.16. Breakpoints, with 95% confidence intervals, for the TP gradient, from piecewise
   regressions	56
Figure 3.17. Piecewise regression plot of diatom All variable, RAWIowP on a TP gradient (top) and SiZer
   map from analysis of the same two variables (bottom)	57
Figure 3.18. Piecewise regression plot of BMI All variable, lntolerant_PercentTaxa on a TN gradient (top)
   and SiZer map from analysis of the same two variables (bottom)	58
Figure 3.19. Piecewise regression plot of BMI All variable, Taxonomic_Richness on a chlorophyll a
   gradient (top) and SiZer map from analysis of the same two variables (bottom)	59
Figure 3.20. Piecewise regression plot of BMI All variable, lntolerant_PercentTaxa on an AFDM gradient
   (top) and SiZer map from analysis of the same two variables (bottom)	60
Figure 3.21. Closeup of piecewise regression plot of hybrid All variable, the IBI H20 on an AFDM gradient
   (top) and SiZer map from analysis of the same two variables (bottom)	61
Figure 3.22. Distribution of All values among sites with stressor gradient (i.e., biomass or nutrient
   concentration) values below vs. above the threshold that had been determined based on piecewise
   regression	62
Figure 3.23. Heat map showing relative influence (%) of predictor variables (biomass, nutrients, and
   environmental co-factors) on All response variables, from 14 independent BRT models.
   Yellow = low influence, red = high	67
Figure 3.24. Summary of the relative influence of biomass and nutrient predictors on Alls,
   from the BRT models	68
Figure 3.25. Partial dependence plots of chlorophyll a from BRT models predicting three All response
   types: CSCI, Taxonomic Richness, and RAWNhet	72
Figure 3.26. Partial dependence plots of AFDM from BRT models predicting four All  response types: the
   metrics EPT_Percent and RAWDO100; and the IBIs H20 and D18	73
Figure 3.27. Partial dependence plots of TN from BRT models predicting three All response types: the
   soft algae IBI, S2; and the BMI Alls, Taxonomic Richness and Intolerant Percent Taxa	74
Figure 3.28. Partial dependence plots of TP from BRT models predicting three All response types: the
   soft-algae All, RAWmeanZHR; and the IBIs,  D18 and H20	75
Figure 3.29. Summary of results across analyses using the chlorophyll a, AFDM, TN, and TP gradients,
   stratified by assemblage type	76
Figure 3.30. Ranges of thresholds of All response by "All category" (as described in section 3.2.2)
   for two biomass and two nutrient gradients	77
Figure 4.1. Example of user interface for the BBST (example highlights output plot for the Dodds 1997
   version of the model) with input and output panels	94
Figure 4.2. Example of user interface for BBST Standard QUAL2K model with input and output panels	96
                                               VIII

-------
Figure 4.3. Sample plots of validation data showing measured versus predicted chlorophyll a by standard
   QUAL2K model, with 1:1 slope lines	104
Figure 4.4. Relative influence of variables for some selected models for chlorophyll a (top row: a, b, c)
   and AFDM (bottom row: d, e, f)	107
Figure 4.5. Heat map showing relative influence (%) of predictor variables (nutrients and environmental
   co-factors) on biomass response variables, from six independent BRT models	120
Figure 4.6. Three-dimensional plot (two views) of NOx and mean monthly maximum ambient air
   temperature (the mean of the month the sample was collected and the two months prior) from BRT
   model for chlorophyll a	113
Figure 4.7. Partial dependence plots of percent fine substrata (left) and percent canopy cover (right)
   from the BRT model for AFDM	113
Figure 4.8. Three-dimensional plot of percent canopy cover (XCDENMID) and conductivity from a BRT
   model for PCMVIAP (percent macroalgal cover)	114
Figure 4.9. Three-dimensional plot of percent sand + fine substrata and days of accrual from BRT model
   for PCT_MCP (percent macrophyte cover)	115
Figure 4.10. Bayesian regression tree for Model 1	117
Figure 4.11. Bayesian regression tree for Model 2	117
Figure 4.12. Location of sampling sites corresponding to nodes in Bayesian CART Model 3	118
Figure 4.13. Sampling station locations corresponding to final nodes in Bayesian CART Model 4	119
Figure B.I. Histograms of biomass and algal/macrophyte cover data, all California probability data
   combined	131
Figure B.2. Boxplots of biomass, ash-free dry mass, and macroalgal percent cover	134
Figure B.3. Cumulative distribution functions of biomass, ash-free dry mass, and macroalgal
   percent cover, by region, for all probability sites	137
Figure B.4. CDFs for benthic chlorophyll a, for the "xeric" and "mountain" Level III ecoregions
     (Omernik 1987) within the South Coast	139
Figure B.5. Within-ecoregion estimated percent of stream kilometers lower than the lowest proposed NNE
   endpoint for chlorophyll a (100 mg m~2), by site disturbance class	140
Figure C.I. Examples of plots of TITAN sum(z) scores fortaxa that decrease in frequency along the gradient
   of interest (in black) and those that increase (in red)	141
Figure C.2 Examples of plots of TITAN change points for individual taxa	142
Figure C.3. Examples of a SIZer map	143
Figure D.I. Comparison of the NNE stations (3053) to the daily precipitation station from  NOAA
   (981, July 2012)	179
                                                IX

-------
Figure D.2. Comparison of the predicted precipitation between the PRISM and ARCGIS
   (predicted for this study)  	180
Figure D.3. Ranking by variable importance for the Dodds 97 model for AFDM using random forests	181
Figure D.4. Ranking by variable importance for the Dodds 02 model for AFDM using random forests	182
Figure D.5. Ranking by variable importance for the QUAL2K model for AFDM using random forests	183
Figure D.6. Ranking by variable importance for the Dodds 97 for Chlorophyll a using random forests	184
Figure D.7. Ranking by variable importance for the Dodds 02 for Chlorophyll a using random forests	185
Figure D.8. Ranking by variable importance for the QUAL2K models for Chlorophyll a 186
   using random forests	186
Figure D.9. Predicted versus observed normalized loglO chlorophyll a biomass (mg/m2) for 1) TNTP
   training set, b) TNTP test set, c) DINDIP training set, and d) DINDIP test sets used in Bayesian CART
   analysis (Dodds-type model, all potential classifiers)	191
Figure E.I. Frequency of a month of a) maximum annual total N and b) maximum annual total P in 47
   California NAWQA streams	195
Figure E.2. Value for a) total N and b) total P by month relative to maximum monthly values in 47
   California NAWQA streams	196
Figure E.3 Relationship between annual maximum and growing season average values for a) total N and
   b) total P in 47 California NAWQA streams	197
Figure E.4. Chlorophyll a levels (log-transformed) across sampling dates, by year, for South coast (blue)
   and all other sites (red) within the state	198
Figure E.5. Chlorophyll a levels (log-transformed) across sampling dates normalized by log total N,
   by year, for South Coast (blue) and all other sites (red) within the state	199

-------
Acronym or abbreviation   Definition
AFDM
AG_2000_1K
AG_2000_5K
AG_2000_WS
All
BCG
BEST
BMI
BRT
BURC
CART
CDF
CODE21_2000_1K
CODE21_2000_5K
CODE21_2000_WS
COLD
COMM
CPOM

CSCI

DO
D18
EMAP
EPT_Percent
EPT_PercentTaxa
EPT_Taxa
EQ
FPOM
GHCND
GRTS
H20
H21
H23
IBI
lntolerant_Percent
lntolerant_PercentTaxa
lntolerant_Taxa
LRWQCB
MIGR
MRDS
MSE
MUN
N
N:P
nCPA
NH4
NMS
NNE
NOAA
NOX
NPS
OBEM
0/E
P
Ash-free dry mass
Percent agricultural land use in catchment within a 1-km radius from sampling site
Percent agricultural land use in catchment within a 5-km radius from sampling site
Percent agricultural land use in catchment
Aquatic life indicator
Biological condition gradient
Benthic biomass spreadsheet tool (Tetra Tech 2006)
Benthic macroinvertebrate
Boosted regression tree analysis
Beneficial use risk category
Classification and regression tree analysis
Cumulative distribution function
Percent "Code 21" land use in catchment within a 1-km radius from sampling site
Percent "Code 21" land use in catchment within a 5-km radius from sampling site
Percent "Code 21" land use in catchment
Cold freshwater habitat beneficial use
Commercial and sport fishing beneficial use
Coarse particulate organic matter
California Stream Condition Index (the  BMI-based statewide index for stream bioassessment; Mazor
  et al., under review)
Dissolved oxygen
Diatom Index  of Biotic Integrity (IBI); Fetscher et al. 2014
Environmental Monitoring and Assessment Program
Percent BMI individuals that are Ephemeroptera, Plecoptera, or Trichoptera
Percent BMI taxa that are Ephemeroptera, Plecoptera, or Trichoptera
Number of BMI taxa that are Ephemeroptera, Plecoptera, or Trichoptera
Equation
Fine particulate organic matter
Daily global historical climatology network
Generalized random tessellation stratified  method for creating a spatially balanced probability survey
Diatom + soft algae ("hybrid") index of  biotic integrity (IBI);  Fetscher et al. 2014
Diatom + soft algae ("hybrid") index of  biotic integrity (IBI);  Fetscher et al. 2014
Diatom + soft algae ("hybrid") index of  biotic integrity (IBI);  Fetscher et al. 2014
Index of biotic integrity
Percent BMI individuals that are "intolerant"; Ode et al. 2005
Percent BMI taxa that are "intolerant";  Ode et al. 2005
Number of BMI taxa that are "intolerant"; Ode et al. 2005
Lahontan Regional Water Quality Control Board
Migration of aquatic organisms beneficial use
Mineral Resources Data System
Mean square error
Municipal (beneficial use)
Nitrogen
Nitrogen-to-phosphorus ratio
Nonparametric change point analysis
Ammonium
Nonmetric multidimensional scaling analysis
Nutrient numeric endpoint
National Oceanic and Atmospheric Agency
Nitrate + nitrite
Non-point source
Out of  bag error method
Observed over expected taxa from RIVPACS models for BMI taxa; Mazor et al., under review
Phosphorus
                                                            XI

-------
PCT_CPOM
PCT_FN
PCT_MAP
PCT_MCP
PCT_MIAT1
PCT_SAFN
PHab
PRISM
propAchMin

propTaxaZHR

PSA
QUAL2K
RARE
RAWD0100
RAWD050
RAWeutro
RAWIowN
RAWIowP
RAWIowTPsp
RAWmeanZHR
RAWNhet
RAWpropBiovolChlor

RAWpropBiovolZHR

RAWpropGreenCRUS

RCMP
REC-1
REC-2
RIVPACS
S2
Shannon_Diversity
Simpson_Diversity
SiZer
SMC
SPWN
SRP
SWAMP
SWRCB
Taxonomic_Richness
TITAN
TN
Tolerant_Percent
Tolerant_PercentTaxa
Tolerant_Taxa
TP
URBAN_2000_1K
URBAN_2000_5K
URBAN_2000_WS
USEPA
USGS
W1_HALL
WARM
WILD
WQO
XDENMID
Percent cover of coarse particulate organic matter in streambed
Percent cover of fine substrata in streambed
Macroalgal percent cover
Macrophyte percent cover
Percent presence of thick (lmm+) microalgae
Percent sand + fines in streambed
Physical habitat (measures collected in a stream reach)
Parameter-elevation relationships on independent slopes model
Proportion of diatom valves that are Achnanthidium minutissimum
Proportion of total of total soft-algae taxa recorded that are in the Zygnemataceae, heterocystous
  cyanobacteria, or Rhodophyta
Perennial stream assessment
A river and stream water quality model
Rare, threatened, or endangered species beneficial use
Proportion diatoms requiring nearly 100% DO saturation; van Dam et al. 1994
Proportion diatoms requiring at least 50% DO saturation; van Dam et al. 1994
Proportion eutrophication indicator diatoms; van  Dam et al. 1994
Proportion low-N indicator diatoms; Potapova and Charles 2007
Proportion low-P indicator diatoms; Potapova and Charles 2007
Proportion of soft algal taxa that are considered "low TP" indicators; Fetscher et al. 2014
Mean of the metrics propTaxaZHR and RAWpropBiovolZHR; Fetscher et al. 2014
Proportion nitrogen-heterotroph diatoms; van Dam et al. 1994
Proportion of total soft algae biovolume that is Chlorophyta
Proportion of total soft algae biovolume that is in  the Zygnemataceae, heterocystous cyanobacteria,
  or Rhodophyta
Proportion of green algal biovolume belonging to  Cladophora glomerata, Rhizoclonium
  hieroglyph/cum, Ulvaflexuosa, or Stigeoclonium species
California's Reference Condition Management Program
Contact water recreation beneficial use
Non-contact water recreation beneficial use
River Invertebrate Prediction and Classification  System
Soft algae index of biotic integrity (IBI);  Fetscher et al. 2014
Shannon Diversity Index for BMI taxa
Simpson Diversity Index for BMI taxa
Significant zero crossings analysis
Stormwater Monitoring Coalition
Spawning, reproduction, and/or early development beneficial  use
Soluble reactive phosphorus
California State Water Resources Control Board Surface Water Ambient Monitoring Program
State Water Resources Control Board
Richness of BMI taxa
Threshold indicator taxa analysis
Total nitrogen
Percent BMI individuals that are "tolerant"; Ode et al. 2005
Percent BMI taxa that are "tolerant"; Ode et al. 2005
Number of BMI taxa that are "tolerant"; Ode et al. 2005
Total phosphorus
Percent urban land use in catchment within a 1-km radius from sampling site
Percent urban land use in catchment within a 5-km radius from sampling site
Percent urban land use in catchment
U.S. Environmental Protection Agency
U.S. Geological Survey
A riparian disturbance index; Kaufmann et al. 1999
Warm freshwater habitat beneficial use
Wildlife habitat beneficial use
Water Quality Objective
Percent canopy cover
                                                           XII

-------

The California Surface Water Ambient Monitoring Program (SWAMP) funded Reference Condition Management
Program and Perennial Stream Assessment, and the Southern California Stormwater Monitoring Coalition (SMC)
provided data. Taxonomic identifications for biotic samples were carried out for the aforementioned monitoring
programs  by the following  individuals: Joseph Slusark,  Dan Pickard, and  Austin  Richards of the California
Department of Fish  and Wildlife Aquatic Bioassessment  Laboratory (for  benthic  macroinvertebrate  data);
Rosalina Stancheva, Robert Sheath, and Christina Fuller of California State University San Marcos (for soft-algae
data); and Patrick Kociolek, Evan Thomas, and Carrie Graeff of the University of Colorado at Boulder (for diatom
data). The draft  CSCI was  developed  by  Raphael Mazor in collaboration  with  members of the California
Biological-Objectives  Technical  Team  and  Science  Advisory  Panel.  Mark  Engeln,  Abel  Santana,  Rebecca
Schaffner, Marco Sigala, Carly Beck and Bryan White assisted with data management and literature review. The
document benefited  from  review comments  on an earlier version  by EPA staff (Nathan Smucker,  Karen
Blocksom, Lester Yuan, Glen Thursby,  and Wayne Munns), SCCW staff (Eric  Stein), San Diego Regional Water
Control  Board  (Lilian  Busse), and Tetra Tech (Michael Paul, Jon Butcher), and valuable feedback on the revised
version  from Walter  Dodds, Michelle Evans-White (University of Arkansas),  R. Jan Stevenson  (Michigan State
University, SCCWRP  (Stephen Weisberg, Kenneth Schiff), Los Angeles County Sanitation District (Anne Heil,
George Gallis, Joshua  West), and two anonymous reviewers.
                                                 XIII

-------
In wadeable streams,  nutrient enrichment,  in concert with  other  site-specific factors, can  result in the
overabundance of algal biomass, low dissolved oxygen and altered biotic communities. These changes can have
adverse effects on stream ecosystem services. Scientifically-based water quality objectives (WQO) and tools that
relate these objectives  to  nutrient management are needed  in  California to prevent eutrophication  from
occurring and to provide targets to restore waterbodies where adverse effects have already occurred.

The California State Water Resources Control Board (SWRCB) is developing nutrient water quality objectives for
the State's surface waters.  USEPA guidance on  nutrient objective development generally recommends three
means to  set nutrient objectives (USEPA 2000): 1) a  reference approach, based on a statistical percentile of
nutrient or  biotic  response indicators in  minimally-disturbed  waterbodies; 2) an  empirical stress-response
approach, based on statistical analyses of field data on nutrients, algal abundance and indicators of aquatic life;
or 3) a process-based approach, involving identification of ecological responses of concern and mechanistically
modeling the linkage back to nutrient loads and other co-factors controlling response.

Among the  approaches that the SWRCB staff is  considering is the process-based approach, known as the
Nutrient Numeric Endpoint (NNE) framework  (Tetra Tech 2006). The NNE framework is intended to  serve as
numeric guidance to translate narrative WQO.  It consists of two tenets:  1) assessment  and recommended
numeric (regulatory)  endpoints based  on the  ecological  response  of an aquatic waterbody to eutrophication
(e.g., algal abundance, dissolved oxygen  [DO]) to assess waterbody condition and 2) scoping-level models that
link the response  indicator endpoints to nutrient inputs and  other  site-specific factors and management
controls. These scoping models were intended to be used to establish default nutrient targets for point source
discharge  and municipal stormwater permits  and total  maximum daily loads (Tetra Tech 2006). Tetra  Tech
(2006) developed the benthic biomass spreadsheet tool (BBST) for use in streams. As the SWRCB  prepares to
propose nutrient objectives for wadeable streams, scientific analyses of improved data from California statewide
stream probabilistic and targeted bioassessment surveys can strengthen the scientific basis for policy decisions.
In the context of this study,  "endpoints" refer to policy decisions on levels at which  point management action
should be taken; "thresholds" refer to the output of scientific analyses.

The objectives of this project are three-fold:

    •  Estimate the natural background and ambient concentrations of nutrients and candidate indicators of
      primary producer abundance in California wadeable streams;
    •  Explore relationships and identify thresholds of adverse effects of nutrient concentrations and primary
      producer abundance on aquatic life indicators in California wadeable streams;
    •  Evaluate the Benthic Biomass Spreadsheet Tool for California wadeable streams using existing data
      sets and recommend avenues for refinement.
The intended outcome of this study is research, NOT recommendations for regulatory endpoints for nutrient
and response indicators for California wadeable streams. The findings of this research study, as well  as other
analyses, may be used as lines of evidence considered to support SWRCB policy decisions on nutrient objectives
for wadeable streams.
                                                 xiv

-------
The majority of the State's Wadeable Streams sampled are below the 75th percentile of minimally disturbed
"reference sites." California's perennial, wadeable streams, as assessed during the bioassessment index period
of late spring through mid-summer, exhibited a skew toward the low end of the primary producer abundance
gradient. Nearly 66% of perennial wadeable stream kilometers had estimated benthic chlorophyll a and 59% had
estimated TN and TP values below the 75th percentile of each variable at reference sites statewide1. Among the
regions, a gradient in algal abundance and nutrient concentrations was observed from high in areas developed
by urban and agricultural land uses (South Coast,  Central Valley) to low in areas  of the state with lower density
development (e.g., North Coast and Sierra regions).

Statistically detectable  thresholds were found  for benthic chlorophyll a, ash-free dry mass (AFDM), and
nutrients; benthic chlorophyll a thresholds were below those of TetraTech (2006). This study found statistically
significant relationships and thresholds  of adverse effects  of benthic chlorophyll a, AFDM, and  TN and TP
concentrations on indicators of benthic macroinvertebrate (BMI) and algal community structure—employed in
this study as indicators of aquatic life. Integrative aquatic life indicators (Alls) such as indices of biotic integrity
corresponded to higher thresholds whereas All  measures specific to constrained groups of "sensitive" taxa
generally corresponded  to lower thresholds, illustrative of the paradigm of the biological condition gradient.
Most  of these thresholds of effect exceeded the 75th percentile of these indicators among reference stream
reaches statewide,  but  they were often less than the 95th  percentile.  The range  of  benthic chlorophyll a
thresholds in this study  were generally substantially below the current NNE endpoints protective of beneficial
uses recommended by TetraTech (2006;  100 and  150 mg/m2 chlorophyll a for cold [salmonid] and warm water
respectively). However,  it should be noted that our results are based on instantaneous measurement at low-
flow conditions, and as  such, do not reflect year-long loads or storm  flows. It is not clear to what degree the
types  of ALI-stressor relationships we observed would hold during rain events.

Validation exercise indicates that there is considerable room for improvement in BBST; inclusion of landscape
and site-scale factors provide  avenue for model refinement. The BBST models  show poor fit,  particularly
among "stressed" sites (one-third of the data set), when validated against a statewide dataset, which contains
benthic chlorophyll  a  data as currently measured in California ambient monitoring programs. The  poor fit is
understandable, given that the BBST was optimized for North American temperate streams and that the model
predicts maximum algal abundance, a value not verifiably captured  during the period in which  sampling to
generate the project data set occurred. Several  landscape- and site-scale explanatory variables were high in
their relative influence in the BBST model predicted-observed variance analysis and in preliminary nutrient-algal
response  models.  Nutrient  concentrations were  important  predictors  in BBST  model  predicted-observed
variance analysis and  boosted regression tree (BRT) models, albeit occupying less prominent roles than other
factors, such as temperature and stream substratum type. This finding validates the fundamental NNE approach:
site-specific co-factors that vary across the California landscape can influence algal response to nutrients. It also
suggests that model refinements are possible; inclusion of these site- and landscape-scale explanatory variables
in preliminary nutrient-algal response models substantially improved model fit over existing BBST models.
1 The analogous values, if considering the 95th percentile of Reference sites, are 90% of stream kilometers for Chlorophyll a
 and approximately 78% for nutrients.
                                                  xv

-------
• Statistical  analyses  for threshold detection were conducted  on statewide scale;  resources  were
  insufficient to look at the question of whether  there is scientific evidence for regionalization of
  thresholds. Additional analyses are recommended to look specifically at this question.
• An alternative approach to establish levels protective of All is to use predictive regression models to
  estimate concentrations of nutrients or algal abundance that are linked to a quantitative All target.
  We recommend such analyses based on the benthic invertebrate and stream algal IBI.
• A comprehensive effort to develop nutrient-algal abundance models for wadeable streams should be
  undertaken, considering a full  range of predictive and probabilistic statistical models. The compiled
  dataset now includes a variety of explanatory variables that are available to begin a more thorough set
  of analyses. More than one model categorized by classes may be necessary in order to  capture the
  range of nutrient-response relationships  statewide. More  complex mechanistic models could be
  considered over  the long-term if the need to offer greater flexibility and applications to  site-specific
  waterbody assessment are warranted.
                                              XVI

-------
I.I                and
Eutrophication1 of water resources  is a major environmental issue in California, with demonstrated links
among anthropogenic changes in watersheds, increased nutrient loading, harmful algal blooms, hypoxia, and
impacts on aquatic food webs. In wadeable streams, elevated nutrient concentrations, in concert with other
site-specific factors, can result in the overabundance  of algal biomass, low dissolved  oxygen and altered
biotic communities, with  a  suite of adverse effects  on  stream ecosystem  services and  beneficial uses
(Appendix A, Table A.I). High algal abundance can alter hydrology and interfere with spawning, foraging, and
shelter  (Biggs  2000,  Quinn and  Mickey 1990), limit the growth  of benthic  diatoms  as food sources  for
scraper/grazers (Steinman 1996), and deteriorate water quality (Quinn and Gilliland 1989). Wadeable stream
algal blooms  can also negatively impact human  health and other ecosystem services  or beneficial uses,
through toxin-forming harmful algal  blooms, proliferation  of pathogenic bacteria, taste/odor problems in
municipal drinking water supplies and compromised aesthetics (Biggs 2000, Lembi 2003, Suplee et al. 2009,
Fovet et al. 2012). In California, examples of eutrophication in wadeable streams have been well-documented
(e.g., Southern California, Mazor et al. 2014). Scientifically-based water quality objectives and tools that
relate these objectives to management controls are needed to prevent eutrophication from occurring and to
provide targets to restore waterbodies where adverse effects have  already occurred.

USEPA  guidance on nutrient objective development  generally recommends  three  means to  set nutrient
objectives  (USEPA 2000):  1)  a  reference approach, 2) an empirical stress-response approach, and 3) a
mechanistic, process-based approach. The reference waterbody approach involves  characterization of the
distributions of nutrients in "minimally disturbed" waterbodies. Nutrient concentrations are chosen at some
statistical  percentile  of those reference waterbodies. The  empirical stress-response  approach involves
establishing statistical  relationships between the causal or stressor variable (in  this  case  nutrient
concentrations or loads) and the ecological response (changes in algal or aquatic plant biomass or community
structure, changes in  sediment  or  water chemistry  such as dissolved oxygen, pH). The  process-based
approach involves identifying the ecological responses  of concern and mechanistically modeling the linkage
back to nutrient loads and other co-factors controlling response (e.g., hydrology, grazers, denitrification,
etc.).

The California  SWRCB is developing nutrient water quality objectives for the State's  surface waters. Among
the approaches that SWRCB staff is considering is a process-based  approach, known as the Nutrient Numeric
Endpoint (NNE) framework (Tetra Tech 2006). The NNE framework, intended to serve as numeric guidance to
translate narrative WQO, consists of two tenets: 1) numeric (regulatory) endpoints based on the ecological
response of an aquatic waterbody to eutrophication (e.g., algal abundance, dissolved oxygen [DO]) to assess
waterbody condition and  2) models  that link the response indicator endpoints (e.g., algal abundance)  to
nutrient inputs and other  site-specific factors and  management controls. These models are intended to be
used to establish nutrient targets for point source discharge and municipal stormwater permits and total
maximum daily loads (Tetra Tech 2006). Tetra Tech (2006) developed the benthic biomass spreadsheet tool
for use in  establishing "scoping  levels" nutrient targets in streams.  As the  SWRCB prepares to propose
nutrient objectives for wadeable streams, analysis  using newly available data from statewide stream
1 See definition of eutrophication and other key terms in Appendix A.

                                                 1

-------
bioassessment surveys can improve the scientific basis for policy decisions on nutrient objectives. In the
context of this study, "endpoints" refer to regulatory decisions at which point management action should be
taken, while "thresholds" refer to the output of scientific analyses.

The objectives of this research project are three-fold:

    • Estimate the natural background and ambient concentrations of nutrients and candidate indicators
     of primary producer abundance in California wadeable streams;
    • Explore relationships and identify thresholds of adverse effects of nutrient concentrations and
     primary producer abundance on indicators of aquatic life in California wadeable streams;  and
    • Evaluate the Benthic Biomass Spreadsheet Tool (BBST) for California wadeable streams using
     existing data sets and recommend avenues for refinement.
The intended outcome of this study is research, NOT final regulatory endpoints for nutrient  and response
indicators for California wadeable streams.  In this  context,  this research  can  provide:  1) improved
understanding of the corresponding quantitative thresholds at which eutrophication  stressors  (e.g., nutrient
concentrations, algal abundance) begin to exert adverse effects on aquatic life measures, 2) context for these
thresholds by summarizing available data on reference  and ambient concentrations of stressors and 3) an
improved understanding of what types of nutrient-response modeling may be appropriate, given existing
data. The findings of this research study,  as  well as other analyses, may be used  as lines of evidence
considered to support SWRCB policy decisions on nutrient objectives for wadeable streams.

  .'" : '„;,,
The document is organized as follows:

Chapter 1: Introduction, Objectives and Document Organization

Chapter 2: Estimation of Reference and Ambient Concentrations of Algal Biomass

Chapter 3: Investigating Nutrient and Primary Producer Abundance Thresholds for Aquatic Life Response

Chapter 4: Validation of NNE Benthic Biomass Spreadsheet Tool and Investigation of Stream  Nutrient
         Relationships with Biomass

Appendices

-------
1.3
Biggs, B.J.F. 2000. New Zealand Periphyton Guideline: Detecting, Monitoring and Managing the
    Enrichment of Streams. Ministry for Environment Publication, Wellington, New Zealand, 151 pp.
Fovet, O., Belaud G., Litrico X., Charpentier S., Bertrand C, Dollet P., and C. Hugodot. 2012. A model for
   fixed algae management in open channels using flushing flows. River Research and Applications
    28:960-972.
Lembi, C.A. 2003. Control of nuisance algae. In: Wehr JD, Sheath RG (eds). Freshwater Algae of North
   America Ecology and Classification. Academic Press, New York, pp 805 - 834.
Quinn, J.M. and B.W. Gilliland. 1989. The Manawatu River cleanup - Has it worked? Transactions of the
    Institution of Professional Engineers, New Zealand, 16:22-26.
Quinn, J.M. and C.W. Mickey. 1990. Magnitude of effects of substratum particle size, recent flooding and
   catchment development on benthic invertebrate communities in 88 New Zealand rivers. New
   Zealand Journal of Marine and Freshwater Research, 24: 411-427.
Steinman, A.D. 1996. Effects of grazers on freshwater benthic algae. In R. G. Stevenson et al. (eds), Algal
    Ecology. Academic Press, San Diego, CA: 341-365.
Suplee, M.W., Watson V., Teply M. and H. McKee. 2009. How green is too green? Public opinion of what
   constitutes undesirable algae levels in streams. Journal of the American Water Resources
   Association, 45:123-140.
Tetra Tech 2006. Technical Approach to Develop Nutrient Numeric Endpoints for California. Tetra Tech,
    Inc. http://rd.tetratech.eom/e pa/Documents/CA  NNE  July Final.pdf
US EPA 2000. Nutrient Criteria Technical Guidance Manual for Rivers and Streams. US. EPA Office of
   Science and Technology. EPA-822-B-00-002. July 2000.
    http://www2.epa.gov/sites/production/files/documents/guidance  rivers.pdf

-------

-------
2.                  of                and                                      of


2.1
As the SWRCB prepares to propose nutrient objectives for wadeable streams, newly available data from
statewide and regional stream surveys can improve the scientific basis for policy decisions on regulatory
endpoints. These policy decisions should  be supported, in part, by the  distribution of natural background
concentrations from minimally disturbed reference sites and the distribution  of  ambient concentrations
across the full population of wadeable streams. At the time in which the  NNE framework was conceived
(Tetra Tech 2006), this had not been  summarized. Distribution of natural background concentrations and
ambient levels are key considerations in the process of determining the scientific basis and the cost/benefits
of policy decisions on regulatory endpoints. Here, natural background refers to the absence or near absence
of anthropogenic effects and ambient levels, refers to all streams, including those affected by anthropogenic
activities.

This section addresses two key questions:

   •  What is the distribution of the values of nutrient and algal abundance indicators at "Reference"
      sites that are subjected to minimal anthropogenic disturbance?
   •  What are the ambient distributions of these indicators in California perennial, wadeable streams
      statewide and by ecoregions of interest?

2,2
2,2,1 Approach
The California NNE  framework proposes  to establish regulatory endpoints for  algal abundance, dissolved
oxygen and pH in order to assess the beneficial use status of wadeable streams (Tetra Tech 2006). The
Surface Water Ambient Monitoring Program (SWAMP) has  since adopted a standardized algal monitoring
protocol which  includes alternate measures of algal abundance (e.g., ash-free dry mass [AFDM] and algal
percent cover)  (Fetscher et al.  2009). Currently, data are  available on  938 sites using this standardized
protocol,  thus  providing  the opportunity to  summarize nutrient  concentrations and  algal abundance
indicators at the statewide and ecoregional scale.

2.2.2 Do to Sources, Site Selection, and Stream Sampling Protocol
Data Sources
Survey data were compiled from the following wadeable stream monitoring programs:

   •  Statewide Perennial Stream Assessment (PSA),
   •  Statewide Reference Condition Management Program (RCMP), and
   •  Southern California Stormwater Monitoring Coalition (SMC)
The probabilistic survey design for the California  ambient surveys (PSA, SMC ) is based on  the methods
described  in Stevens and  Olsen  (2004). The quality assurance parameters for  the California datasets are

-------
based on those established for the Surface Water Ambient Monitoring Program (SWAMP 2008). In some
places (where noted), non-probability data (i.e., from sites subjectively selected for "targeted" sampling) are
also  included. In probability surveys, sites are selected using a combination of stratification and unequal
probability weighting that yields a spatially balanced distribution of sites. Because of the objective  way in
which sites are selected, regional/statewide estimates of perennial wadeable stream condition with  known
confidence limits can be generated from the survey data. For more information on probability surveys, see
Stevens and Olsen (2004). All references to statewide or regional stream percentiles in this report are based
on this sampling framework and the operational definition of perennial wadeable streams.

The probability surveys reported on here are those of 1) the State of California Perennial Stream Assessment
(PSA), and 2) the  southern California  Stormwater Monitoring Coalition (SMC). Results from these two
programs  were used to generate regional and statewide estimates of stream condition for nutrients and
indicators of primary producer abundance. In addition to  probability data, data from targeted sampling sites
were also included in the analyses. These data from targeted sites come from the state's Reference Condition
Management Program (RCMP) and a recently completed  project geared toward developing stream algal
assemblage data for use in bioassessment of stream condition. Taken together, the available data represent
938 wadeable, perennial2 stream reaches sampled from 2007 through 2011, including the sampling  frames
for probability surveys throughout the state (National Hydrography Data  Set  [NHD]  v2, www.horizon-
systems, com/nhdplus; Figure  2.1). Of these, 575 of the reaches were sampled as  part  of the probability
surveys, and the remaining 363 were targeted. Sampling was largely conducted as one-time site visits (91% of
samples) within the time frame spanning late spring to early fall,  with the vast majority occurring  in May
through August.  For sites with both benthic macroinvertebrate and algae data, the  two assemblages were
sampled during the same visit.

Site Selection and Evaluation for Probability Surveys

The spsurvey package (Kincaid and Olsen 2008) in  R (R Core Team 2008) was  used in establishing the list of
"probability sites" for each year's statewide (PSA) and regional (SMC) probability survey. This involved  using a
technique called Generalized Random Tessellation Stratified sampling site selection (GRTS; Stevens and Olsen
2004) to create spatially-balanced survey designs. As long  as sites are sampled in the order in which they
appear  on  the list,  spatial  balance among them  is preserved, and the  resulting dataset can be used to
generate estimates of natural  resource extent and condition with known confidence limits. The design of
each survey was based on a "linear" resource sensu  Kincaid and Olsen (2008). The reporting unit for this type
of survey was in terms of length (e.g., stream kilometers). Once sampling sites were identified, they were
inspected  to determine whether they  belonged to the  target sampling population  (perennial,  wadeable
streams in California), whether permission could be secured for sampling, whether they were safe to  access,
and  whether they could be reached within  a timeframe  that would not compromise  holding times for
analytes.
 We used the PSA operational definition of "perennial", i.e., those stream reaches with surface flow during the sampling
 period. A "wadeable" reach was defined as that which is < 1 m deep for at least 50% of its length.

-------
Based on these factors, as well as whether a sample was successfully collected, sites were then classified into
one of four "evaluation categories":

    • site is part of survey's "target population", and was sampled

    • site is part of "target population", but was not sampled

    • site is not part of "target population"

    • unknown
                                                                  ^£te^
                                                                      &-   >
                                                                       ^kv 
-------
Description of Stream Algal Field Sampling and Laboratory Analysis Protocols Utilized in Compiled
Wadeable Stream Survey Data

The field sampling and laboratory analyses protocols used in the compiled stream survey data are briefly
described in this section. The types and distribution of primary producer abundance across channel habitats
can vary widely among stream types. For this reason, it is important to assess primary producer abundance
within a stream  in a number of different ways,  because each individual indicator captures this distribution
differently. For example, both  benthic chlorophyll a and ash-free dry mass (AFDM) measure algal biomass,
but chlorophyll a is a proxy for the measurement of live algal biomass, while AFDM measures both live and
dead biomass, as well  as organic matter imported into the survey site. Furthermore, algae and macrophytes
can occupy  different  "compartments" within  the  stream (i.e.,  floating on  the surface, attached to
cobbles/boulders, interstitially  distributed within the upper layer of gravel and fine sediments), all of which
are included across the sample types  upon which results are reported here. The ability to  look  at a
combination  of measures may provide a more  robust overall assessment of algal/macrophyte abundance.
Based on this rationale, the SWAMP standardized algal  assessment protocol yields the following data types
for indicators of stream primary producer abundance (Fetscher et al. 2009):

  • Algal biomass:
    •   benthic chlorophyll a
    *   benthic ash-free dry mass (AFDM)
  • Algal cover:
    •   macroalgal percent cover
    •   microalgal percent cover and thickness
  • Macrophyte percent cover3

In addition to primary  producer abundance indicators, total and dissolved inorganic nitrogen and phosphorus
concentrations were also assessed. Chlorophyll a is under consideration for  use within the current NNE
framework. Other indicators (e.g., percent cover, AFDM, or other measures) may be considered for inclusion
in the future.

A  "multi-habitat" method was  employed to quantitatively collect benthic algae at each sampling site4. This
method, SWAMP's Standard Operating Procedures (Fetscher et al. 2009), is based largely on the procedures
of EPA's Environmental  Monitoring and Assessment Program (EMAP; Peck et al. 2006) and is analogous to
SWAMP's method for collecting benthic macroinvertebrates (Ode  2007). It involves objectively collecting
from a known surface area specimens from a variety of stream substrata, in proportions aligning with relative
abundances of substratum types in the stream.  Specifically, eleven subsamples are collected at objectively
determined locations,  one from each of 11 transects that are spaced equidistantly from one another, across
the 150-m long sampling reach. For systems with a mean wetted width >10, the sampling reach is 250 m
long. The subsamples are then  combined into a single "composite" sample for laboratory analyses. As such, a
given  composite sample may have been collected from  any combination of cobbles, gravel, sand, and other
substratum types. The goal  is to achieve  a representative sample of the benthic algae from each sampling
reach, in terms of both community composition and biomass.
3 Macrophytes technically refer to both macroalgae and rooted aquatic vegetation. In this context, we define
4
macrophytes as rooted aquatic vegetation.
BMIs and algae were collected in tandem at each of the 11 subsampling locations described at each study site; first
BMIs, then algae, slightly offset so that sampling locations did not interfere with one another.

                                                8

-------
Various measures of algal and macrophyte cover were carried out using the methods outlined in Fetscher et
al.  (2009). This involved recording point-intercept presence/absence of microalgae, macroalgae,  and
macrophytes at each of 105 points objectively positioned (in a pre-determined grid) throughout each stream
reach.  Macroalgae that was attached to the stream bottom was recorded  separately from that which  was
unattached  and  free-floating  at  the  time  of  assessment.  Microalgae  was  measured  based  on
presence/absence of a biofilm on stream substrata. The thickness of the microalgal biofilm was also recorded
using ordinal thickness codes.

For algal biomass, filtered aliquots of quantitatively sampled algal material were analyzed for chlorophyll a
content using EPA 445.0, and for AFDM using WRS 73A.3. Chlorophyll a and AFDM concentrations measured
in the laboratory were transformed into mass per area of stream bottom sampled (e.g., mg/m2).
Most algal/macrophyte field metrics were calculated as percent cover estimates based on the percentage of
sampling points at which the type of algae/macrophyte was observed. The midpoint values of the ranges
corresponding to each thickness code for mean microalgal thickness were averaged across all 105 sampling
points  per site (Fetscher et al. 2009). A "nuisance algae" metric combining information from both macroalgae
and thick microalgae (>1 mm) was also calculated. A summary with descriptions of the metrics associated
with algal/macrophyte cover is provided in Table 2.1.
Sites were grouped into "disturbance classes"  throughout  the following  analyses.  To  assign  sites to
disturbance classes, we used the same set of screening criteria as that employed by the State of California's
Biological Objectives initiative (Ode et al., under review). Under this approach, sites are  classified according
to the  degree  of anthropogenic disturbance they are exposed to, based on surrounding  land uses and local
riparian disturbance measures. Table 2.2 provides a list of the factors that were used for classifying sites  into
one of the three disturbance  classes: "Reference", or those sites that are  exposed to the lowest levels of
anthropogenic disturbance based on the variables considered,  "Stressed", or those  sites exposed to the
highest levels,  and "Intermediate", or those sites falling between the "Reference" and "Stressed" groups.

-------
Table 2.1. Metric
descriptions and codes for stream primary producer abundance indicators.
Metric Code
  Description
   Percent Presence of Attached Macroalgae (defined as algal mats or filaments easily
     visible to the naked eye)
   Percent Presence of Macroalgae (Attached and/or Unattached)
   Percent Presence of Unattached Macroalgae
PCT_MAA
PCT_MAP
PCT MAU
PCT_MIAT1
PCT_MIAT1P
PCT_MIATP
   Percent Presence of Thick Microalgae (lmm+)
   Percent Presence of Thick Microalgae (lmm+), where Microalgae Present
   Percent Presence of Microalgae
PPT I\KA            Percent Presence of Nuisance Algae (Macroalgae and/or Thick Microalgae [lmm+]
                      counts as "presence" at a given point)
XMIAT              Mean Microalgae Thickness (mm)
XMIATP             Mean Microalgae Thickness (mm) where Microalgae Present
PCT_MCP            Percent Presence of Macrophytes
                                                  10

-------
Table 2.2. Variables used for assigning sites to "site disturbance classes" per the state's bio-objectives
process (adapted from Ode et al., under review). WS: Watershed. 5K: Watershed clipped5 to a 5-km buffer of
the sample point. IK: Watershed clipped to a 1-km buffer of the sample point. W1_HALL:  proximity-weighted
human activity index (Kaufmann et al. 1999). In order to be considered "Reference" condition, all criteria
listed in the "Threshold" column for "Reference" must be met. If any of the criteria in the  "Stressed" column
apply, that site is considered "Stressed". Sites not falling into either of these categories default to
"Intermediate".  Data sources are as follows: A: National Landcover Data Set (2006,
http://www.epa.gov/mrlc/nlcd-2006.html). B: Custom roads layer (P. Ode, pers. comm.). C: National
Hydrography Dataset Plus (v2, http://www.horizon-systems.com/nhdplus/). D: National Inventory of Dams.
E: Mineral Resource Data System (MRDS 2014). F: Field-measured variables (Fetscher et al. 2009).
Variable
% Agriculture
% Urban
%Ag + % Urban
% Code 21s

Road density
Road crossings


Dam distance
% Canals and pipelines
Instream gravel mines
Producer mines
W1_HALL
Scale*
Ik, 5k, WS
Ik, 5k, WS
Ik and 5k
Ik and 5k
WS
Ik, 5k, WS
Ik
5k
WS
WS
WS
5k
5k
reach
Threshold
(Reference)
<3
<3
<5
<7
<10
<2
<5
<10
<50
>10
<10
<0.1
0
<1.5
Threshold
(Stressed)
>50
>50
>50
>50
>50
>5
-
-
-
-
-
-
-
>5
Unit
%
%
%
%
%
km/km2
crossings/ km2
crossings/ km2
crossings/ km2
km
%
mines/km
mines
NA
Source
A
A
A
A
A
B
B, C
B,C
B,C
D
C
C, E
E
F
*For variables in which multiple spatial scales are used for determining site classification, in the case of the "Reference"
 boundary, the value indicated must apply to all spatial scales listed, whereas for the "Stressed" boundary, the indicated value
 need only apply for one of the listed spatial scales.


Secondary data for watershed characterization were derived from the sources described below. Watershed
and  local habitat characteristics are  required  both as co-variates  in periphyton and  macroinvertebrate
response models and as predictors  of watershed disturbance regimes. Factors affecting instream periphyton
growth and biomass accrual  include nutrients (and their ratios),  solar radiation, temperature, shading from
riparian cover, incised stream channels, local topography, mean stream velocity, substratum type, abundance
of grazers, and frequency, magnitude, and time since droughts or  scouring flows. Field data were collected by
PSA, RCMP, and  SMC monitoring programs. Sources of landscape, meteorological, and geology data are listed
in Table 2.3.
 Only the land within the catchment contributing to the sampling site was included within the indicated radii (i.e., the
area was clipped at the watershed boundaries).
6 "Code 21" encompasses a wide range of land uses primarily characterized by heavily managed vegetation (e.g., low-
 density residential development, parks, golf courses, highway medians)

                                                  11

-------
Table 2.3. Sources of data for landscape, meteorological, and geological explanatory variables used in
predictive models. DEM = digital elevation model.
Data Type/Variable
Minimum and maximum air
temperature per month
(2007-2012)
Solar Radiation (for
topographic shading)
Cloud cover, mean percent
per month (2007-2012)
Land cover/land use
Hydrology
Elevation
Drainage area (from DEM)
Geology maps
Total precipitation per
month (2007-2012)
Basin slope (from DEM)
Data Source
PRISM
ArcMap 10 tool Solar
Radiation using DEM
data from NHDPIus
MODIS Cloud data
from NASA
National Landcover
Data Set, 2006
National Inventory of
Dams and NHD Plus
National Elevation
Dataset
NHDPIus
uses
PRISM
NHDPIus
Description or Download
http://www.prism.oreRonstate.edu/products/matrix.phtml,
http://www.prism.oreRonstate.edu/docs/index.phtml

http://www.horizon-systems.com/nhdplus/

http://ladsweb.nascom.nasa.Rov/data/

http://www.epa.Rov/mrlc/nlcd-2006.html

http://Reo.usace. army. mil/pRis/f?p=397: 1:0;
http://www.horizon-systems.com/nhdplus/index.php
http://ned.usRS.Rov/
http://www.horizon-systems.com/nhdplus/

http://mrdata.usRS.Rov/ReoloRV/state/
http://www.prism.oreRonstate.edu/products/matrix.phtml,
http://www.prism.oreRonstate.edu/docs/index.phtml
http://www.horizon-systems.com/nhdplus/

2.2,3 Distribution of Wadeable Stream Nutrients    Primary Producer Inldicator Values
To provide an overview of the values for each of the indicators of primary producer abundance in  the
ambient surveys' target population (i.e., California perennial, wadeable streams), we generated descriptive
statistics for estimated  data distributions  and cumulative distribution functions (CDFs) (Kincaid and  Olsen
2009), using the spsurvey package  in  R on the probability subset of data. A CDF depicts  the estimated
probability distribution of values of a given indicator relative to the cumulative proportion of the geographic
unit of interest, i.e., percent of stream length in the state.

Each site in the combined probability surveys for the different programs/years has an associated weight in
units  of stream  length,  which reflects  how much of the state's  stream network,  within the stratum (e.g.,
landcover type, region, watershed) in which that site is found, is "represented by that site". The more sites in
a given stratum, the less weight  each  site is assigned. Because  data from multiple surveys with  different
stratification schemes were combined for this report, it was necessary to create mutually exclusive "cross-
categories" corresponding to the  intersection of the different  strata from  the various surveys. Once  cross-
categories were created, the weights of all sites had to be adjusted to reflect the combined numbers of sites
within each new cross-category. Adjusted weights were calculated for each cross-category by dividing  the
total stream length within that cross-category by the number of sites evaluated during site reconnaissance.
Once  weights were adjusted, statewide extent and magnitude estimates for the various primary  producer
indicator values could be computed (see below).
                                                 12

-------
It is not uncommon for some of the sites generated  in a probability-based design to prove unsuitable for
sampling for a variety of reasons that include: 1) the site being found, during reconnaissance, not to be part
of the survey's designated "target population"; or 2) the site is within the target population, but for some
logistical reason, it cannot be sampled (e.g., access denial, physical barriers or sheer distance of the site from
nearest roads).  Comprehensive documentation is  required  in  order  to  classify  sites into  "evaluation
categories" based on the results of site reconnaissance. If insufficient information regarding why samples
were not collected is provided by field crews, the default classification for a site is "Unknown".

We chose to describe the ambient distribution of nutrients and primary producer abundance statewide and
by ecoregion relative to the 75th  percentile of reference sites. The percent of stream kilometers with indicator
values below the 75th percentile of reference were calculated  using the Horvitz-Thompson estimator (1952),
which  is  a weighted  average of sample  values  where  weights  are adjusted  according  to  design
implementation. Confidence intervals were based on  local neighborhood variance estimators (Stevens and
Olsen 2003), which assumes that samples located close together tend to be more alike than samples that are
far  apart. Graphical  output for all  analyses in the report was  generated  using the R package ggplot2
(Wickham 2009). All graphics and statistical analyses in the report were carried out using R (version 2.15.1, R
Core Team 2012), unless otherwise noted.

23 Res  . • ;
2,3.1 Distnouaon of Nutrients    Primary Producer Indicators  at Reference Sites
For the most part, quality of "Reference" sites, as identified by our standard set of screens, did not noticeably
vary among regions in terms of the distribution of percent open (undeveloped) space within the contributing
watersheds. The one notable exception was the South Coast (and particularly the xeric portion thereof),
which did have a somewhat lower overall  percentage of open  space (96%) than other regions (which ranged
from 98 to 99%).

Chlorophyll a, AFDM, macroalgal percent cover, and nutrients (TN and TP) exhibited a considerable degree of
variability in values among Reference sites, but their distributions were highly skewed toward the low end of
the stressor gradients  (Table 2.4).  At the 75th  percentile, the ranges in nutrients  and primary producer
indicator values  among ecoregions  were fairly  narrow (i.e.,  0.10-0.31 mg/L TN, 0.02-0.04 mg/L TP, 8-27
mg/m2 chlorophyll a, 6-27 g/m2 AFDM, and 15-37% cover of  macroalgae). For primary producer indicators
and TN, North Coast and Sierra Nevada reference sites represented the lower end of that range, while South
Coast, Desert-Modoc and Central Valley represented the upper end of the range.
                                                13

-------
Table 2.4. Median, 75th, and 95th percentiles of raw (unweighted) TN, TP benthic chlorophyll a, AFDM,
and macroalgal percent cover (PCT_MAP), statewide and by region, at Reference sites (both probability
and targeted datasets included).
Statistic by Primary
Producer Indicator type
Chlorophyll a
(mg/m2)
AFDM
(g/m2)
Macroalgal
percent cover
(%)
TN (mg/L)
TP (mg/L)
Median
75th
95th
Median
75th
95th
Median
75th
95th
Median
75th
95th
Median
75th
95th
Statewide
n=263
6.9
14.6
44.1
5.4
11.9
34.0
7.0
22.9
45.7
0.091
0.161
0.462
0.019
0.032
0.074
Central
Chaparral Valley1
n=56 n=l
8.9
16.4 23.0
46.2
6.2
10.0 12.9
19.7
3.5
15.9 41.0
38.9
0.090
0.144 0.155
0.264
0.022
0.042 0.027
0.088
Deserts-
Modoc
n=10
10.7
26.5
32.0
13.4
23.9
36.7
30.5
36.8
55.9
0.223
0.281
0.467
0.027
0.041
0.079
North
Coast
n=41
6.2
9.2
25.1
4.0
6.0
14.8
5.5
15.0
36.5
0.090
0.117
0.212
0.016
0.020
0.045
South
Coast
n=74
12.5
24.4
124.8
16.3
26.8
130.6
9.5
26.0
60.0
0.138
0.308
0.925
0.018
0.035
0.106
Sierra
Nevada
n=81
3.1
7.9
28.3
3.7
5.8
12.2
7.0
23.0
50.3
0.065
0.100
0.185
0.021
0.032
0.060
1 The Central Valley ecoregion had only one site in the Reference site disturbance class; values in the table represent the results
of this single site.

2,3,2 Ambient           of Nutrients     Primary Producer Abundance
The proportions  of sites falling into the four site "evaluation categories" are shown in Table 2.5. By far, the
majority of stream kilometers in the state were estimated to fall outside of the surveys' "target population",
either because they were non-perennial or non-wadeable stream reaches. The proportion of sites for which
samples were collected represented about 10% of the state total stream kilometers.

Analysis of the statewide ambient wadeable stream data showed that algal biomass parameters, (chlorophyll
a, AFDM) and nutrients exhibited broad ranges in concentrations, but their distributions were very highly
skewed toward the low end (Table 2.6, Figure B.I). This was also generally true of primary producer percent
cover metrics, with the exception of percent presence of microalgae (PCT_MIATP).

CDFs of site disturbance classes show a good amount of separation of reference, intermediate and stressed
sites for chlorophyll a and AFDM, but not for macroalgal % cover (Figure 2.2). Boxplots of the distributions
are provided  in Appendix B, Figure B.2.
                                                 14

-------
Table 2.5. Extent estimates for the site-evaluation categories based on reconnaissance information
across the PSA and SMC probability surveys from 2008-2011.
Number of Sites
Site Evaluation Category Sampled*
Part of survey's "target
population", and sampled
Part of "target population",
but not sampled
Not part of "target
population"
Unknown
572
400
3362
174
Estimated Stream Kilometers Confidence Interval
(% of State Total) (95%)
33,499
43,438
238,195
9,510
(10)
(13)
(74)
(3)
29,101
37,973
231,300
7,270
- 37,897
- 48,903
- 245,089
- 11,750
* Note that each sample for the input data used in the analysis represents either a one-time sampling event, or an average (for
the small subset of stream reaches for which multiple samples over time were available).
Table 2.6. Statewide estimates for distributional properties of primary producer abundance indicator
values in California perennial, wadeable streams. Data are from combined PSA and SMC probability
surveys from 2008-2011. SE: standard error of the mean; Cl: confidence interval (95%). Indicator acronyms
are defined in Table 2.1.
Indicator
TN (mg/L)
TP (mg/L)
Chlorophyll o (mg m-2)
AFDM (g m-2)
PCT_MAP (%)
PCT_MAA (%)
PCT_MAU (%)
PCT_MCP (%)
PCT_MIAT1(%)
PCT_MIAT1P (%)
PCT_MIATP (%)
PCT_NSA (%)
XMIAT(mm)
XMIATP (mm)
Range of
Measured Values
(N)
0.01-26 (538)
0.002-4.5 (536)
0.22-1504 (536)
0.07-489 (525)
0-98 (480)
0-98 (480)
0-87 (480)
0-98 (480)
0-94 (478)
0-100 (464)
0-100 (478)
0-100 (478)
0-6 (478)
0-20 (464)
Estimated Mean
(SE)
0.533 (0.074)
0.086 (0.008)
21(2)
16(2)
16(1)
14(1)
2 (0.5)
10(1)
7(1)
8(1)
76(2)
20(2)
0.5 (0.03)
0.6 (0.03)
Estimated Median
(Cl)
0.131(0.111-0.156)
0.028 (0.024-0.031)
8 (6-12)
7 (6-8)
6 (4-9)
5 (3-7)
0 (0-0)
4 (2-5)
2 (0.5-2)
2 (1-3)
86 (83-93)
11 (9-13)
0.3 (0.3-0.4)
0.4 (0.4-0.5)
Estimated
90th percentile
(Cl)
1.035 (0.846-1.428)
0.190(0.150-0.280)
47 (39-64)
40 (23-50)
51 (41-56)
43 (36-52)
3 (2-9)
25 (20-39)
20 (13-32)
22 (16-41)
99 (99-100)
52 (50-62)
1(0.8-1.5)
1(0.8-1.6)
                                                 15

-------
                          100-
                           90-
                           80-
                         r 70-
                         \ 60-
                         »
                         j 50-
                         ! 40-
                         ' 30-
                           20-
                           10-
                            o-
                          100-
                           90-
                           80-
                           -
                           ee-
                           50-
                           40~
                           30-
                           20-
                           10-
                            0-
                          100-
                           90-
                           80-
                         g70-
                         •£ 60-
                         S 50-
                         | 40-
                         " 30-
                           20-
                           10-
                            0-
                               0  25  50 75 100 125  150 175 200 225 250 275 300
                                          benthic chlorophyll a (mg/m2)
Subpopulation
     Stressed
     Intermediate
     Reference
                                   20  40  60  80  100 120 140  160 180 200  220
                                             benthic AFDM (g/m2)
                               0   10  20  30   40   50   60  70  80   90   100
                                            macroalgal percent cover
Figure 2.2. Statewide CDFs for biomass measures and macroalgal percent cover (attached and/or
unattached combined) by site disturbance class. The graphs show the estimated probability distributions
of the three types of primary producer abundance indicators relative to the cumulative proportion of stream
length. Highlighted areas delineate the 95% confidence intervals for each estimate.
                                                 16

-------
As with the reference sites, the ranges in median values of nutrients and primary producer indicator values
among ecoregions were fairly narrow (i.e., 0.05-0.48 mg/L TN, 0.02-0.09 TP, 6-26 mg m~2 chlorophyll a, 5-17 g
m~2 AFDM, and 1-20%  cover of macroalgae, Table 2.7, Figure  2.3).  North  Coast and Sierra Nevada sites
represented the lower end of that range, while South Coast and Central Valley consistently represented the
upper end.

Table 2.7. Estimated median values (with 95% confidence intervals) for selected ambient stream nutrient
and primary producer abundance indicators statewide and by region. Data are from combined PSA and
SMC probability surveys from 2008-2011.
Indicator
Chlorophyll o
(mg m"2)

AFDM (g m )
Macroalgal percent
cover (PCT_MAP, %)

TN (rne/L)
1 ' " \> • '£>/ •-/

TP (mg/L)
Chaparral
13
(5.6-17.4)
6.6
(5.8-9.3)
5
(3-17.7)
0.251

(0.135-0.365)
0.034
(0.023-0.094)
Central Valley
12.6
(7.5-21.6)
13
(10.3-18.6)
16.9
(4.9-33.9)
0.480

(0.332-0.890)
0.095
(0.041-0.196)
Deserts-Modoc
8.9
(5.8-11)
10.2
(6.9-12.4)
11.9
(7-21.7)
0.257

(0.192-0.364)
0.041
(0.028-0.053)
North Coast
5.7
(4-11.3)
5.5
(4.6-6.5)
7
(3-12.9)
0.104

(0.080-0.130)
0.028
(0.020-0.029)
South Coast
25.7
(19.2-40.7)
17.2
(10.9-23.9)
20.1
(14.6-29.8)
0.744

(0.540-0.989)
0.050
(0.045-0.090)
Sierra Nevada
5.7
(2.9-12)
4.8
(4.1-9.4)
1
(0.2-4)
0.052

(0.042-0.081)
0.020
(0.016-0.021)
Statewide, the percentage of stream kilometers that exceeded the 75th percentile of statewide reference
values ranged from 27 % for macroalgal percent cover to a high of 41 % for TP and TN (Figure 2.3, Table 2.8).
This  range was generally greater among the  regions and inconsistent by indicator group by region. For
example, regions that were on the lower end  of the absolute concentration range (North Coast and Sierra
Nevada)  had a higher  percentage  of  miles  exceeding their respective  75th percentile  of eco-regional
reference, putting them within range of South Coast, a region consistently at the upper edge of concentration
range (Figure 2.2, Table 2.8). This is due to a proportionally lower Ecoregional reference value.

Table 2.8 Percent of perennial wadeable stream kilometers exceeding 75th and 95th percentiles of
statewide or regional Reference values for nutrient and primary producer abundance gradients. By this
definition, 25% and 5% of Reference sites, respectively, exceed the indicated value as well. Data are from
combined PSA and SMC probability surveys from 2008-2011.
Statewide
Gradient
TN (mg/L)
TP (mg/L)
Chlorophyll o
(mg/m2)
AFDM (g/m2)
Macroalgal
percent cover
(PCT_MAP; %)
75th
41
41
34
27

26
95th
22
21
10
11

11
Chaparral
75th
58
42
39
32

35
95th
44
38
8
12

17
Central Valley
75th 95th
85
58
34
55

33
Deserts-
Modoc
75th
48
70
14
19

16
95th
30
26
7
10

4
North
75th
40
69
43
48

22
Coast
95th
12
4
10
2

9
South
75th
75
62
54
34

42
Coast
95th
2
32
15
6

16
Sierra Nevada
75th
27
23
43
41

16
95th
6
2
10
18

9
                                                17

-------
                                  25  50 75 100 125 150  175 200 225 250 275 300
                                         benthic chlorophyll O (mg/m2)
                                                        Subpopulation

                                                           Chaparral
                                                           Deserts-Modoc

                                                           North Coast

                                                           South Coast

                                                           Sierra Nevada
                                  20 40  60  80  100 120 140 160 180 200 220
                                            berthic AFDM (g/m2)
                                   10   20
                                           30   40   50   60   70  80  90  100
                                           macroalgal percent cover
Figure 2.3. CDFs for biomass measures and macroalgal percent cover (attached and/or unattached
combined), broken down by PSA6 ecoregion. The graphs show the estimated probability distributions of
the three types of primary producer abundance indicators relative to the cumulative proportion of stream
length. The vertical dashed line on each graph denotes the 75th percentile among Reference sites,
statewide. Confidence intervals for each CDF can be viewed on the individual graphs for each ecoregion
provided in Figure B.3. In addition, a further breakdown of the CDFs within the South Coast ecoregion (i.e.,
"xeric" and "mountain" subregions) is provided in Figure B.4.
                                                  18

-------
 "'. :'  "
California's perennial, wadeable streams, as assessed during the PSA index period7, exhibited a skew toward
the lower end of the nutrient and primary producer abundance gradient, although nutrients  and primary
producer abundance were understandably higher in intensively developed  regions like South  Coast  and
Central Valley. Statewide, only an estimated 34% of perennial stream kilometers had chlorophyll  a values
exceeding the 75th percentile of Reference sites8 statewide. This percentage was slightly higher for nutrients
(41% of stream kilometers for both TN and TP).

Interpretation of the ambient distribution  of nutrients and algae should be tempered by an understanding
that the data may not represent the peak concentrations. Algal abundance and nutrient concentrations vary
seasonally as a function of stream flow, temperature, available light, grazing pressure, nutrient source  and
other factors (Dodds et al. 2002). The data  utilized in this survey represent a single time point taken during a
late spring-summer index period. This  index period was established to optimize condition assessment for
benthic macroinvertebrates, not stream  algal  abundance, per  se. The optimum period  for stream algal
assessments has not been established (see  Appendix E for discussion). Atmospheric deposition is a significant
component  of  N  loading to California ecosystems, with a  much more significant contribution of  dry
deposition to loading. Atmospheric deposition  can have a more far-reaching effect than point sources  and
can affect otherwise pristine montane streams due to atmospheric transport. Impacts of dry deposition on
stream  water chemistry  can  be  delayed from the  dry  summers  until  fall/winter  when  rains  occur
(Bytnerowicz and Fenn 1996).

The 75th percentiles for TN and TP estimated from the  probability-based samples of reference streams are
similar to those modeled for RF1 reaches  in the corresponding nutrient ecoregions by Smith et al. (2003).
Smith's values were based on  models developed from  estimated yields of USGS reference gaging stations
(1976 - 1997), with corrections for wet atmospheric N deposition based on interpolated NADP values from
1980 - 1993. Smith et al. estimated an upper quartile of 0.21  (s.d. = 0.07) mg N/L with wet deposition  and
0.18 (s.d. = 0.07) after correction for wet atmospheric deposition for annual flow-weighted instream nutrient
concentrations in  the Central Valley and Western Forested Mountain nutrient ecoregions, and 0.11 (s.d. =
0.04)  with and 0.05 (s.d. = 0.07) without wet deposition  for the  Xeric  West  ecoregion.  For total  P
concentrations, Smith et al. estimated 75th  percentiles of 0.02 (s.d. = 0.005) and 0.03 (s.d. = 0.015) mg P/L for
Central Valley and Western Forested Mountain ecoregions or Xeric West ecoregions, respectively. There are
likely to be some differences between our estimates and those of Smith et al. due to differences in sampling
dates as atmospheric  N deposition has  been  declining, and because Smith et al. calculated  estimated
concentrations across the RF1 stream network which extends to larger systems than the perennial wadeable
stream dataset. Corrections by Smith et al for atmospheric deposition are likely underestimates because of
the dominance of dry deposition N sources  in the arid west (Bytnerowicz and Fenn 1996).
7 The PSA index period for stream sampling starts in May for drier parts of the state and June or July in colder/wetter
parts of the state (depending upon stream flow conditions), and lasts for two to three months.
8 In the case of the reference sites, values are given here for all available data combined (i.e., probability plus non-
probability, or "targeted", sites)
                                                19

-------
- .'-;
Bytnerowicz, A. and M. E. Fenn. 1996. Nitrogen deposition in California forests: a review. Environmental
    Pollution, 92:127-146.
Dodds, W.K., Smith V.H., and K. Lohman. 2002. Nitrogen and phosphorus relationships to benthic algal
    biomass in temperate streams. Canadian Journal of Fisheries and Aquatic Sciences, 59: 865-874.
Fetscher, A.E., Busse L.B., and P.R. Ode. 2009. Standard Operating Procedures for Collecting Stream Algae
    Samples and Associated Physical Habitat and Chemical Data for Ambient Bioassessments in
    California. California State Water Resources Control  Board Surface Water Ambient Monitoring
    Program (SWAMP) Bioassessment SOP 002. (updated May 2010)
Horvitz, D.G. and D.J. Thompson. 1952. A generalization  of sampling without replacement from a finite
    universe. Journal of the American Statistical Association 47:663-685.
Kaufmann, P.R., Levine P., Robinson E.G., Seeliger C, and D.V. Peck. 1999. Surface waters: Quantifying
    physical habitat in  wadeable streams. EPA/620/R-99/003. US EPA. Office of Research and
    Development. Washington, DC.
Kincaid, T. and A.R. Olsen. 2008. SPSurvey package for R. Available from http://www.epa.gov/nheerl/arm
Mineral Resources Data System (MRDS). 2014. http://tin.er.usgs.gov/mrds/. Accessed 18 January 2014.
Ode, P. 2007. SWAMP  Bioassessment Procedures: Standard operating procedures for collecting benthic
    macroinvertebrate samples and associated physical  and chemical data for ambient bioassessment in
    California. Available from http://mpsl.mlml.calstate.edu/phab_sopr6.pdf
Ode, P.R., Rehn A.C., Mazor R.D., Schiff K. C, May J. T., Brown L R., Gillett D.,  Herbst D., Lunde K. and C.P.
    Hawkins. In review. Evaluating the adequacy of a reference pool for use in environmentally complex
    regions.
Peck, D.V., Herlihy AT., Hill B.H., Hughes R.M., Kaufmann P.R., Klemm D., Lazorchak J.M., McCormick
    F.H., Peterson S.A., Ringold P.L, Magee R., and M. Cappaert. 2006. Environmental Monitoring and
    Assessment Program-Surface Waters Western Pilot Study: Field operations manual for wadeable
    streams. U.S. Environmental Protection Agency, Washington, D.C. EPA/620/R-06/003.
R Core Team. 2008. R:  A language and environment for statistical computing.  R Foundation for Statistical
    Computing, Vienna, Austria. ISBN 3-900051-07-0. http://www.R-project.org/
R Core Team. 2012. R:  A language and environment for statistical computing.  R Foundation for Statistical
    Computing, Vienna, Austria. ISBN 3-900051-07-0. http://www.R-project.org/
Smith, R. A., Alexander R.B., and G.E. Schwarz. 2003. Natural background concentrations of nutrients in
    streams and rivers of the conterminous United States. Environmental Science and Technology,
    37:3039-3047.
Stevens, D.L. and A.R. Olsen. 2004. Spatially balanced sampling of natural resources. Journal of the
    American Statistical Association, 99:262-278.
Stevens, D.L. and A.R. Olsen. 2003. Variance estimation for spatially balanced samples of environmental
    resources. Environmetrics, 14:593-610.
Tetra Tech. 2006. Technical Approach to Develop Nutrient Numeric Endpoints for California. Tetra Tech,
    Inc. http://rd.tetratech.com/epa/Documents/CA NNE July  Final.pdf
Wickham, H. 2009. ggplot2: elegant graphics for data analysis. Springer, New York.
                                                20

-------
3.                  of                         of
                      and                on                                         Life


3.1
Nutrient overenrichment,  in concert with other site-specific factors, can result in the overabundance of
organic matter in a process known as eutrophication (Nixon 1995). The adverse effects of eutrophication on
stream ecosystem services such as biodiversity, water quality and aesthetics have been well vetted in the
literature (Figure 3.1).  Nutrients, together with a  complex suite of reach scale environmental factors, can
have direct and indirect effects on biotic communities (Wotton et al. 1996, Stevenson et al. 1997, Risang et
al. 2004). Nutrients stimulate autotrophic production. This  high algal abundance can alter hydrology  and
interfere with spawning, foraging, and shelter (Biggs 2000, Quinn and Mickey 1990). Filamentous algae that
proliferates in high nutrient conditions can limit the growth of benthic diatoms as a food source for primary
consumers such as scraper/grazers (Steinman  1996).  Excessive organic matter  accumulation can cause
declines  in dissolved oxygen, leading to deteriorated habitat quality (Quinn and Gilliland,  1989). Nutrients
also increase heterotrophic production, a pathway  much less well studied than autotrophic pathways (Evans-
White 2014, Dodds 2007). Most studies to date have demonstrated short-term stimulation of bacterial  and
fungal growth with nutrient additions, improved nutritional  quality  of leaf litter (decreased C:N ratios)  and
concomitant increases in detritivore biomass (Greenwood et al. 2007, Connolly and Pearson 2013, Tant et al.
2013).  Longer-term enrichment of headwater streams can lead to a decreased efficiency of trophic transfers,
with an increased loss  of carbon downstream and reduced  productivity  of top predators if nutrients have
differential effects on primary consumers with different degrees of resistance to predators (Davis et al. 2010,
Suberkropp et al. 2010). Algal blooms can also negatively impact ecosystem services by causing taste/odor
problems, cyanobacterial toxin production (Chorus and Bartram 1999; Aboal et al. 2002;  Douterelo et al.
2004), blocked of filtration systems, and compromised aesthetics (Biggs 2000, Lembi 2003, Suplee et al. 2009,
Fovet et al. 2012).

While the conceptual models for adverse effects are generally well accepted, research is needed to better
quantify relationships  among  nutrients,  stream landscape- and site-scale  environmental co-factors  and
ecological responses (Stevenson et al. 2011). In particular, thresholds in the ecological responses along these
environmental gradients help develop stakeholder  consensus for management action and provide a basis for
evaluating the cost vs.  benefits of different options (Stevenson and  Sabater 2010, Muradian 2010). Used in
this context, an ecological threshold refers to a marked change in a dependent variable (ecological response)
within a small range of the independent variable (stressor).

Over the past decade, there has been a tremendous increase in  peer-reviewed science examining levels or
thresholds of nutrients adversely affecting aquatic life indicators. Most examples to date have used benthic
macroinvertebrates, algal,  and/or fish  community  composition as AL measures in empirical  stress-response
relationships with  nutrients. In addition,  most of these studies have focused on streams in temperate
climates. It is unclear how  applicable these thresholds are in California's Mediterranean climate, when many
of the major co-factors controlling response to nutrients or algal abundance are fundamentally different (e.g.,
rainfall frequency, flow, temperature, and light availability). In some cases, benthic algal abundance
                                                21

-------
     ECOSYSTEM SERVICES
        Biodiversity
         Aesthetics
       Nitrogen Cycling
     Primary Production
     Good Water Quality
                                      RESPONSE INDICATORS
  Altered Aquatic Life
 (e.g. Benthicand Algal
 Community Structure)
Increased abundance of
primary producers (e.g.
 algae) & heterotrophs
    (e.g. bacteria)
                                       Altered DO and pH
                                NUTRIENTS AND STREAM
                                     CO-FACTORS
                                                                         Light Availability
                                                                           Hydrology
Nutrients (Nitrogen,
   Phosphorus)
And Organic Matter
                                 Biological Communities
                                                                          Temperature
                                                                          Et al. Factors
Figure 3.1. Simplified conceptual model of eutrophication in wadeable streams depicting the
relationship between nutrients, stream co-factors, ecological response and ecosystem services.
and/or dissolved oxygen measures have  served  as  intermediate  response variables for inferring  impacts
thereto (Wang et al. 2007, Weigel and Robertson  2007, Stevenson  et al. 2008, Miltner 2010, Smith and Iran
2010, and Suplee and Watson 2013).

Measures of algal abundance such as benthic chlorophyll a, algal percent cover, or ash-free dry mass are of
interest to California water quality managers for  use in assessment of eutrophication. This is because such
measures have a strong mechanistic linkage with  nutrients, but are more robust measures of the impacts of
nutrient  enrichment on the ecosystem services because they integrate stream  co-factors (Figure 3.1).
However, few studies  have identified thresholds in relationships among algal abundance measures and
aquatic life (e.g.  benthic macroinvertebrate [BMI] or algal community composition). Most studies are based
on watershed- or reach-scale mechanistic  models that link algal abundance to dissolved oxygen. Very few
empirical stress-response studies have been published looking  specifically at thresholds of algal abundance
(benthic  chlorophyll a, ash-free dry mass,  or macroalgal percent cover) that adversely affect BMI  or algal
communities in  wadeable streams. Of these, the majority have been conducted in New Zealand and are,
therefore,  of uncertain applicability to California's wadeable streams. As  California state water quality
managers are interested in using  biological  response to assess status of stream  beneficial uses vis-a-vis
nutrients, this study focused on investigations of thresholds of algal abundance measures as well as nutrients
on aquatic life measures.

Over the past 10 years, California's investment in a stream bioassessment program has produced a large data
set that  can be  used to investigate the linkage  of  nutrient and  algal  abundance  to All, employing both
reference and empirical stress-response approaches. The previous section of this  report summarized the
reference distribution of nutrients and algal abundance  indicators. The objectives of this section are to 1)
investigate  relationships of nutrients  and primary producer  abundance indicators  with BMI  and algal
community measures of All and 2) determine levels of nutrients and primary producer abundance indicators
associated with adverse effects to these AL  measures.
                                                 22

-------
3.2 Methods
3.2.1 Conceptual Approach
Three basic approaches have been used in establishing levels of stream nutrients and algal abundance that
are protective  of aquatic  life  (U.S. Environmental  Protection  Agency 2000):  1) reference approach,  2)
empirical stress-response approach, and 3) process-based approach. Of these approaches, the reference and
empirical stress-response versions are among the most commonly used quantitative approaches to establish
WQOs across large geographic areas,  such as California wadeable streams. The latter approach involves
quantifying the  relationships among stressor gradients (e.g.  nutrients, algal  abundance)  and aquatic life
measures that are representative of ecosystem services. The Biological Condition Gradient (BCG) is a useful
conceptual model for empirical stress-response studies. The BCG model describes the changes in  aquatic
communities, measured by aquatic life indicators, as a function of stress (Davies and Jackson 2006; Figure
3.2). It predicts the transition of biotic communities, as measured by All indicators, as a function of increasing
stress, from pristine to slightly modified ecological condition, then moderate, and finally, very low ecological
condition. These relationships can be linear or non-linear in nature.
  Natural
       o
       T3

       §
       tB
       O
       o
       CO
Degraded
                Native or natural condition
                             Minimal loss of species; some
                             density changes may occur
Some replacement of
sensitive-rare species;
functions fully
maintained
                Tolerant species show increasing
                dominance; sensitive species are
                rare; functions altered
                                    Some sensitive species
                                    maintained but notable
                                    replacement by moreHolerant
                                    taxa; altered distributions;
                                    functions largely maintained
                             Severe alteration of
                             structure and function
             Low
                     Stressor gradient
High
Figure 3.2. Conceptual model depicting stages of change in biological conditions in response to an
increasing stressor gradient. Reproduced from Davies and Jackson (2006).
In this study, we investigated the relationships between eutrophication stressors (e.g., nutrient and primary
producer abundance indicators) and BMI and algal community structure as measures of All (Figure 3.1). BMI
and algal community structure were selected as All measures because: 1) they are the assemblages of choice
for  bioassessment  in  California  statewide and regional  programs, 2)  BMI  and/or diatom  community
composition have been used as the basis for WQO development in various other states and countries, and 3)
a large  and  geographically broad set  of ambient survey data is available for both assemblages, using
standardized field and laboratory protocols (Ode  2007 and Fetscher et al. 2009). Other potential indicators of
All attainment, such as dissolved oxygen or pH,  could be used for setting biomass/nutrient WQOs, but data
on diel ranges and fluctuations are not available statewide. A large number of BMI and algal IBI metrics were
                                                 23

-------
evaluated in order to better understand the complexity of the community response to the chosen stressor
gradients.

A wide  variety of  statistical methods  have been used  to  model the fundamental relationships among
stressors, community responses, and environmental co-factors that mediate response to stress. With respect
to setting quantitative water quality goals, two approaches  are commonly used (Figure 3.3): 1)  statistical
change  point detection and 2) regression methods  to relate stressors to quantitative ecosystem service
targets (e.g. percentile of index of biological integrity corresponding to a percentile of reference sites) (EPA
2010). In this study, we investigated where along the gradients of nutrients and primary producer abundance
these All measures exhibited  break  points, as evidence  of thresholds of adverse  effects. We focused on
breakpoints or thresholds rather than regressions that extrapolate to quantitative ecosystem service targets
because: 1) California has not yet officially adopted into policy stream BMI or algal  IBI targets, and 2) break
points and other types of thresholds are valuable to describe the response surface  of ecosystem change to
stressors over the BCG (Stevenson et al. 2011).
       Statistical Change Point
               Detection
                                  Estimate Levels Corresponding to Quantitative
                                   Ecosystem Service Target (e.g.  Biodiversity)
  ^v^
                                              .  •"•'.•  •..  ".  •  ••-;••••'
                                                                 • T
           Stressor Gradient
                                                          Stressor Gradient
Figure 3.3. Examples of two statistical approaches used to derive quantitative water quality goals.
(EPA 2010)

3.2.2 Aquatic Life and Stressor Indicators
We utilized several biotic assemblages in order to  examine multiple lines of evidence for effects of  algal
biomass and nutrients  on stream communities. Some  measures from these assemblages were selected
because they have explicit connections to eutrophication as known indicators of dissolved oxygen, organic
matter, or stream nutrient levels, allowing us to remove to some degree the effect of other confounding
stressors. Other Alls were selected because they were  developed to serve as indicators of overall stream
condition. Thus they may be  responsive to  changes in the stream  environment resulting from nutrient
enrichment,  even if they were not developed specifically  for assessing nutrient (and excessive biomass)
impacts. Furthermore, within assemblages, we used  several types of metric and index that describe different
aspects of the communities or summarize community composition as a whole. This facilitated an evaluation
of how different levels of stress  may have different degrees of effect on  stream communities. An
understanding  of the  magnitude and  extent of effects of stress on  biotic  communities across such  a
"response surface" conveys information relevant for risk  classification (Tetra Tech 2006). Within the BMI and
                                               24

-------
algal assemblages,  we  used three basic types of All measures:  1)  "raw" community  composition,  as
summarized in the  form of axes from  non-metric multidimensional scaling ordinations (NMS; see Section
3.2.5),  2) calculated metrics that describe specific aspects of a biotic community according to taxon-specific
attributes/ecological preferences, and 3) calculated multimetric indices that provide more holistic ways of
summarizing community composition. Unlike the NMS axes, metrics and indices have pre-established polarity
of scoring that is indicative of "good" vs. "bad" water-body condition.

To facilitate interpretation of the results of our analyses within the context of the BCG conceptual model, we
grouped the metric- and index-based All measures into four categories (Table 3.1):

    • Sensitive: metrics based on "sensitive" taxa, i.e., those that are known, based on the literature, to
     be highly responsive to relatively low levels of generalized stress. Also  included in this group are
     "tolerant" taxa because of the loosely inverse relationship between metrics describing proportion
     of sensitive taxa and proportion of tolerant.
    • Low-nutrients: metrics based on taxa that have been associated with low-nutrient conditions by
     previous studies in the literature
    • Eutrophication: metrics based on taxa that are tolerant to various aspects of eutrophication,
     according to the literature
    • Integrative: indices that provide an integrative measure of community composition to provide
     inference into overall water-body condition
Along the BCG gradient,  sensitive metrics would be expected to respond at level 2, with functional
changes (eutrophication metrics) occurring at level 4 and integrative indices  showing significant impacts
at levels 4 and 5.

Table 3.1. Description of aquatic life indicators (Alls) used in the analyses.  Indicators are listed in the
table alphabetically according to the shorthand version of the name, grouped by assemblage type. The All
categories used, INT= integrative, SEN= sensitive, EUTRO= eutrophication and NUT = low nutrient, are
defined above.  IBI =index of biotic integrity.

                          All                                                       All Response
Abbreviated Name      Category  Description of Variable                                to Stress
All - Benthic Macroinvertebrate	
                                 California Stream Condition Index (the BMI-based
CSCI                      INT     statewide multimetric index for stream                 decrease
                                 bioassessment; Mazor et  al., under review)
,_„_,_ „      9              „,_„.     percent BMI individuals that are Ephemeroptera,
EPT Percent               SEN     „,             T . ,              H       H             decrease
    ~	Plecoptera, orTrichoptera	
                                 percent BMI taxa that are Ephemeroptera, Plecoptera,
EPT PercentTaxa          SEN     H     ,                   v       v    ,      v    ,    decrease
    ~                             orTrichoptera
,_„_,_ T                     SEN     number of BMI taxa that are Ephemeroptera,
EPT Taxa                         „.             T . ,                                    decrease
    ~                             Plecoptera, orTrichoptera
9 Not considered "integrative", because EPT account for only a subset of the whole community.

                                                 25

-------
Table 3.1 (continued)
Abbreviated Name
lntolerant_Percent
lntolerant_PercentTaxa
lntolerant_Taxa
0/E
Shannon_Diversity
Simpson_Diversity
Taxonomic_Richness
Tolerant_Percent
Tolerant_PercentTaxa
Tolerant_Taxa
BMI community
All
Category Description of Variable
SEN percent BMI individuals that are "intolerant"; Ode et
al. 2005
SEN percent BMI taxa that are "intolerant"; Ode et al. 2005
SEN number of BMI taxa that are "intolerant"; Ode et al.
2005
INT observed over expected taxa from RIVPACS models for
BMI taxa; Mazor et al., under review
INT Shannon diversity index for BMI taxa
INT Simpson diversity index for BMI taxa
INT richness of BMI taxa
SEN percent BMI individuals that are "tolerant"; Ode et al.
2005
SEN percent BMI taxa that are "tolerant"; Ode et al. 2005
SEN number of BMI taxa that are "tolerant"; Ode et al.
2005
INT NMS axis 1 score (from ordination of BMI community
composition data)
All Response
to Stress
decrease
decrease
decrease
decrease
decrease
decrease
decrease
increase
increase
increase
no
expectation
All - Diatom
D18
propAchMin
RAWDO100
RAWDO50
RAWeutro
RAWIowN
RAWIowP
RAWNhet
diatom community
All - Hybrid of Diatoms
H2010
H21
INT diatom IBI; Fetscher et al. 2014
proportion of diatom valves that are Achnanthidium
minutissimum
proportion diatoms requiring nearly 100% DO
saturation; van Dam et al. 1994
EUTRO proportion diatoms requiring at least 50% DO
saturation; van Dam et al. 1994
EUTRO proportion eutrophication indicator diatoms; van Dam
etal. 1994
proportion low-N indicator diatoms; Potapova and
Charles 2007
proportion low-P indicator diatoms; Potapova and
Charles 2007
proportion nitrogen-heterotroph diatoms; van Dam et
EUTR° al.1994
NMS axis 1 score (from ordination of diatom
community composition data)
and Soft-bodied Algae
INT diatom + soft algae ("hybrid") IBI; Fetscher et al. 2014
INT diatom + soft algae ("hybrid") IBI; Fetscher et al. 2014
decrease
decrease
decrease
decrease
increase
decrease
decrease
increase
no
expectation

decrease
decrease

  H20, H21, and H23 differ in terms of the type of soft-algal information they include. H20 includes only species
presence/absence, H21 includes only species biovolumes, and H23 includes both types of data. For more details, see
Fetscher et al. (2014).

                                                    26

-------
Table 3.1 (continued)
Abbreviated Name
H23
All All Response
Category Description of Variable to Stress
INT diatom + soft algae ("hybrid") IBI; Fetscher et al. 2014
decrease
All - Soft Algae
propTaxaZHR
RAWIowTPsp
RAWmeanZHR
RAWpropBiovolChlor
RAWpropBiovolZHR
RAWpropGreenCRUS
S2
proportion of total of total soft-algae taxa recorded
NUT that are in the Zygnemataceae, heterocystous
cyanobacteria, or Rhodophyta
proportion of soft algal taxa that are considered "low
TP" indicators; Fetscher et al. 2014
mean of the metrics propTaxaZHR and
RAWpropBiovolZHR; Fetscher et al. 2014
proportion of total soft algae biovolume that is
EUTRO
Chlorophyta
proportion of total soft algae biovolume that is in the
NUT Zygnemataceae, heterocystous cyanobacteria, or
Rhodophyta
proportion of green algal biovolume belonging to
EUTRO Cladophora glomerata, Rhizoclonium hieroglyphicum,
Ulva flexuosa, or Stigeoclonium spp.
INT soft algae IBI; Fetscher et al. 2014
decrease
decrease
decrease
increase
decrease
increase
decrease
The focal stressor gradients used in the analyses were of two categories: 1) primary producer abundance,
including measures of benthic chlorophyll  a,  AFDM and algal and  macrophyte  percent  cover  and 2)
concentrations  of  total  and dissolved  inorganic  nitrogen  and phosphorus. In addition to  these, other
environmental gradients that could influence relationships between the focal stressor gradients and Alls
were included in analyses, where possible. These are listed in Table 3.2.

Table 3.2 Descriptions of the focal stressor gradients and other explanatory variables used in the
analyses, listed in the table alphabetically according to the shorthand version of the name, within their
respective categories.
Abbreviated Name
Description of Variable
PRIMARY PRODUCER ABUNDANCE
AFDM
benthic ash-free dry mass
PCT  MAP
macroalgal percent cover
PCT  MCP
macrophyte percent cover
PCT  Ml ATI
percent presence of thick (lmm+) microalgae
none
benthic chlorophyll a
none
soft algal total biovolume
NUTRIENT
NH,
                       ammonium
NOX
nitrate + nitrite
SRP
soluble reactive P
                                               27

-------
Table 3.2 (continued)
Abbreviated Name      Description of Variable
TN                     total nitrogen
TP                     total phosphorus
LANDSCAPE- Development
AG_2000_1K            percent agricultural land use within a 1-km radius from sampling site
AG_2000_5K            percent agricultural land use within a 5-km radius from sampling site
AG_2000_WS           percent agricultural land use in catchment
CODE21_2000_1K       percent "Code 21"11 land use within a 1-km radius from sampling site
CODE21_2000_5K       percent "Code 21" land use within a 5-km radius from sampling site
CODE21_2000_WS       percent "Code 21" land use in catchment
                       percent urban land use in catchment within a 1-km radius from sampling
URBAN 2000 IK
      ~     ~          site
                       percent urban land use in catchment within a 5-km radius from sampling
      ~     ~          site
URBAN_2000_WS       percent urban land use in catchment
none                   site disturbance class
LANDSCAPE-Geographic
none                   ecoregion
none                   latitude
none                   longitude
none                   site elevation
none                   watershed area
none                   percent sedimentary geology in the catchment
LANDSCAPE - Meteorological
none                   mean monthly % cloud cover (3-mo antecedent mean)
none                   mean monthly max temperature (3-mo antecedent mean)
none                   mean monthly solar radiation (3-mo antecedent mean)
none                   total precipitation (3-mo antecedent total)
LOCAL PHYSICAL HABITAT (PHab)
PCT_CPOM             percent cover of coarse particulate organic matter in streambed
PCT_FN                 percent cover of fine substrata in streambed
PCT_SAFN              percent sand + fines in streambed
W1_HALL              a riparian disturbance  index; Kaufmann et al. 1999
XDENMID              percent canopy cover
none                   days of accrual (i.e., estimated number of days since last scour event)
none                   mean stream depth
none                   mean stream width
11 "Code 21" encompasses a wide range of land uses primarily characterized by heavily managed vegetation (e.g., low-
density residential development, parks, golf courses, highway medians)
                                              28

-------
Table 3.2 (continued)
Abbreviated Name
Description of Variable
none
slope, reach-level
none
stream discharge
none
stream temperature
WATER CHEMIISTRY (GENERAL)
none
alkalinity
none
conductivity
none
turbidity
3.2.3 Detection of Ecological Thresholds
Different types of ecological threshold exist (Figure 3.4). The relationship between gradient and response can
involve a change in the slope relating the response variable to the stressor gradient (as in the graph on the
left of Figure  3.4), or a change in magnitude of the response variable's value (as depicted in the "step"
response model on the right side;  Brendan et al. 2008). The dashed line in each figure represents a response
threshold. It is important to note by the time the threshold is reached in Figure 3.2, an ecologically significant
change may have occurred in the value of that ALL

Some stressor-response  relationships may  involve a more complex type of change-in-slope  than that
described in Figure 3.4.  For instance, low values of the stressor gradient may be tolerated within a certain
range at the low end of the stressor gradient (i.e., the "reference  envelope") without a concomitant decline
in ALI response. In addition, Cuffney et al. (2010) distinguished between "resistance thresholds" (marked by a
sharp decline in ecosystem  condition following  an initial no-effect  zone)  and "exhaustion thresholds"
(marked by a sharp transition to zero slope at the end of a stressor gradient at which point the ALI response
is  essentially  saturated; Figure  3.5). Furthermore, different ALI  measures,  within or between  biotic
assemblages, can exhibit different thresholds of response to a given stressor gradient depending upon their
varying levels  of susceptibility. In  aggregate, this array represents the response surface of the BCG  (Davies
and Jackson 2006; Figure 3.2).
                                                                        change in
                                                                        magnitude
                                         gradient
Figure 3.4. Examples of types of threshold relationships.
                                                29

-------
                                    gradient (e.g., chlorophyll a)
Figure 3.5. ALI response to a stressor gradient showing the "reference envelope" along with
"resistance" and "exhaustion" thresholds.
Statistical techniques  vary in terms of what types of threshold they are most appropriate for detecting
(Brenden et al. 2008).  We employ multiple statistical analyses in examining thresholds because each method
has a unique set of attendant advantages and limitations and no one technique is universally  accepted
among scientists. Furthermore, when results of different tests converge on similar values, there is  a greater
level of confidence  and less likelihood that a  given result  was merely an artifact of the statistical method
used. We began the study by conducting a small set of initial analyses across a broad swath of the available
ALI measures and biomass/nutrient gradients, then  conducted additional analyses on the subset of ALI-
biomass/nutrient combinations that yielded the strongest relationships, in order to look for  support for
preliminary thresholds identified. Section 3.2.2 and 3.2.4 summarize the techniques used in this section and
what threshold types they can detect. Once we identified thresholds across analyses, we summarized results
by grouping them according to the four ALI categories  listed above, in order to provide a snapshot of the
study's findings within the context of the BCG concept.

For the purposes of this study, we define "endpoints" or "objectives" to refer to policy decisions to regulate
levels that are deemed an unacceptable risk, while "thresholds" refer to the output of statistical  analyses.
The results of this study may be among those that the SWRCB considers in its synthesis of the science that
will support policy decisions. However, the thresholds produced in the course of this study should not be
construed as policy.

3.2.4 Data Sources
The dataset used for the analyses in this chapter is described in detail in Chapter 2, Section 2.2.1. Included
are sites from  both  the  probability  and  targeted  surveys. Table  3.1 lists  the  response variables (ALI
indicators), while Table 3.2 provides a list of the stressor variables, as well as site-specific and landscape-level
co-factors. Sample sizes for analyses varied according to the variables used (Table 3.3).
                                                30

-------
Table 3.3. Sample
sizes.



All Type
Gradient
Chlorophyll a
AFDM
PCT_MAP
PCT_MCP
PCT_MIAT1
TN
TP
NOX
SRP
BMI
545
521
452
452
452
574
582
593
581
diatom
850
819
745
745
744
873
892
931
935
hybrid
784
756
679
679
679
775
767
769
769
soft
804
776
700
700
700
796
788
790
790
In looking for evidence of biomass thresholds for impacts to Alls, our primary focus was on algal biomass as
measured by benthic chlorophyll a concentrations. Chlorophyll  a is a primary line  of  evidence  in  the
recommended NNE endpoints for wadeable streams (TetraTech 2006), and in general, it is the typical means
of quantifying eutrophication of wadeable streams (US Environmental Protection Agency  2000). However,
alternative indicators of stream primary producer abundance were explored 1) to which Alls may be more
directly responsive and/or 2) which may be more directly tied to nutrient impacts. These included AFDM, soft
algal total biovolume, macroalgal percent cover (PCT_MAP), macrophyte cover (PCT_MCP), and  percent
presence of thick (lmm+) microalgae (PCTJV1IAT1), for certain analyses.

For  the BMIs, Alls included the California Stream Condition Index (CSCI), a draft statewide multimetric index
for  stream  assessment recently developed by  Mazor  et  al. (under  review), as well  as a  statewide
"Observed/Expected" RIVPACS-type (Wright et al., 1993) predictive  model based on BMI taxa, also developed
by Mazor et  al. (under review). In addition, several classical metrics based on the BMI community were used.
Finally, community composition data were ordinated using Nonmetric Multidimensional Scaling (NMS) to
reduce the dimensionality of the dataset, allowing the use of NMS axis scores  as response variables that
summarize  information about the BMI community in each sampling site. We made the  assumption that
values  of the All response to the right  of an identified  threshold along an increasing stressor represent
adverse effects relative to the values on the left. Similarly, King and Richardson (2008) used an NMS-based
approach to assess  biological impairment in the Everglades resulting from experimental  P additions.  For
benthic stream  algae, some  Alls were based on indices developed  by Fetscher et al. (2014), which  use
community  composition of  diatoms and/or non-diatom  ("soft")  algae. Although developed  for  use in
southern California streams, they have some applicability in other parts of the state (Fetscher et al. 2013).
Selected metrics that comprise the IBIs, and NMS scores based on diatom community composition, also were
included. Lists of the All variables,  primary producer biomass variables,  and other variables  (landscape,
meteorological, local physical  habitat, and water chemistry) used in the analyses are provided in Tables 3.1
and 3.2.
                                                31

-------
3.2.5 Data Analyses
There are a number of challenges to determining the existence of ecological thresholds. First, the noisiness of
the dataset may interfere with threshold detection by making it difficult to discern whether or where there is
a clear, abrupt change in the response variable along the stressor gradient. Second, different taxa within any
given biotic assemblage may respond somewhat differently to any given stressor (Baker and King 2010).
Finally, multiple factors potentially influence the value of any given All response measure, making it difficult
to ascribe All response solely to the stressor of interest.

Analytical techniques differ in terms of whether and how confounding factors such as other sources of stress
can be taken into account, and also differ in their susceptibility to outliers. As such, we used a variety of
techniques to attempt to mitigate these challenges and seek consensus in results among different techniques
(Dodds et al. 2010; Smucker et  al. 2013a,b). We  also looked at different measures of All within and across
biotic assemblages. In some analyses, we were able to control for potential confounding factors that could
influence All response variables. This was made  possible by the large size of our dataset and the fact that
sites throughout California, and  across varying levels and types of anthropogenic disturbance, were sampled.
In addition, a large number of local physical habitat (PHab), landscape-level  geographic, meteorological
variables, and water chemistry measures were available for most of the sampling sites.

The analytical techniques used for exploring potential biomass thresholds for All response can be grouped
into two broad categories. Table 3.4 provides a summary of the key assumptions, strengths and limitations of
the different approaches:

    • Analyses that use basic species data for evaluating shifts in "raw" community composition
    • Nonmetric multidimensional scaling (followed by classification and regression tree analysis;
     N MS/CART)
    • Threshold indicator taxa analysis (TITAN)
    • nonparametric change point analysis (nCPA)
    • Analyses that include higher-order variables, such as biotic metrics and indices, as integrative
     measures reflecting aspects of community composition
    • Piecewise regression
    • Significant zero crossings (SiZer)
    • Boosted regression trees (BRT; including partial Mantel tests to pre- and post-screen predictor
     variables. Boosted regression tree and partial Mantel tests were also used for examination of
     nutrient and other environmental co-factor  effects on biomass—see Chapter 4).
The following section provides a brief introduction to each analytical technique.
                                                 32

-------
Table 3.4. Summary of analytical techniques used for threshold estimation.
Analytical
Technique
Strengths
Limitations
Type of Threshold
(refer to Figure 3.4)
            Number of thresholds does not
            have to be established a priori but
            can be manually limited by user.
CART       Least absolute deviation method
            can be used to reduce sensitivity
            to outliers. Can handle multiple
            potential predictors of thresholds.
                                  This technique can overfit
                                  classification and regression
                                  trees. Bootstrapping is
                                  desirable to determine
                                  robustness and level of
                                  confidence associated with
                                  solutions. Will find a break-
                                  point whether one exists or
                                  not.
                                   magnitude
            Provides separate change points
            for taxa to allow user to assess a
            community-level change point (if
TITAN       it exists); multiple assessment
            measures are available for
            determining confidence in change
            points
                                  Some degree of
                                  interpretation is involved in
                                  determining what constitutes
                                  a "community-level change
                                  point"
                                   magnitude
Piecewise
Regression
Intuitive, conceptually easy for
non-experts to grasp; provides
several measures of uncertainty
for determining confidence in the
breakpoint
User must specify number of
breakpoints a priori; this
technique will "find a
breakpoint" whether a true
threshold exists or not;
sensitive to outliers
        slope
            No requirement for a priori
SiZer        determination of the number of
            break points
                                  SiZer maps can be difficult to
                                  interpret; output does not
                                  include a numeric threshold
                                  (only visual, subject to
                                  interpretation); no measure
                                  of uncertainty
                                     slope
BRT
Insensitive to data distributions as
well as the presence of outliers,
can fit both linear and nonlinear
relationships, and automatically
handles interaction effects
between pairs of predictors
Partial effects plots are
created using the mean of
other predictor variables so
care must be taken in
interpretation if interactions
exist.
slope (thresholds
identified from partial
dependence plots);
magnitude thresholds
can be deter-mined
through subsequent
CART analysis
                                                 33

-------
Nonmetric multidimensional scaling (NMS)

Nonmetric  multidimensional  scaling  is  an ordination technique  that  reduces  the  dimensionality  of
information  in a dataset in order to  summarize its major gradients. The product of an NMS ordination
conducted on community composition data is a series of  scores. The plots produced provide insight into
similarity in  species composition among samples. The  closer two sample points are to one another within
NMS ordination  space, the more similar they  are in terms of the  types  and proportions of species  they
contain.

NMS analyses were performed separately for BMI and diatom community composition.  Proportion data were
used for both the BMI and diatom-based ordinations. Only sites with at least 450 BMI (or diatom) individuals
were used in the analyses. Furthermore, only taxa that represented at least 1% relative abundance for at
least two sites in the dataset  were included. NMS was run using PC-ORD software (version 6; McCune and
Grace,  2002) with the Bray-Curtis distance measure and "slow and thorough" autopilot mode. This measure
and mode  runs  initial ordinations to determine the  best dimensionality  (stability criterion  of  0.00001,
maximum of six axes, 40 runs  with real data, and 50 randomized runs). A second round of ordinations is run
using the selected dimensionality (stability criterion of 0.00001, one run with  real data,  up to 400 iterations).
From  each NMS ordination,  we selected the  axis  that was  most  strongly associated  with biomass and
nutrients for use as a response variable in subsequent analyses (e.g., CART, see below).

For determining Pearson correlation coefficients that incorporate sample weights to describe the relationship
between NMS scores and biomass/nutrient gradients, we  used the R package "weights" (Pasek and  Tahk
2012).  Sample weights were calculated as the number of stream kilometers represented by  each  sampling
site (weights account for differences in the number of sites in each stratum and stream kilometers in the
stratum; see Chapter 2). To facilitate use of the "weights" package, which provides only Pearson correlation
coefficients,  the non-normal data were first rank-transformed.

Classification and regression tree (CART) analysis

Classification and regression trees (De'ath and Fabricius 2000) is an analytical method that "builds trees"  via a
recursive series of binary splits of the data  set into successively smaller  groups  of observations. Splitting
occurs along one or more explanatory variables, which can be categorical, or continuous. For classification
trees, during each recursion, the split chosen maximizes homogeneity of response values within the resulting
two groups.  We used CART to identify "cut points" (i.e., locations of the split) in explanatory variables,  such
as biomass and nutrients, as an indication of thresholds of their effects on All response variables. The NMS 1
axis scores for the BMI and diatom communities were used as response variables in the CART analyses.

Depending upon the version of CART analysis run, explanatory variables included either all of the following:
Chlorophyll a, AFDM, the different nutrient  types, ecoregion, and site disturbance class (in which  case, the
output of the analysis is referred to as "ALL"), or only a single explanatory variable was used (either AFDM
alone or only chlorophyll a ["CHLA"] alone).  For the latter two versions of the analysis, the number of splits
used in the tree  building was  restricted to two. Because community composition  could vary  geographically
and, therefore, might influence the outcome of the analysis,  CART analyses were run both  statewide and
within the South Coast ecoregion (Figure 2.1), where the highest density of data were available. These latter
groupings of data facilitated an assessment of the  possible effect of biogeographic variation on cut point
values.

                                                34

-------
CART analysis was carried out using SYSTAT v. 13 software (Systat Software, Inc.,  Chicago,  IL)  using the
following options: least absolute deviation (which renders the analysis less sensitive to outliers), a  maximum
number of 2 splits, p-values of 0.05 for total and incremental variance explained and minimum of five objects
in final classes. One thousand bootstrap replicates were run to generate confidence intervals for split values.
In this and subsequent analyses, the number of bootstrap replicates  run is chosen to ensure robust and
reasonably precise results within practical limits.

Threshold indicator taxa analysis (TITAN)

Threshold indicator taxa analysis (TITAN) is an analytical technique that represents a combination of indicator
analysis and change point analysis. TITAN identifies if synchronous declines occur in multiple species along an
environmental gradient12 of interest. To conduct TITAN, we  used the R package "mvpart" (version 1.6-0,
Therneau  and Atkinson 2009) with scripts modified by Baker and King (2010). We used TITAN to look at BMI
and diatom species responses to biomass and to nutrient levels in  order to identify potential thresholds.
TITAN is still  being debated as to the validity of change point values identified and for that reason may be
considered more exploratory. TITAN uses indicator value scores from  indicator species analysis (Dufrene and
Legendre 1997) to integrate occurrence, relative abundance, and directionality of taxa responses. It identifies
the optimum value (i.e., "change point") of a continuous variable, x, that partitions sample units while
maximizing taxon-specific scores. Indicator z scores standardize original  scores relative to the mean and
standard deviation of permuted samples along x, thereby emphasizing the relative magnitude of change and
increasing the contributions of taxa with low occurrence frequencies but high sensitivity to the gradient.
TITAN  distinguishes negative  (z-) and positive (z+) taxa  responses  to the gradient. It  tracks cumulative
responses of declining sum(z-), which we refer to as  "decreasers", and increasing sum(z+) taxa,  which we
refer to as "increasers", in the community. Narrow peaks in sum(z) scores along the environmental gradient
of interest (x-axis) and the presence of many taxa with  change  points at similar levels of that gradient
indicate a community threshold.

Bootstrapping is used to estimate indicator taxon "reliability"  and "purity" as well as uncertainty around the
location of individual taxa and community change  points.  Indicator  "purity" as defined by Baker and  King
(2010) is  the proportion of  change-point  response directions  (positive or negative)  among  bootstrap
replicates that agree with the  observed response. As such,  "pure indicators" are those that are consistently
assigned the same response  direction,  regardless of abundance and frequency distributions generated by
resampling the original  data.  For the purposes of this report, "pure taxa" are defined as those  for which
purity > 0.95. Indicator "reliability" is defined by Baker and King (2010) as the proportion of bootstrap change
points  whose indicator value  scores consistently result in P-values  below one or more user-determined
probability levels.  For the purposes of this report,  "reliable indicators"  are those with repeatable and
consistently large indicator value maxima (specifically, > 0.95 of the bootstrap replicates achieving P < 0.05).
Examples of TITAN output and its interpretation are provided  in Figures C.I and C.2. We used 500  bootstrap
replicates in  order to identify pure  and reliable indicator taxa, and to  establish uncertainty around  taxa
change-points (i.e., 5 and 95% quantiles; Baker and King  2010). In order to downweight the  influence on
12 A multivariate version of this package, that would allow multiple stressor gradients to be used in determining taxon-
specific z-scores, is currently under development (M. Baker, personal communication 2014), and not available for use in
the present version of the report.

                                                 35

-------
indicator values of taxa with high  relative  abundances, both  BMI and diatom  data  were  analyzed  as
Iog10(x+l)-transformed species relative abundances. Taxa with fewer than five occurrences on either side of a
partition (during the TITAN binary partitioning process) were eliminated.

Nonparametric change point analysis (nCPA)

Nonparametric change point analysis (Qian et al. 2003) is an analytical technique used when a step function
(i.e., change in magnitude, as described in Figure 3.4) is assumed.  It seeks the point along a gradient at which
the sum of the deviance in the response variable, to the left of the point, plus the deviance to the right of the
point, is maximally lower than that across  the data set as a whole. The sum  of the  deviance values is
calculated iteratively  along the gradient, and the point at which  maximal deviance reduction is  realized
reflects a community-level change-point (or threshold) in  the relationship.

To conduct  nCPA on  BMI and diatom  community data, we used the R package "mvpart" (version 1.6-0,
Therneau and Atkinson 2009) with scripts modified by Baker and  King (2010). All response variables for the
nCPA analyses  were  based on  Bray-Curtis  and  Euclidean distances as the dissimilarity metrics  for the
community data.  In addition, 5th and 95th quantiles were determined using 500  bootstrap replicates.  Data
were prepared for analysis in the same way as described above for TITAN.

Piecewise regression

We used piecewise linear regression to detect change  in slope  in the relationship between All response
variables and biomass/nutrient gradients  to  search for possible "breakpoints"  in the response  of  each
available All variable to biomass and nutrient gradients (Muggeo 2003). Before running piecewise regression
analysis, scatterplots for all Alls against the various biomass and nutrient gradients were visualized (Muggeo
2008), and there were no cases  in which it was clear, based on the plots, that >1  breakpoint was present.
Therefore, as a conservative default, all analyses were run coercing a single breakpoint. Piecewise regression
was  one of  the few  analyses for  which sample weights  could be incorporated. For  each  All/gradient
combination, the analyses were run both with and without incorporating sample weights as described above.

Because piecewise regression will always "find a break point" whether or not one truly exists, we created a
set of four criteria against  which to evaluate the output of each analysis, in order to distill the full list of
All/gradient  combinations  into a subset for which high confidence  could be  ascribed to the  breakpoints
identified. Two levels of criteria were employed: "strict" and "relaxed". In order for the piecewise regression
output for a  given All/gradient combination to be assigned to one of these levels: 1) it had to result in a
significant Davies' (1987)13 test (indicating that the slopes on either side of the break point were significantly
different from one another at a = 0.05); 2) at least one of the two slopes had to  be significantly different
from zero (as assessed by ensuring that the 95% Cl [confidence interval] around at least one of the slopes did
not straddle zero); 3) the Cl around the  break point had to be sufficiently narrow (i.e., the Cl width divided by
the breakpoint value had to be <0.5 for the "relaxed" level, and <0.3 for the "strict").; and 4) the adjusted R2
for the regression had to be sufficiently high (i.e., at least 0.1 for the "relaxed" level, and at least 0.25 for the
"strict"). To conduct piecewise regressions, we used the R package "segmented"  (Muggeo 2008).

Significant Zero crossings (SiZer)
13 Using a standard "k" value of 10, per Muggeo (2008).

                                                 36

-------
The "significant zero crossings" analytical technique fits  multiple smoothing  curves through a scatterplot
using locally weighted polynomials, with the goal of assessing the nature of the first derivative of each curve
(indicating the direction of slope) at intervals along a gradient (Chaudhuri and  Marron 1999). SiZer attempts
to distill  the significant, "real" features of a  curve  in a dataset by "looking  past" any noise that may be
present. To accomplish this, curve smoothing of the relationship of a response variable (e.g., an All measure)
to a gradient (e.g., a biomass or nutrient gradient)  is  conducted at various bandwidths. The bandwidth is the
ranges along the x-axis over which the polynomial smoothing is conducted. Unlike the approach for piecewise
regression using the "segmented" package, SiZer does not require the user to propose a priori the number of
breakpoints upon which to base output. Thus the data can freely "speak for  themselves" as to how many
thresholds may be present.

The output of a SiZer analysis is a "SiZer map", which  uses color coding to allow users to visualize where
along the gradient (the x-axis) the first derivative  is  significantly positive (depicted by blue) and where it is
significantly negative (depicted by red) for different curve-smoothing bandwidths (which are represented on
the y-axis as "h"  (Iogi0-tranformed). Areas  in  which the derivative is neither increasing nor decreasing
significantly are indicated by purple, and grey means that data are insufficient to make a determination for
that gradient-bandwidth combination. SiZer does not explicitly provide threshold values, but the SiZer map
supplies the user with the means to make inferences. Specifically, a point along the stressor gradient where a
narrow band of transition  from purple to red, purple to blue, or vice versa  is consistent across many
bandwidths is a  compelling  indication  of a  significant and  robust change  in  slope (and corresponding
threshold of response). Examples  of SiZer maps  and their  interpretation are provided in Figure C.3. To
produce SiZer maps, we used the R package "SiZer" (Sonderegger 2011).

Boosted regression trees (BRT)

With the exception of the CART analysis, the analyses described above all involved looking at  response of
some form of All  to a single independent variable (i.e.,  chlorophyll a concentration, or some other biomass
or nutrient measure). In order to look at the effect  of biomass/nutrients within  the context of other potential
predictors,  which  could confound All responses, and to facilitate an evaluation of the relative importance of
biomass/nutrients as compared to other potential determining factors, we employed boosted regression tree
analysis. BRT analysis was used for two purposes in this report to assess:

   • Biomass and nutrient relationships with Alls, as well as to look for evidence of thresholds of
     response to biomass/nutrients, while holding other predictors constant
   • Nutrient and other environmental co-factor relationships with biomass of various types
One of the daunting aspects of determining nutrient effects on stream primary producer biomass is the fact
that nutrients do  not act in isolation. Rather, their influence on biomass is mediated by any of a number of
environmental co-factors, which can limit the potential for biomass accrual even when nutrient levels  are
high. As such, determining the influence of nutrients on biomass requires accounting for the effects of the co-
factors. To this end, we used BRT analysis to investigate nutrient effects on biomass levels in conjunction with
other environmental co-factors.

BRT combines the strengths of regression trees with a machine-learning algorithm called "boosting", which is
an adaptive method for combining many simple models to give  improved predictive performance. The final
BRT model is essentially an additive regression model in which individual terms are simple trees, fitted in a

                                                 37

-------
forward, stage-wise fashion. BRTs randomly add predictor variables and identify "relative influence" of each
predictor based on how often it is selected and whether it improves the model. Advantages of BRT analysis
are that it is insensitive to  data distributions (thus requiring no transformation) as well as the presence of
outliers, can  fit both  linear and nonlinear relationships,  and automatically handles interaction effects
between pairs of predictors. BRTs can also be used to plot the "partial dependence" of the response variable
on an  individual predictor, which is a way of looking at the relationship of  the  response variable to  the
predictor when all other predictors are held constant, at their mean values in the dataset (Elith et al. 2008).

Rather than conducting BRT analyses on  all possible All response variables, we selected a cross-section of
variable  types from each  assemblage  and  sought to reduce redundancy.  For example,  for  the BMI
assemblage, only one type  of diversity index was used, as well  as only one metric  each from the BMI  "trio"
metric groups (i.e., Taxa, Percent, and PercentTaxa).

BRTs were run with tree complexity = 5,  learning rate = 0.001, and bag fraction = 0.5, and all final models
were built with >1000 trees, the number of which was optimized per model (for most models, except where
noted) to maximize model  performance while reducing overfitting. A  10-fold cross-validation procedure
without  replacement (90% training, 10% validation) was  employed  that used  all data for training and
validation steps (Elith et al. 2008). We utilized the model-simplification procedure described in Elith  et al.
(2008) to reduce the number of predictor variables in the final model for each ALL  For BRT analysis, we used
the R package "dismo" (Hijmans et al. 2013).

Partial Mantel tests

Because some of the predictors available  for use in the BRT analyses (i.e., the  landscape variables at nested
spatial scales; Table 3.2) had a high  likelihood of being correlated with one another, we used partial Mantel
tests in a "prescreening" step to determine if effects of  the  landscape variables at  each scale  could  be
detected after accounting for other scales. Any non-significant  land-use  variables were then excluded from
the BRT analyses for that  ALI type. We  also  used  partial  Mantel tests for "post-screening" the  suite of
predictor variables remaining in each final BRT model after having  completed the BRT model-simplification
procedure (see above)  in  order to determine which variables had significant partial Mantel correlation
coefficients when the other predictor variables from the final BRT models were taken into account. This was
accomplished by including  all top-ranked (i.e., those with  the highest  relative influence) biomass and/or
nutrient predictor variables, as well  as any non-nutrient/non-biomass predictors that ranked above them. A
geographic-distance  variable was  also  incorporated,  in  order  to  evaluate the  potential for spatial
autocorrelation. Note that  the Mantel test can return erroneously low  p-values in the presence of spatial
autocorrelation, thus it is important to rule out.

To prepare data for the partial Mantel tests, we first transformed all non-normal data, using arcsine-square-
root for proportion  data (such as land use) and logic for other data types (such as  chlorophyll a, AFDM, and
nutrient concentrations). We also included information on  geographic distance among sites in order to test
for potential  spatial autocorrelation (hereafter  referred to as "space") in the ALI relationship  with  the
nutrients/biomass and other variables (King et al. 2005). To accomplish this, we first transformed latitude and
longitude into Universal  Transverse Mercator (UTM)  coordinates.  Euclidean  distance matrices were then
calculated for each variable based on the transformed values. For partial Mantel tests, we used the R package
"ecodist" (Goslee and Urban 2007).
                                                 38

-------
 • ."•'
This section provides results from each statistical technique we employed, organized analysis-by-analysis. The
final  part of the results section provides a summary of all thresholds identified per All category, to facilitate
visualization and interpretation of study results within the context of the BCG concept.

3,3,1         Diatom Responses to             Nutrient Gradients      on Shifts in Community
       Composition

NMS and CART
NMS was used on BMI and diatom  data to depict sampling site relationships to one another based on
community composition. Proximity of sites to one another along an NMS axis is an indication that those sites
share a more similar  community composition than sites that are further away in ordination space. The same
NMS ordination axes from this analysis were used as response variables in  CART models (see Methods). The
spline fits to NMS and subsequent CART analyses with NMS scores as the response variable revealed similar
qualitative  (NMS)  and quantitative  (CART) thresholds for the biomass  and  nutrient stressor gradients
examined. CART-derived thresholds were  slightly higher than perceived resistance thresholds on NMS spline
plots  and well  below  perceived  exhaustion thresholds.  CART analyses carried out with all explanatory
variables ("ALL") generally included AFDM, TN, or TP as splitting variables, but chlorophyll a was rarely
included.

All four versions of the NMS analyses resulted in 3-axis solutions. In each case, NMS axis 1 had the strongest
relationship with biomass and nutrients  and was therefore selected. For the BMI  analyses statewide, final
stress was 17.7 and  percent variance explained by NMS 1 was 28.7 while the results were 17.2 and  29.3,
respectively, for the South Coast. For the diatom  analyses, final stress was  19.2 and percent variance
explained by NMS 1 was 27.3 for the statewide data set, and 19.7 and 20.6, respectively, for the South Coast.
Scatterplots  of  statewide  NMS axis 1 against biomass and  nutrient gradients are provided in Figures 3.6
and 3.7.

All graphs show consistent and significant  relationships between biomass/nutrient gradients and NMS scores,
indicating that  sites that share  similar biomass/nutrient  concentrations are  also  similar in   species
composition.  Pearson correlation coefficients  indicate  that, for  both assemblages, the relationships are
strongest for nutrients, chlorophyll a, and AFDM, whereas  the weakest relationships  are between  NMS
scores and  the percent  cover  metrics.  Particularly strong are  the  relationships between the  diatom
community composition and  nutrients (especially TP). Furthermore, the scatterplots show that the  most
pronounced relationship between the diatom  community (NMS axis 1) and TP occurs between  a lower
qualitative threshold of approximately 0.01 and a higher  one at 0.1 mg/L,  whereas for TN,  the  most
pronounced relationship for  both diatom  and BMI communities  occurs  between a  lower threshold  of
approximately 0.1 and  a  higher one at 1 mg/L. These observations are corroborated by the results of the
CART analyses of diatom and BMI NMS axis 1 scores (Table 3.5, Figure 3.8), in which median cut point values
for TP  and TN were consistently <0.1 and <1 mg/L, respectively, and closer to visually perceived resistance
thresholds in spline fits. All median cut points for chlorophyll a were <31 mg/m2,  and for AFDM were <42
g/m2. Note that CART analyses carried out with all explanatory variables ("ALL") generally included AFDM, TN,
or TP as splitting variables, but chlorophyll a was rarely included. Ecoregion and site disturbance class  were
never included in final trees.
                                                39

-------
        r = 0.29

  m
     o-
    '-1 -
           0.1      1      10     100
               chlorophyll o (mg/m2)

        r = 0.19
  CD
  CO
     o-
    -1 -
                       10
                  PCT_MAP
                     100
        r = 0.30
  CD
  co
     0-

         0.01
0.1        1
  TN (mg/L)
10
                                   = 0.35
                                                 1 -
                                              CO
                                              CO
                                                 o-
                                                -1 -
                01


                 = 0.19
                                                    10       100
                                            AFDM (g/m2)
                                              co
                                10
                          PCT_MCP
                                  100
                                 r = 0.24
                                                 i -
                                              DO
0001     001      0.1
              TP (mg/L)
Figure 3.6. Scatterplots and splines for non-metric multidimensional scaling (NMS) axis 1 values from
the benthic macroinvertebrate (BMI) community against biomass and selected cover and nutrient
gradients on log scale, using the statewide data set. The Pearson correlation coefficient for each
relationship is provided to the upper left of each graph. Correlation analyses were performed on rank-
transformed data, and sample weights were used in the analyses. All relationships were highly statistically
significant (p <0.0001). PCT_MAP and PCT_MCP are percent cover of macroalgae and macrophytes,
respectively.
                                              40

-------
         f = 0.42
                                  r = 0.50
            01     1     10     100
                 chlorophyll o(mg/m2)
            0.19
                         10
                   PCT_MAP
          = 0.54
    I
      o-
    05
                -.'
          0.01
0.1        1
  TN (mg/L)
10
                                                p 11
                                                ro


                                                T-" 0"
                                                O>
                                  0.1
                                   = 0.24
            1       10
             AFDM (g/m2)
                                          100
                                                o , .
                                                "5
                                                CO
                                                  10
                                            PCT_MCP
                                  r = 0.63
                            I1"1
0.001     0.01     01
              TP (mg/L)
                                               100
Figure 3.7. Scatterplots and splines for NMS axis 1 values from the diatom community against
biomass and selected cover and nutrient gradients on log scale, using the statewide data set. The
Pearson correlation coefficient for each relationship is provided to the upper left of each graph. Correlation
analyses were performed on rank-transformed data, and sample weights were used in the analyses. All
relationships were highly statistically significant (p <0.0001). PCT_MAP and PCT_MCP are percent cover of
macroalgae and macrophytes, respectively.
                                              41

-------
Table 3.5. Results of CART analyses with NMS axis 1 scores for either the BMI or the diatom community
as the response variable. Separate analysis were run for the statewide dataset and for the South Coast
ecoregion. Model runs included either the full set of explanatory variables ("ALL", see Methods), or
chlorophyll a or AFDM alone. Cut points are the median values, from 1,000 bootstrap runs, at which the
first split in the indicated splitting variable was made during tree building. "Frequency" refers to the
number of bootstrap replicates in which the variable in question was the splitting variable at the first node.
Splitting
Variable
Chlorophyll a
(mg/m2)



AFDM (g/m2)





TN (mg/L)


TP (mg/L)


NH4 (mg/L)
SRP (mg/L)


Cut Point (95% Cl)
6.2(6.2-6.2)
23.6(4.1-61.9)
21.9(11.4-30.1)
30.8 (2.7 - 86.0)
23.6(4.1-61.9)
23.6(4.1-61.9)
12.6 (4.8 - 35.8)
30.8(4.1-88.7)
41.8(6.0-159.3)
25.2 (4.0 - 75.0)
25.9 (3.2 - 103.3)
18.5(3.6-54.1)
18.5(3.6-54.1)
0.29(0.09-0.75)
0.65 (0.22 - 1.8)
0.61(0.12-2.2)
0.60(0.18-1.7)
0.058(0.021-0.12)
0.055(0.017-0.12)
0.080 (0.012 - 0.25)
0.070(0.01-0.19)
0.013 (0.008 - 0.018)
0.045(0.005-0.18)
0.080(0.055-0.12)
0.074 (0.016 - 0.20)
0.078(0.012-0.15)
Assemblage
BMI
BMI
BMI
BMI
diatom
diatom
BMI
BMI
BMI
BMI
diatom
diatom
diatom
BMI
BMI
diatom
diatom
BMI
BMI
diatom
diatom
BMI
BMI
BMI
diatom
diatom
Region
statewide
statewide
South Coast
South Coast
statewide
South Coast
statewide
statewide
South Coast
South Coast
statewide
statewide
South Coast
statewide
South Coast
statewide
South Coast
statewide
South Coast
statewide
South Coast
statewide
South Coast
South Coast
statewide
South Coast
Explanatory
Variables in
Model
ALL
chlorophyll o
ALL
chlorophyll o
chlorophyll o
chlorophyll o
ALL
AFDM
ALL
AFDM
ALL
AFDM
AFDM
ALL
ALL
ALL
ALL
ALL
ALL
ALL
ALL
ALL
ALL
ALL
ALL
ALL
Frequency
1
810
3
173
810
810
634
239
59
988
50
840
840
179
42
554
82
9
435
221
523
2
11
43
21
304
Model
Fit
0.21
0.14
0.15
0.09
0.14
0.14
0.22
0.10
0.18
0.13
0.24
0.15
0.15
0.22
0.19
0.25
0.24
0.19
0.21
0.25
0.25
0.19
0.18
0.20
0.24
0.25
                                                42

-------
              chlorophyll o (mg/m2}
75-
50-
25- 1

*- 2 0-
c
o> 1.5-
"S 1.0-
0) 0.5- 1
o.o-


0.15-
0.10-
0.05-
n nn-


' 4
TN(mg/y



j

NH4 (mg/L)



k


i




L





A
150-
100-
' 50" i 1'
	 o- 	 — 	 ^-
IIIIIIIl ^^^^TPimg/LEI
0.25-
0.20-
0.15-
0.10- I I
JO.05- f f
1 0.00 -, 	
i | P04 (mg/L)
{0.20-
0.15-
0.10- 1
1 0.05-


L
-~~~~~~~~~~~~~~~~~~~~~~< assemblage
"
diatoni

analysis_type
• AFDM
A ALL

j • CHLA




            statewide
                         South Coast
                                                statewide
                                                             South Coast
                                     region
Figure 3.8. Cut points from CART analyses using NMS axis 1 scores from either the BMI or the
diatom community as the response variable. Cut points are the median values, from 1,000 bootstrap
runs, at which the first split in the indicated splitting variable was made during tree building. Error bars
correspond to cut point 95% confidence intervals. Separate analyses were run for the statewide  dataset and
for the South Coast ecoregion. Model runs included either the full set of explanatory variables ("ALL", see
Methods), or chlorophyll a ("CHLA") or AFDM alone. Y-axes correspond to the stressor gradients, which are
labeled in the upper strip of each panel.

TITAN and nCPA
TITAN and nCPA were used on BMI and diatom community composition  data in order to detect change points
in biotic response along biomass and  nutrient stressor gradients. Based on the nCPA results  and results for
the TITAN "decreaser" taxa: Chlorophyll a change points were always <27 mg/m2, AFDM change points were
always <13 g/m2, TN change points were always <0.5 mg/L, and TP change points were always <0.09 mg/L.

Table 3.6 provides the mean change points (i.e., points along biomass/nutrient gradients where taxa  show
the greatest  change in frequency  and relative abundance, and which, therefore, can be  interpreted as
thresholds) derived from the nCPA analyses.  Only results for the  pure and reliable taxa from the TITAN
analyses  are included. Table C.I provides TITAN change point values for individual taxa. TITAN change points
for  "increaser" taxa were  invariably higher, sometimes substantially so, than  those for  "decreaser" taxa.
Numbers of pure and reliable taxa were low for the percent cover ALIs relative to the other biomass/nutrient
gradient types. For macroalgal percent cover (PCTJV1AP), change points from nCPA and TITAN "increasers"
were all <36%, and for macrophyte percent cover (PCT_MCP), change points were all <19%. Figures 3.9-3.11
show examples of TITAN and nCPA change points for BMI and diatom communities along several biomass
and nutrient gradients, and Figure 3.12 provides a graphical summary  of change points from all TITAN and
nCPA analyses. TITAN analyses show a narrow range of response to AFDM and TP for sensitive (decreaser)
taxa with relatively narrow confidence intervals. Appearance of tolerant (increasing) taxa was more gradual
with much wider  confidence intervals (Figure 3.10). Overall,  BMI community composition showed a very
sharp threshold of response  along a gradient of TN, while responses along  gradients of macroalgal and
macrophyte cover were more diffuse (Figure 3.11).
                                                43

-------
Table 3.6. TITAN and nCPA results for BMI and diatom community composition data. Included are change
points from TITAN (sum[z+] and sum[z-]) analyses and nCPA analyses based on Euclidean and Bray-Curtis
distance measures (nCPA.euc and nCPA.bc). Also provided are quantiles (tau = 0.05 through 0.95) of each
estimated change point distribution. The values provided for the TITAN analysis are mean values among
only the "pure" and  "reliable" taxa (see  Methods for more details). "Tau = 0.95, max." is the tau = 0.95
value for the taxon  (among the pure and reliable taxa) that had the highest tau = 0.95 value for the
analysis in question.
Gradient
Chlorophyll
(mg/m2)





AFDM (g/m






PCT MAP
(%)





PCT MAP
(%)






Analysis Type
o TITAN
TITAN

nCPA.

nCPA.

2) TITAN

TITAN

nCPA.

nCPA.

TITAN

TITAN

nCPA.

nCPA.

TITAN

TITAN

nCPA.

nCPA.

.decreasers
.increasers

euc

be

.decreasers

.increasers

euc

be

.decreasers

.increasers

euc

be

.decreasers

.increasers

euc

be

Assemblage #Taxa
BMI
Diatom
BMI
Diatom
BMI
Diatom
BMI
Diatom
BMI
Diatom
BMI
Diatom
BMI
Diatom
BMI
Diatom
BMI
diatom
BMI
diatom
BMI
diatom
BMI
diatom
BMI
diatom
BMI
diatom
BMI
diatom
BMI
diatom
92
72
23
61
217
409
217
409
90
65
34
100
217
408
217
408
74
47
24
35
203
387
203
387
52
23
23
80
203
387
203
387
Change
* Point
16.51
16.06
43.40
83.98
15.98
26.73
17.15
26.73
7.05
10.19
23.06
73.01
11.42
7.80
10.86
12.71
13.53
11.63
25.96
37.44
35.00
14.00
19.78
14.00
5.37
3.46
34.92
29.89
18.05
7.00
18.05
7.00
i
0.05
4.45
4.54
7.90
18.78
5.63
12.13
10.28
13.88
3.27
4.26
6.67
14.74
6.13
5.34
6.25
5.98
2.00
1.99
8.16
15.10
12.00
7.81
11.00
12.00
0.29
0.17
12.54
6.80
6.00
2.00
6.00
2.00

0.1
6.02
6.19
10.01
27.31
6.15
12.98
11.03
15.21
3.80
5.10
7.45
20.96
6.61
5.59
6.61
6.80
3.60
2.62
11.35
19.10
15.00
9.69
16.00
12.92
0.66
0.55
16.02
10.02
8.00
2.86
8.00
2.86

0.5
13.28
14.13
37.75
69.01
17.37
26.65
17.15
28.39
6.67
9.42
20.78
60.52
10.92
8.23
10.91
11.61
12.66
10.80
27.57
37.81
33.00
18.10
23.00
18.00
5.20
4.21
32.99
28.83
14.00
6.26
15.00
6.92
tau =
0.9
22.83
26.92
84.58
188.87
49.28
49.55
20.99
50.73
10.85
16.48
55.71
130.20
16.10
23.08
12.52
21.87
25.82
23.55
56.10
58.62
60.05
36.84
39.00
35.65
15.72
13.81
52.27
54.54
20.98
18.07
21.00
14.57


0.95 (max)0.95
27.74
32.26
108.32
233.19
57.22
55.87
22.87
56.84
13.04
18.72
71.87
153.92
18.75
26.81
12.93
24.70
29.99
28.34
60.86
65.32
70.74
37.07
41.37
36.97
18.97
17.19
56.66
60.60
23.09
26.13
25.53
26.68
95.17
108.96
580.45
747.25
-
-
-
-
52.91
66.52
185.37
304.33
-
-
-
-
82.95
73.78
89.26
93.00
-
-
-
-
59.02
41.68
80.02
82.00
-
-
-
-
* Number of taxa, for TITAN, is the number of pure and reliable taxa, not the total number of taxa evaluated in the analysis
  (the latter of which is the same number as that provided in the corresponding nCPA analysis).
                                                 44

-------
Table 3.6 Continued.
Gradient

TN
(mg/L)

TP
(mg/L)

Analysis Type
TITAN. decreasers
TITAN. increasers
nCPA.euc
nCPA.bc
TITAN. decreasers
TITAN. increasers
nCPA.euc
nCPA.bc
Change
Assemblage #Taxa* Point
BMI
diatom
BMI
diatom
BMI
diatom
BMI
diatom
BMI
diatom
BMI
diatom
BMI
diatom
BMI
diatom
117
96
31
103
220
407
220
407
103
68
21
98
220
406
220
406
0.20
0.29
1.49
1.75
0.32
0.37
0.32
0.48
0.04
0.04
0.11
0.18
0.08
0.03
0.08
0.05
t
0.05
0.08
0.07
0.27
0.43
0.29
0.32
0.25
0.33
0.01
0.02
0.04
0.05
0.04
0.02
0.04
0.02
tau =
0.1
0.09
0.09
0.38
0.56
0.30
0.34
0.26
0.35
0.01
0.02
0.05
0.06
0.05
0.02
0.05
0.03
0.5
0.18
0.25
1.16
1.48
0.37
0.45
0.31
0.47
0.03
0.04
0.09
0.15
0.08
0.03
0.08
0.05
0.9
0.34
0.53
3.57
4.63
0.54
0.60
0.45
0.60
0.06
0.06
0.41
0.45
0.10
0.06
0.10
0.09
0.95 (max) 0.95
0.39
0.63
4.52
5.91
0.59
0.61
0.48
0.63
0.08
0.07
0.61
0.56
0.12
0.08
0.11
0.10
1.33
4.12
13.96
16.42
-
-
0.46
0.35
2.07
2.07
-
_
* Number of taxa, for TITAN, is the number of pure and reliable taxa, not the total number of taxa evaluated in the analysis.
                                                      45

-------
                               o
                               o
                                           X,
                                       100   200   300   400   500
                                        chlorophyll a (mg/m2)
                               o
                               o
                               CD
                               s
                               CM
                               O

                                          50      100     150
                                            AFDM (g/m2)
                                                                      o
   Z5
   CT
CO CD

  •V
   O
   
-------
                             50      100     150     200    250
                               benthic ash-free dry mass (g/m2)
                                     300
                     0.0
0.5
  1.0
TP (mg/L)
1.5
2.0
Figure 3.10. Plots  of taxon-specific change from TITAN analysis of diatom community data along
AFDM and TP gradients. Black plots refer to sum(z) scores for "decreased taxa, and red plots correspond
to "increaser" taxa. Horizontal lines overlapping each symbol represent 5th and 95th percentiles from 500
bootstrap replicates. See Table C.1  for lists of the decreaser and increaser taxa (too numerous to label
legibly here), their individual change point values, and the ranks thereof.
                                               47

-------
                     c
                     .Q ID
                     Q
                     (U

;






c


!l
jl
\
\
C
\.
^••w/v
i i i 1*1 n!
2 4 6 8 10 12 14
o
co £
o OJ
«!
0 "§
i
SJ
ci aj
j3
3 1
o
o
O

                     5
                       If)
                       Q
                       m
                       c4
                     t3
                     (Li
                     D
                     Q O
                       lO
                       O
                                       TN (mg/L)
 ?     i
u.
                                20
              I	I	l_
              40     SO     SO
            macros Igal % cover

                                                           100
    ff>'h   •
   S     /' .-

  ^    ;^
                                              :^.
                                                                D"
                                                                P
                          0   10  20  30  40  50  60  70  80

                                  mar.rnnhvte
Figure 3.11. Nonparametric change point analysis (nCPA) results. Shown are deviance reduction values
across TN, macroalgal percent cover (PCT_MAP), and macrophyte percent cover (PCT_MCP) for the BMI
community (distance measure = Bray-Curtis).
                                           48

-------


200-
150-
100-
50-
o-

- 60-
«40-
CD
0) 20-
o-

6-
4-
2-
n-
chlorophyll a (mg/m2)





.0 \
0



•6 j


B
PCT MAP(%)


I I *
TV
1
9 1 <
n li 1
TN (mg/L)



<


> i
fi



»?


•o y *o •©

150-

100-
50-
n-

60-
HU
20-
n-

0.6-
0,4-
0.2-
n n-
AFDM (g/m2)



, (
• 0




i
•o •<•>
PCT MCP (%)

<
li

>f


s
' 5 tt
TP (mg/L)



i



i


)
•o v ' *0 »6
                                                                                    assemblage
                                                                                    ••• BMI
                                                                                    -Q diatom
        TITAN.dec TITAN.inc  ncpa.bc   ncpa.euc        TITAN.dec TITAN.inc  ncpa.bc  ncpa.euc
                               change-point analysis type

Figure 3.12. Summary of TITAN and nCPA change points along biomass/nutrient gradients, based on
BMI and diatom community composition using the statewide dataset. Vertical lines associated with
each change point represent 5th - 95th percentiles from 500 bootstrap replicates. TITAN values represent
means among pure and reliable taxa. Y-axes correspond to the stressor gradients, which are labeled in the
upper strip of each panel.

3.3.2 Biotic Responses to Biomass Gradients Based on Shifts in Integrative Measures of Community
        Composition (Metrics and Indices)
Piecewise regression and SiZer

Piecewise regression and SiZer are different approaches to evaluating relationships (and identifying potential
thresholds or response) between biomass and nutrient stressor gradients and Alls. We used a variety of Alls
that included both metrics and more integrative indices, such as IBIs, as opposed to the previous analyses
that focused on "raw" community data. Chlorophyll a breakpoints, as estimated via piecewise  regression,
ranged from approximately 25 to 150 mg/m2. For AFDM, over half of the All breakpoints  were estimated (in
the unweighted analyses) to be <20 g/m2. Estimated TN breakpoints from unweighted piecewise  regressions
ranged from approximately 0.5 to  1.1 mg/L, while those for TP ranged from 0.075 to 0.12 mg/L. There was
generally a high degree of correspondence between the piecewise regression output and the SiZer map for
the various ALI/gradient combinations.

Table 3.7 provides a summary of the piecewise regression analysis output for all  ALI/gradient combinations
for which at least one member of each analysis pair (weighted/unweighted) passed all four "strict" evaluation
criteria; Table C.2 is an extended version, providing  the output for alj_ALI/gradient combinations. Breakpoint
estimates arising from analyses not incorporating sample weights were  almost invariably lower than  those
with weights (and CIs for the latter tended to  be broader). Furthermore, analyses including weights were less
likely than those without weights to result in output that successfully met all four screening criteria (even for
the "relaxed" version).
                                                49

-------
For chlorophyll a breakpoints, over half of the Alls have values of <100 mg/m2 (among the analyses that did
not incorporate sample weights). However, there was little agreement among All  variables, and CIs were
generally broad, especially for the Alls with the higher breakpoints (Figure 3.13). Furthermore, breakpoints
were generally not well supported, as few of them  (N=10; Table C.2, Table 3.7) passed the "relaxed", and
none passed the "strict", screening criteria. Nonetheless, of those that did achieve the "relaxed" criteria, all
four assemblages were represented. Breakpoint values were still highly variable, ranging from  23 to 113
mg/m2. In general, there was a  high degree of interdigitation of breakpoints among assemblages (e.g., BMI
Alls were represented across the full range of values  generated; Figure 3.13). Exhaustion thresholds occurred
at lowest levels for SENS indicators, followed by EUTR indicators (decreased DO, increasing saprobicity, then
increasing green algal  biovolume and finally  nuisance green algae), NUTR indicator taxa, and finally INT
indicators.  Sensitive and eutrophication indicators tended  to have the  smallest Cl  for  chlorophyll  a
breakpoints as compared to nutrient and integrative indicators.

Table 3.7. Summary of piecewise regression results for all All response types for which at least one
version of the analysis (weighted or unweighted) fulfilled  all four "strict" criteria, as described in the
Methods.
Analysis
Gradient Response Type
unweighted
TN (mg/L) D18
weighted
unweighted
EPT_PercentTaxa
weighted
unweighted
EPT_Taxa
weighted
unweighted
H20
weighted
unweighted
H21
weighted
unweighted
H23
weighted
TN ( /L) Intolerant-Percent unweighted
1 3X3
weighted
Breakpoint (SE), Slope 1 Slope 2
95% Confidence (95% Confidence (95% Confidence
Interval Width Interval) Interval)
0.88 (0.07),
0.26
1.29(0.13),
0.50
0.68 (0.04),
0.16
0.72 (0.06),
0.22
0.63 (0.04),
0.14
0.62 (0.05),
0.21
1.06 (0.06),
0.25
1.29(0.12),
0.46
0.68 (0.05),
0.18
1.19(0.12),
0.47
0.77 (0.04),
0.18
1.21(0.11),
0.45
0.62 (0.04),
0.15
0.58 (0.05),
0.19
-45.66
(-53. 16 --38.16)
-34.89
(-39.88 --29.91)
-0.56
(-0.63- -0.48)
-0.55
(-0.62- -0.47)
-27.25
(-30.69 --23. 81)
-31.09
(-35.72 --26.46)
-40.18
(-45.02 --35. 33)
-32.65
(-36.85 --28.46)
-58.63
(-67.41 --49.85)
-35.13
(-40.14 --30.12)
-56.32
(-63.40 --49.25)
-34.86
(-39.56 --30.16)
-0.57
(-0.65- -0.50)
-0.65
(-0.75- -0.55)
0.38
(-0.22-0.97)
0.25
(-1.54-2.04)
0.00
(0.00-0.01)
0.00
(-0.01-0.01)
0.01
(-0.22-0.23)
-0.07
(-0.81-0.67)
0.29
(-0.22-0.80)
-0.14
(-1.64-1.37)
-0.19
(-0.72-0.34)
-0.33
(-2.02-1.37)
-0.18
(-0.69-0.33)
-0.31
(-1.90-1.27)
0.00
(-0.01-0.00)
0.00
(-0.02-0.01)
Adjusted
R2
0.37
0.31
0.59
0.46
0.60
0.41
0.53
0.39
0.46
0.34
0.53
0.36
0.57
0.41
all 4
criteria
(relaxed)
fulfilled?
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
all 4
criteria
(strict)
fulfilled?
yes
no
yes
no
yes
no
yes
no
yes
no
yes
no
yes
no
                                                 50

-------
Table 3.7 Continued
Analysis
Gradient Response Type
unweighted
lntolerant_Taxa

weighted
Breakpoint (SE), Slope 1 Slope 2
95% Confidence (95% Confidence (95% Confidence
Interval Width Interval) Interval)
0.
0.

.52 (0.03),
0.13
.51 (0.04),
0.18
-25.78
(-29.45 --22. 10)

(-36
0.83 (0.06),
unweighted
S2

weighted

unweighted
Taxonomic_Richness

weighted

Tolerant_Percent unweighted
Taxa
weighted
TP (mg/L) D18 unweighted

weighted

H20 unweighted

weighted

H23 unweighted

weighted

RAWIowP unweighted

weighted

SRP(mg/L) H20 unweighted

weighted



0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.24

.93 (0.14),
0.53
.71 (0.05),
0.19
.71 (0.07),
0.27
.67(0.05),
0.18
.73 (0.06),
0.23
.12(0.01),
0.03
.14(0.01),
0.05
.11(0.01),
0.03
.13(0.01),
0.05
.11(0.01),
0.03
.14(0.01),
0.05
-31.35
.28 --26.43)
-52.74
(-60.75 --44.72)


(-43

(-35


-34.70
.39 --26.01)
-31.26
.72 --26.80)
-33.33
(-38.93 --27.72)

(0

(0

0.42
.35-0.48)
0.41
.35-0.46)
-352.10
(-406.00 --298.20)

-290.00
(-335.00 --245.00)

(-420

(-315

-369.00
.80 --317.20)
-275.10
.90 --234.30)
-371.90
(-426.40 --317.30)

(-309
0.08(0.01),

0.

0.

0.

0.02
.08(0.01),
0.03
.13(0.01),
0.04
.14(0.01),
0.05
(-7

(-7

-266.10
.70 --222. 60)
-6.78
.97 --5. 60)
-6.02
.24- -4.81)
-315.70
(-360.00 --271.40)

(-330
-284.30
.80 --237.80)
(-0.

(-0.

-0.06
25-0.
-0.13
81-0.
-0.80
(-1.43 --0


(-3.

(-0.

(-1.

(-0.

(-0.

(-9.

(-19

(-7.

(-15

(-6.

(-15

(-0.

(-0.

(-7.


-1.38
23-0.
-0.07
40-0.
-0.11
08-0.
0.00
01-0.
0.00
01-0.
-4.04
01-0.
-9.16
.92-1
-3.79
96-0.
-6.31
.21-2
-2.34
92-2.
-5.41
14)

55)

.17)


46)

27)

86)

00)

01)

93)

.60)

38)

.59)

24)

Adjusted
R2
0.53

0.37

0.46


0.20

0.54

0.37

0.50

0.45
0.41

0.35

0.50

0.38

0.46

0.34
all 4
criteria
(relaxed)
fulfilled?
yes

yes

yes


no

yes

yes

yes

yes
yes

yes

yes

yes

yes

yes
all 4
criteria
(strict)
fulfilled?
yes

no

yes


no

yes

no

yes

no
yes

no

yes

no

yes

no
.22-4.40)
-0.03
10-0.
-0.03
20-0.
-1.12
26-5.
4.09

03)

13)

02)

0.35

0.22

0.37

0.29
yes

yes

yes

yes
yes

no

yes

no
(-8.30-16.48)
                                              51

-------
             OoverE-
    Simpson_Diversity-
               CSCM                   	A—•-
    Shannon_Diversity -
           RAWIowNi       		•—A-
           RAWIowP-
                 S2-             		
        propTaxaZHR-                	•—
        RAWIowTPsp-	
                H20i                   	•-A-
  RAWpropGreenCRUS i               	•-A-
                H23i                   	•—A-
  Taxonomic_Richness-                  	•—A-
,       RAWmeanZHR-                	»	
;                Dl8i                  	•—A-
|   RAWpropBiovolZHR		•	A
!                H211                  	•—A-
;           RAWNhet-
l  RAWpropBiovolChlor-
     Tolerant_Percent-
    EPT_PercentTaxa-
          EPT_Taxa-
  Tolerant_PercentTaxa -            	•	     	A •
          RAWsapro-
          RAWD050-          	•-
           RAWeutro-
          propAchMin-
         RAWDO100-
       lntolerant_Taxai
 lntolerant_PercentTaxa -        -•-  	A	
        EPT_Percent-       -•	A---
    lntolerant_Percent-       ••• — A —
       Tolerant_Taxa-_   —€^-
                    ^0   25   50   75  100   125  150  175  200  225  250  275  3C5cT
                                          chlorophyll a (mg/m2)
                                                                                       analysisType
                                                                                       •• Unweighted
                                                                                       A weighted

                                                                                       assemblage
                                                                                       • BMI
                                                                                       [•^ diatom
                                                                                       • hybrid
                                                                                       • soft
Figure 3.13. Breakpoints, with 95% confidence intervals, for the chlorophyll a gradient, from
piecewise regressions using all available ALI data types. Triangles correspond to analyses using sample
weights and circles correspond to unweighted. BMI ALI measures are in pink, diatom = green, hybrid = blue,
and soft = purple. Solid lines are the 95% Cl for unweighted analyses, and dashed are for weighted. Note
that fewer than half of the ALI measures' piecewise regressions met the "relaxed" criteria for confidence in
the breakpoint, as described in Methods, and none met the "strict" criteria. Details on analysis results are
provided in Table 3.7 and Table C.2.
                                                   52

-------
Breakpoints for AFDM (Figure 3.14) exhibited a higher degree of consensus among Alls than was observed
for chlorophyll a. In addition, a higher number of Alls achieved the "relaxed" criteria (but still none achieved
the "strict"). Among those Alls achieving the "relaxed" criteria, all  four assemblages were represented, and
estimated AFDM breakpoint values occupied the relatively narrow range of 7 to 39 g/m2; Table C.3, Table
3.7). Again, exhaustion thresholds tended to be lower for SENS and  EUTR indicators and greatest for INTI and
NUTR indicators although ranges of mean breakpoints tended to be narrower than for  chlorophyll a. Highest
weighted mean breakpoints were associated with proportional biovolume  in green algae, the soft algal IBI
and  proportion nuisance green algae. For the AFDM gradient, BMI breakpoints  tended to  occur at lower
values than diatom breakpoints. Again, indicators of initial DO depletion and increasing saprobicity had
exhaustion thresholds lower than those for filamentous greens and in the same range  as those for SENS and
INTI  BMI indicators.
    RAWpropBiovolZHR-
        RAWmeanZHR-
         RAWIowTPsp-
                  32-
         propTaxaZHR-
    RAWpropBiovolChlor-
       Tolerant_Percent-
                 H20-
   RAWpropGreenCRUS -
            RAWIowN-
                 H23-
                 H21-
            RAWIowP-
0)               D18-
3 Tolerant_PercentTaxa-
g       "  RAWeutro-
^  Taxonomic_Richness-
f    Simpson_Diversity-
--i    EPT_PercentTaxa-
      Shannon_Diversity-
            RAWsapro -
            RAWDO50-
            EPT_Taxa-
         EPT_Percent-
  lntolerant_PercentTaxa -
        lntolerant_Taxa-
                CSCI-
              OoverE-
         Tolerant_Taxa -
            RAWNhet-
      lntolerant_Percent-
           propAchMin-
           RAWDO100-
                                 •A
                                 •A-
                                                                                  •0 Unweighted

                                                                                  •A weighted

                                                                                  assemblage

                                                                                  • BMI
                                                                                     diatom
                                                                                     hybrid
                                                                                     soft
                                25      50       75       100
                                             AFDM (g/m2)
                                                                125
150
Figure 3.14. Breakpoints, with 95% confidence intervals, for the AFDM gradient, from piecewise
regressions using all available ALI data types. Triangles correspond to analyses using sample weights
and circles correspond to unweighted. BMI ALI measures are in pink, diatom = green, diatom+soft hybrid =
blue, and soft algae = purple. Solid lines are the 95% Cl for unweighted analyses, and dashed are for
weighted.  Note that fewer than half of the ALI measures' piecewise regressions met the "relaxed" criteria for
confidence in the breakpoint, as described in Methods, and none met the "strict" criteria. Details on analysis
results are provided in Table 3.7 and Table C.2..
                                                 53

-------
Results of unweighted piecewise regression analyses against the TN gradient achieved the "strict" criteria for
eleven Alls (Table 3.7). The Alls represented  all  four assemblages, and their estimated TN  breakpoints
ranged from approximately 0.5 to 1.1 mg/L, lending support, via multiple lines of evidence, that a variety of
instream ecological changes occur  below  1.1  mg/L TN.  Breakpoints were generally  lower for the BMI
assemblage relative to the algal assemblages (Figure 3.15), and for this assemblage, break points  were very
similar for the weighted and unweighted versions of the analysis.

H20-
Ulo"
52 •
0) H23'
jjj Taxonomic Richness*
03
j2 H21 -
	 i EPT_Peic8iilTaxa"
Tolerant_PercentTaxa •
EPT_Taxa-
lntolerant_PercentTaxa •
lntolerant_Taxa •
• A

.^

•^









_MI:I| ; _.i _• I ,'| _
A weighted
assemblage
BMI
diatom
0 hybrid
[•jsoft


                      0.4   0.5   0.6  0.7  0.8   0.9    1    1.1   1.2   1.3   1.4   1.5
                                              TN (mg/L)
Figure 3.15. Breakpoints, with 95% confidence intervals, for the TN gradient, from piecewise
regressions. This graph include only the All measures for which at least one analysis type (usually the
unweighted version) fulfilled all four of the "strict" criteria for confidence in the breakpoint, as described in
Methods. Triangles correspond to analyses using sample weights and circles correspond to unweighted.
BMI All measures are in pink, diatom = green, hybrid = blue, and soft = purple. Solid lines are the 95% Cl
for unweighted analyses, and dashed are for weighted.


Results of unweighted piecewise regression analysis against the TP gradient achieved the "strict" criteria for
four Alls (Table 3.7). The Alls represented only two of the assemblages (the diatoms and the hybrids, the
latter of which include information about the diatom community), and their estimated TP breakpoints ranged
from approximately 0.075 to 0.12 mg/L (Figure 3.16).
                                                 54

-------
 v>
 co
         D18-
H23-
analysisType
•^ UNweighted
JK  - - • -•

assemblage
9 diatom
0 hybrid
            0.06   0.07  0.08  0.09
                            0.1   0.11   0.12   0.13  0.14  0.15  0.16
                                 TP (mg/L)
Figure 3.16. Breakpoints, with 95% confidence intervals, for the TP gradient, from piecewise
regressions. This graph includes only the All measures for which at least one analysis type (usually the
unweighted version) fullfilled all four of the "strict" criteria for confidence in the breakpoint, as described in
Methods. Triangles correspond to analyses using sample weights and circles correspond to unweighted.
Diatom All measures are in  pink and hybrid = blue. Solid lines are the 95% Cl for Unweighted analyses, and
dashed are for weighted.
Whereas piecewise regression is focused on identifying the breakpoints (and associated uncertainty levels
around them) in All response along a gradient, SiZer plays the complementary role of establishing if, where,
and to what level of resolution, one or more slopes in  the relationship between response variable and
gradient are "real" and significant. Examples across a diverse array of All and gradient types are provided in
Figures 3.17 through 3.21.  In each of these cases, the mean All value decreased from the lowest to the
highest biomass (or nutrient gradient) value, and the portion of the gradient where a downward slope in All
value was most strongly supported by SiZer immediately preceded where the estimated breakpoint in slope
occurred,  as identified  by  piecewise regression.  Thus the two  methods, which are  based on  different
approaches, were always in close agreement. As such, different lines of evidence supported essentially the
same location for each threshold, thereby reducing the possibility that that location of any given estimated
threshold was a mere artifact of the analytical method employed.

In general, for All/gradient relationships in which the first slope  was particularly steep and the estimated
breakpoint based on  piecewise  regression had a narrow Cl, the SiZer map  exhibited a correspondingly
sharp/narrow downward red "peak",  indicating high confidence,  at fine as well as coarse  resolution (i.e.,
bandwidth)  leading up to that breakpoint. This is exemplified by relationships between Alls and nutrient
gradients, as shown in Figures 3.17 and 3.18. More  gradual initial slopes in the piecewise regressions, and
softer antecedent,  downward "peaks" in the red portion  of the SiZer maps were characteristic  of the
chlorophyll a and AFDM gradients (Figures 3.19 and 3.20).  Figure 3.21 provides an example of a more weakly
supported threshold for the All response variable, in this case, for the All H20  (a "hybrid" algae IBI) against
the AFDM gradient. Here, the Cl around the piecewise regression break point  is broad, and the downward
"peaks" in the red portion of the SiZer map are broad rather than sharp and localized over a narrow range in
the gradient.
                                                55

-------
                        0.0
0.2
                                                TP (mg/L)
Figure 3.17. Piecewise regression plot of diatom ALI variable, RAWIowP on a TP gradient (top) and
SiZer map from analysis of the same two variables (bottom). Note that x-axis was truncated to focus on
the area of the break point. On the piecewise regression, the green line under data points along the x-axis is
the 95% confidence interval14 for the breakpoint, and the green triangle is the breakpoint. On the SiZer map,
the downward-extended ("downward peak") red portion of the graph (yellow arrow at ~0.02 mg/L TP)
indicates an area of well-supported, significantly negative slope that holds even at narrow bandwidths (see
Methods for interpretation of color coding on a SiZer map, as well as definition of the y-axis).
  Note that the Cl on this graph is so narrow that it is barely discernible.

                                                 56

-------
                 in
                 a
             a
             s   ^
             t   o
                 **>
                 O
                 CN
                 a
                                              It   1
                 o
                 0
                s
                a
Figure 3.18. Piecewise regression plot of BMI ALI variable, lntolerant_PercentTaxa on a TN gradient
(top) and SiZer map from analysis of the same two variables (bottom). Note that x-axis was truncated to
focus on the area of the break point. On the piecewise regression, the green line under data points along the
x-axis is the 95% confidence interval for the breakpoint, and the green triangle is the breakpoint. On the
SiZer map, the downward-extended ("downward peak") red portions of the graph indicate areas of well-
supported, significantly negative slopes that hold even at narrow bandwidths (see Methods for interpretation
of color coding on a SiZer map, as well as definition of the y-axis). The slope that holds for the broadest
range of bandwidths is marked with a yellow arrow at ~0.14 mg/L TN, however note that there is a second
slope further down the gradient, near 0.8.
                                                57

-------
                s -
                a
                ft
                      0
100
200
300
400
500
Figure 3.19. Piecewise regression plot of BMI ALI variable, Taxonomic_Richness on a chlorophyll a
gradient (top) and SiZer map from analysis of the same two variables (bottom). Note that x-axis was
truncated to focus on the area of the break point. On the piecewise regression, the green line under data
points along the x-axis is the 95% confidence interval for the breakpoint, and the green triangle is the
breakpoint. On the SiZer map, the downward-extended ("downward peak") red portions of the graph indicate
areas of well-supported, significantly negative slopes that hold even at narrow bandwidths (see Methods for
interpretation of color coding on a SiZer map, as well as definition of the y-axis). The slope that holds for the
broadest range of bandwidths is marked with a yellow arrow at ~15 mg/m2 chlorophyll a, however note that
there is a second slope further down the gradient, near 55.
                                                58

-------
                 iq
                 B
*+ *
•r •* *
**•:** *
.
* ,
'..
•
*•
•• •
* * *
* * . *. * *
.."X.:.?. .•;. :-. ', • I-
tmttm*»t*m ** •*»•**«* M ***•**• ». _ _ . » *« « «
0 10 20 30 40 50
•
* 4*
1
60
               |
Figure 3.20. Piecewise regression plot of BMI ALI variable, lntolerant_PercentTaxa on an AFDM
gradient (top) and SiZer map from analysis of the same two variables (bottom). Note that x-axis was
truncated to focus on the area of the break point. On the piecewise regression, the green line under data
points along the x-axis is the 95% confidence interval for the breakpoint, and the green triangle is the
breakpoint. On the SiZer map, the downward-extended ("downward peak") red portions of the graph indicate
areas of well-supported, significantly negative slopes that hold even at narrow bandwidths (see Methods for
interpretation of color coding on a SiZer map, as well as definition of the y-axis). The slope that holds for the
broadest range of bandwidths is marked with a yellow arrow at ~6 g/m2 AFDM, however note that there is a
second slope further down the gradient, near 13.
                                                59

-------
                 8 -
                 o -
                         «*.     ****
                         v*    * .   *.
                            •••   •
                                *   •'
 •*
.. •
                                                              50      60       70
Figure 3.21. Closeup of piecewise regression plot of hybrid ALI variable, the IBI H20 on an AFDM
gradient (top) and SiZer map from analysis of the same two variables (bottom). Note that x-axis was
truncated to focus on the area of the break point. On the piecewise regression, the green line under data
points along the x-axis is the 95% confidence interval for the breakpoint, and the green triangle is the
breakpoint. On the SiZer map, the downward-extended ("downward peak") red portions of the graph  indicate
areas of significantly negative slopes that hold even at narrow bandwidths (see Methods for interpretation of
color coding on a SiZer map, as well as definition of the y-axis). The slope that holds for the broadest range
of bandwidths is marked with a yellow arrow at ~4 g/m2 AFDM, however note that there is a second slope
further down the gradient, near 32.


The thresholds identified in Figures 3.17-3.21 all exhibited hallmarks of "exhaustion" thresholds (see Section
3.2.3). To illustrate this, we generated boxplots showing distributions of the ALI values for sites  binned by
gradient values (specifically, those falling below the identified threshold vs. those falling above;  Figure 3.22).
All five Alls in this example were expected to  decrease in value with increasing stress, and in each case the
mean ALI value below the threshold exceeded that above the threshold. The reason the thresholds were
interpreted as being of the "exhaustion" variety is that the distribution of values  above the threshold
exhibited a substantially narrower interquartile range (IQR) than those below the threshold. The response of
H20  ALI  along the  AFDM  gradient was an exception,  in which  IQRs were similar above  and  below the
                                                60

-------
threshold. However, even in this latter case, the range in AFDM values below the threshold was only 1/10 the
range in values above the threshold within the project dataset (which extends to 450 g/m2 — note that the x-
axis in Figure 3.21 is truncated to allow easier viewing of the break point). As such, the ratio of IQR to range in
gradient was substantially higher below the threshold than above it. In summary, the higher variability in All
values below the threshold strongly supports the threshold as being  "exhaustion" rather than "resistance".
This same pattern is evident across by far the majority of All-gradient combinations we examined (Table C.3).
This  result and  others  (see below) provides  support that,  except  in  the rare  cases  where  noted, the
thresholds we identified were exhaustion thresholds.
  1.00-

  0.75-

  0.50-

  0.25-

§0.00-
>
<

   40-


   20-


    0-
               RAWIowP, TP
0.6-
0.4-
0.2-
n n-
Intolerant PercentTaxa,








i
4
1
4
AFDM
1


0.6-
0.4-
0.2-
n n-
Intolerant PercentTaxa, TN







I I
below
above
                                           below        above
       Taxonomic_Richness, chlorophyll o              H20, AFDM
                                  100-
                                   75-
                                   50-
                                   25-
                                    0-
            below
                        above
below
                                                              above
                   below        above
           gradient value relationship to threshold
Figure 3.22. Distribution of ALI values among sites with stressor gradient (i.e., biomass or nutrient
concentration) values below vs. above the threshold that had been determined based on piecewise
regression. The strip above each panel in the plot indicates the type of ALI followed by the type of gradient
in question. The ALI values are indicated  by the y-axis.
                                                 61

-------
3,3.3 Examining         Influence of Biomass, Nutrients,     Other Factors on           ALI Measures
BRT and partial Mantel tests

BRT, a modeling approach, is the one type of analysis used in this study that allowed us to incorporate effects
of other potential confounding factors (other stressor types as well as natural gradients) on the relationship
between biomass/nutrients and  ALI  responses.  Partial  Mantel tests  were used to  determine whether
important predictors of ALI response, based on BRT models, were statistically significant when other factors
were controlled for.

For most ALIs, nutrients outranked biomass variables in terms of their relative influence in BRT models
(Tables 3.8 and 3.9)15. The exceptions were for the soft-algae ALIs (in which for two of the  four ALI types
tested, biomass in the form of soft algal total biovolume was the biomass/nutrient predictor with the highest
relative influence) and for the  BMI metric EPT_Percent  (for which AFDM ranked higher than any of the
nutrients). In general, among biomass types, AFDM was the highest-ranked predictor for the greatest number
of ALIs, followed  by soft algal total biovolume and chlorophyll a. PCT_MAP and PCT_MCP  were not top-
ranked predictors for any of the ALIs examined.

The overall top-ranked predictor for most of the BMI ALIs was TN (Tables 3.8 and 3.9), and for diatoms, it was
phosphorus  (either as TP or SRP). TN was also a top predictor for one of the soft-algae  ALIs (the index, S2),
and NOX was the top-ranked predictor for another soft-algae ALI (RAWmeanZHR). Summaries of the relative
influence of all predictor variables, and specifically among the biomass and nutrient variables, are depicted
graphically in Figures 3.23 and 3.24, respectively.

Results of the  partial  Mantel tests on the top-ranked  predictors from the final BRT models are provided in
Table 3.10. Most of the top-ranked predictors were found to be significantly correlated with their respective
ALI response  variables  when  the  effects of  the  other  top-ranked  predictors,   as  well  as  spatial
autocorrelation—in terms of geographic distance between sites, were controlled for. The latter generally did
not have a significant effect on the ALI response variables (or the effect was relatively  small, if significant),
suggesting that predictor-response relationships observed in the BRT analyses  were not merely artifacts of
spatial autocorrelation.
15 Note that we re-ran two test cases for BRT with the input data transformed to improve normality and found that the
results were nearly identical to those we provide in Table 3.8 (which was based on untransformed data), thus confirming
that data transformation is not necessary for BRT analysis.

                                                 62

-------
Table 3.8. Summary of boosted regression tree models of All variables, and relative influence (and rank)
of biomass and nutrient predictors used in each. Boldface type corresponds to biomass or nutrient
predictors that ranked highest within each model. Each model contained only one type of biomass
predictor (as indicated by the column, "biomass type included in model"). Biomass type selected for each
model was based on what biomass type ranked highest in an analogous model containing all five predictors
(data not shown). "Model cv correlation (se)" refers to the cross-validation correlation coefficient (with
standard error), indicating reliability of each model (Elith et al. 2008). Dashes indicate that the predictor in
question was not part of the final BRT model for that All  variable.


Assemblage



BMI
(N = 611)



diatom
(N = 888)

Hybrid
algae
(N = 809)


soft algae
(N = 845)









All type
lntolerant_
PercentTaxa
Taxonomic_
Richness

CSCI16
Shannon_
Diversity
EPT_Percent
D18
RAWeutro
RAWD0100
RAWNhet
H20

S2

RAWprop
GreenCRUS

RAWprop
BiovolChlor

RAWmean

ZHR
Biomass
type(s)
included in
model
AFDM
chlorophyll

chlorophyll
soft algal
total
biovolume
AFDM
AFDM
AFDM
AFDM
chlorophyll
AFDM
soft algal
total
biovolume

soft algal
total
biovolume

soft algal
total
biovolume

soft algal
total
biovolume
Highest
ranked
predictor
(relative
influence)
TN (27.12)
oTN (30.96)
URBAN_
02000 5K
(16.2)
TN (15.23)
ecoregion
(10.75)
TP (20.53)
ecoregion
(9.64)
SRP (11.31)
conductivity
° (11.89)
URBAN_
2000_WS
(24.12)
TN
(25.99)

soft algal
total
biovolume
(35.01)
soft algal
total
biovolume
(31.07)
NOX

(18.37)
Model cv
#
„__ j.^ 	 Relative influence of (rank)





correlation # in final
(se)
0.932
(0.005)
0.847
(0.009)

0.829
(0.012)
0.727
(0.016)
0.717
(0.021)
0.773
(0.015)
0.664
(0.021)
0.648
(0.025)
0.641
(0.035)
0.847
(0.009)
0.781
(0.022)

0.727
(0.015)

0.658
(0.021)

0.624

(0.015)
Trees model biomass TN
-° » 2<9? ™
5050 - IS T

5850 31 U2) %?
4900 35 £} ^
5750 2° ™ (3?
5400 31 m1 5(53)8
6850 29 ?10) S
7050 - ?S ™
= 36 S %
2.57 18.47
(9) (2)
6.90 25.99
5950 15 (5) (1)

5500 " T (53)6

4500 18 31'07 5'13
(1) (6)

8.96 10.90
4650 22
(3) (2)
NOX
-
0.27
(35)

0.43
(31)
0.61
(35)
3.47
(15)
0.80
(30)
1.33
(27)
2.61
(16)
1.03
(32)
1.52
(10)
11.82
(3)

9.28
(3)

7.33
(3)

18.37

(1)
NH4
-
0.47
(31)

-
1.21
(27)
-
0.94
(26)
1.38
(26)
-
1.86
(21)
0.77
(23)

"

-




1.42

(20)
TP
-
3.07
(6)

2.66
(11)
2.12
(13)
-
20.53
(1)
9.26
(3)
8.42
(2)
7.10
(3)
12.37
(3)
6.62
(6)

-

3.29
(10)

8.68

(4)
SRP
-
2.24
(10)

1.06
(22)
1.72
(19)
-
4.89
(6)
9.26
(2)
11.31
(1)
5.52
(5)
2.92
(7)

"

-

2.22
(16)

2.21

(14)

  Note that the CSCI scoring tool was in draft form at the time of preparation of this report and is subject to change
before being finalized.

                                                 63

-------
Table 3.9. Relative influence of predictors from BRT models. The top-ranked predictor in each model is in
bold. Dashes indicate that the predictor in question was not a part of the final model for the All measure  in
question.
All Measure
Predictor
Type

Biomass



Nutrient







Other



Predictor
soft algal
biovolume
AFDM
Chlorophyll o
TN
TP
NOX
SRP
NH4
conductivity
ecoregion
URBAN_2000_
5K
slope, reach
elevation
CODE_21_
2000_5K
canopy cover
URBAN_2000_
WS
discharge
stream
temperature
URBAN_2000_
IK
alkalinity
sands & fines
u
0
-
-
2.4
5.4
2.7
0.4
1.1
-
7.3
2.9
16.2
12.1
1.4
-
2.7
-
3.6
2.0
6.6
2.3

Ol
OJ
Q.
t'
LLJ
-
10.7
-
8.6
-
3.5
-
-
4.5
10.8
5.4
8.4
-
-
-
-
6.7
3.5
4.3
2.7

lntolerant_
PercentTaxa
-
2.6
-
27.1
-
-
-
-
3.6
22.4
9.5
5.4
4.0
1.1
2.2
-
1.2
12.0
1.5
2.9

Shannon_
Diversity
2.3
-
-
15.2
2.1
0.6
1.7
1.2
9.8
7.3
11.2
2.4
0.7
0.9
2.0
-
1.9
2.5
1.6
1.9

Taxonomic_
Richness
-
-
1.6
31.0
3.1
0.3
2.2
0.5
5.9
9.0
13.6
1.2
0.4
0.6
1.9
-
1.0
0.8
2.7
0.8

00
Q
-
3.7
-
5.4
20.5
0.8
4.9
0.9
14.4
2.5
-
2.0
1.9
2.5
1.5
9.1
0.7
1.1
2.0
1.2

RAWDOIOO
-
3.6
-
1.7
8.4
2.6
11.3
-
4.1
5.8
-
2.5
3.3
2.1
7.0
-
2.5
3.9
2.6
7.3

RAWeutro
-
2.9
-
1.3
9.3
1.3
9.3
1.4
6.2
9.6
-
2.3
4.8
2.1
7.5
-
1.8
4.8
2.1
6.4

RAWNhet
-
-
3.0
5.3
7.1
1.0
5.5
1.9
11.9
0.7
-
2.2
1.5
1.1
5.9
5.2
1.1
2.2
7.6
1.9

0
-
2.6
-
18.5
12.4
1.5
2.9
0.8
7.0
1.3
-
1.1
3.8
3.2
0.8
24.1
0.5
0.7
1.5
1.2

RAWmeanZHR
9.0
-
-
10.9
8.7
18.4
2.2
1.4
2.9
6.1
-
3.5
3.0
7.1
2.3
-
4.0
2.1
-
2.2

1 1 is
Ills; N
QC 00 QC 13 l/l
31.1 35.0 6.9
-
-
5.1 7.4 26.0
3.3 - 6.6
7.3 9.3 11.8
2.2
-
3.8 7.2 2.3
2.6 4.0 4.0
-
5.0 - 3.9
9.3 10.4 7.3
6.0 9.0 15.0
2.5 4.9 3.3
-
5.4 5.3 2.4
-
-
-

         PH
                      0.6   3.0   1.4   2.7    2.3    6.6    3.4    2.6    3.4   2.7

                      1.0   3.4    -    1.2    0.8    1.1    1.8    1.7    2.5   0.7   4.4   3.2    3.2    2.5
                                                   64

-------
Table 3.9 Continued.
All Measure

Predictor
Type Predictor

01
Ol
D t<
(/) Q.
\J LU

lntolerant_

Percent Taxa

i
0
c
ra
.c
1/1

Diversity

Taxonomic_

Richness
D18

RAWDOIOO

RAWeutro

RAWNhet
H20
cc
RAWmean-Zh

RAWprop
BiovolChlor

RAWprop-
GreenCRUS

fM
l/l
          longitude          1.3     -       -      3.5     5.0    1.3     1.6     1.5    1.2    0.4    3.7    3.1     -     2.8
          mean monthly
          max temp (3-mo    1.0     -       -      2.4     1.2    0.9     2.7     2.4    1.0    0.6    1.3    2.3    4.3    2.9
          span)
          watershed area     2.2    3.5      -      0.9     0.6    1.0     2.2     2.2    1.1    1.1    2.1    3.6     -     2.1
          CODE_21_2000_
          WS                 -.........----
          stream depth       1.1     -       -      5.6     2.0    1.8      -      2.0    3.5    0.7
          coarse particulate
          organic matter     0.7    3.9      -      0.7     0.5    0.9     4.8     2.6    1.9    0.9
site disturbance
. -LJ..O ~
class
turbidity 1.2
sedimentary
geology (%)
Other latitude 1.4
total precipitation
(3-mo span)
fines (%) 3.6 3.7 1.2
stream width
mean monthly
solar radiation (3- 0.9
mo span)
W1JHALL
(riparian
disturbance
index)
mean monthly %
cloud cover (3- - 2.6
mo span)
Ag_2000_WS 1.0
Ag_2000_lK
0.2 1.4
1.7 0.8 2.2 2.0
0.9 0.6 2.1 1.8
4.3 2.9 - 2.1
1.2 0.6 1.0
0.6 0.8 0.9
0.7 0.4 - 4.3
1.8 0.8 - 1.7
1.6 0.8 1.8
1.8 0.4
1.7 1.3
.
2.2 1.3
2.2 1.8 1.4 1.8 -
1.3 2.4 1.4 - - - -
1.5 0.5 0.5 1.6 -
4.0 0.5 - 2.2 -
0.4 - - 2.1 -
3.0 1.3 0.4
1.6 0.9 0.6 1.4 -
2.1 1.0
1.1 0.5
-
2.3 0.6
                                                            65

-------
Ag 2000 1K
Ag_2000 WS-
meart monthly % cloud cover (3-mo span)
W1 HALL (nparian disturbance index)
mean monlhly solar radiation (3-mo span)
stream width
fines (%}
total precipitation (3-mo span) -
latitude
sedimentary' geology £%) -
turciditj
site disturbance dass
coarse paniculate organic matter (%}
stream depth
CODE 21 2000 WS-
watershed area
mean monthly max temp (3-mo span )
._ longitude
2 pH
-S sands* fines (%)

-------

35-

30-


25-
&
o>
£20-
3
—

-------
Table 3.10. Partial Mantel coefficients (95% CIs) for correlation between biomass/nutrient predictors and All variables and p-values. Grey boxes
correspond to explanatory variables that were not included in the partial Mantel test for the All variable in question. "Space" refers to the geographic
distance between sites (for testing the significance of spatial autocorrelation). "NS" = not significant; dashes correspond to predictors that were
included as explanatory variables in the partial Mantel tests for the indicated All variables, but (because they did not fall under the categories of
biomass, nutrients, or "space") were not the focal variable in the tests. Values in bold correspond to significant partial Mantel tests.



L.
2
_u
'-a
K
a.




u
S3



4-»
1 <"
1-' ^
Q. 01
III Q
ra
X
1 m
= "7
E c
01 01
"o y
= &



= ,>
= S2
c Ol
5 £
in O

l
u
E »
0 X
11
y




00
iH
Q

O
O
!H
6
Q
1
rv*



E
4-»
3
01
1
rv*



tt
.c
1
rv*

M

ra
01
|«
o: i

i_
, o
is
f 8
5 .2
rr m
i
O
>
Q. .2
0 CO
f " 3
| 2 2
^ y? u




fM
1/1




O
fM
X
0-06
chlorophyll (004^Q8)
° 0.005
AFDM
soft algal
volume
TN
NOX
TP
SRP
space
(0.16-0.20) (0.13-0.17)
0.001 0.001
3iO-
0.17 0.24 0.34
(0.14-0.18) (0.22-0.26) (0.32-0.36)
0.001 0.001 0.001

0.07
(0.04-0.09)
0.003

-°'°4 0.03 0.07
(" • " (0.01-0.04) (0.06-0.09)
' 0.012 0.001
NS
0.04
(0.02-0.06)
0.022

0.00
(-0.01-0.01)
NS
0.15 0.26
(0.12-0.17) (0.24-0.28)
0.001 0.001

-0.05 0.04
(-0.07- -0.03) (0.03-0.06)
NS 0.004
-0.05
(-0.06- -0.03)
NS
-0.02 -0.02
(-0.03- -0.01) (-0.03- -0.01)
NS NS

0.12
(0.11-0.14)
0.001

0.07
(0.05-0.09)
0.001

0.16
(0.14-0.18)
0.001
0.01
(-0.01-0.03)
NS
0.02
(0.01-0.03)
0.026

0.01
(0.01-0.03)
NS



0.08
(0.06-0.09)
0.001
0.04
(0.03-0.05)
0.023
-0.002
(-0.02-.01)
NS

0.01
(-0.001-
0.022);
NS



0.09
(0.07-0.10)
0.001
0.037 (0.021
- 0.051);
0.013
0.004
(-0.01-0.01)
NS
0.07
(0.04-0.09)
0.002


0.06
(0.03-0.08)
0.006

0.10
(0.08-0.12)
0.001
-0.03
(-0.05- -0.01)
NS
-0.03
(-0.04- -0.02)
NS


0.01 0.210 (0.194
(0.00-0.01); - 0.224);
NS 0.001
0.06
(0.04-0.07)
<0.001
0.12 0.14
(0.11- 0.14) (0.12-0.15)
<0.001 0.001
0.06
(0.05-0.07)
<0.001

0.04 0.01
(0.03-0.05); (0.0 -0.02)
<0.001 NS


0.18 0.05
(0.17-0.20) (0.04-0.061)
0.001 0.001
0.07 0.10
(0.06-0.08) (0.09-0.12)
0.001 0.001
0.11 0.19
(0.09-0.12) (0.18-0.21)
0.001 0.001
0.08
1 (0.07-0.09)
0.001

, -°'03 -0.01
7-- (-0,2-0,1)
NS

0.11
(0.09-0.13)
0.001

0.17
(0.15-0.19)
0.001

0.16
(0.14-0.18)
0.001
0.03
(0.01-0.05)
0.049
-0.03
(-0.04- -
0.02)
NS
conductivity
                                                                      68

-------
Table 3.10. (continued)



L.
5
_u
'-a
s
ft




U
S3



4-»
1 «
1-' ^
Q. 01
III Q
ra
X
1 m
= "7
E c
01 01
"o y
= S



= ,>
i S
y

i
u
E »
o S
11
y




00
iH
Q

O
O
!H
6
Q
1
rv*



E
4-»
3
01
1
rv*



tt
^
1
rv*

M

ra
01
i«
o: i

i_
, o
is
f 8
5 .2
r^ m
i
O
>
Q. .2
0 CO
f £ ^
1 S! i
r^ y? u




fM
1/1




O
fM
X
canopy cover
stream
temperature
alkalinity
stream width
CPOM (%)
elevation
URBAN_
2000 IK
sand & fines
URBAN_
2000 WS
stream depth
total precipi-
tation  (3-mo
span)	
URBAN_2000_
5K

slope, reach

discharge

fines (%)

CODE_21_2000
_5K
latitude

longitude

mean  monthly
max temp (3-
mospan)
                                                                                    69

-------
Chlorophyll a had low relative influence in all three BRT models in which it was included as a predictor (Tables
3.8 and 3.9), and changes in slope in the corresponding partial dependence plots (Figure 3.25) were gradual
along the chlorophyll a gradient, making it difficult to discern clear thresholds. However, at least for the BMI
Alls, a  possible threshold was weakly evident in the vicinity of 60 mg/m2, and for all three Alls, the partial
dependence plots leveled off by 100  mg/m2, suggesting that any thresholds of effect of chlorophyll a  on
those Alls is <100 mg/m2.
                             .§
                             "O   CM -
                             1
                                 o -
                                           Taxonomic  Richness
                                                i     i     i      r
                             s
                                                           CSCI
                             1
                             o
                             n
                             
-------
AFDM was included as a predictor in the BRT models for six Alls spanning the BMI and diatom assemblages
as well as the diatom/soft algae ("hybrid") IBI, H20. Examples of partial dependence plots from these models
are provided in Figure 3.26. AFDM was the predictor with the second highest relative influence on the BMI
All, EPT_Percent, among a total of 20 predictors in the final model  (Tables 3.8 and 3.9). There was a
precipitous drop in the fitted value for EPT_Percent along the AFDM gradient until around 25 g/m2, beyond
which no further decline was evident. A roughly similar pattern, albeit less pronounced, was realized for the
BRT model with the diatom All, RAWDO100, as the response variable (however, AFDM was not a significant
explanatory variable for this All in the partial Mantel test; Table 3.10). The IBIs H20 and D18 exhibited similar
initial  breakpoints of approximately  35 g/m2 in their partial dependence plots; beyond that point, the fitted
response variables continued to decline (although much more shallowly; Figure 3.26). Only at approximately
180 g/m2 AFDM was no further decline evident for any of the Alls.
          £=
          O
              o
              O
              in
              o
          TJ
          I   8
  EPT Percent
                                       H20
          I
              O
              °
              o
              o
              o
                                   RAWDO100
                                              D18
                         50
100
150
200
50
100
150
200
                                            AFDM (g/m2)
Figure 3.26. Partial dependence plots of AFDM from BRT models predicting four ALI response types:
the metrics EPT_Percent and RAWDO100; and the IBIs H20 and D18. Y-axes correspond to the
standardized, fitted ALI variables. Graphs do not show entire gradient length, but are cut off at the point
beyond which there are no further changes in slope.

Whereas biomass was  rarely  a top-ranked predictor among the BRT models for the  fourteen Alls,  the
opposite was true for nutrients (Tables 3.8 and 3.9). Nitrogen in one form or another was the top-ranked
predictor for three BMI Alls and two soft-algae Alls, and phosphorus in one form or another was the top-
ranked predictor for two diatom Alls. Partial  dependence plots of TN from BRT models  for three Alls are
provided in Figure 3.28. Three breakpoints were observed in the plots, depending upon the ALI in question.
For two of the ALIs (S2 and Taxonomic Richness), there was an initial breakpoint at 0.3 mg/L TN, where the
curve transitioned from more-or-less flat to a  strong negative slope, and a final breakpoint at 0.8 mg/L TN,
                                               71

-------
after the curve became essentially flat again. The former may be considered a "resistance" threshold, and the
latter an "exhaustion" threshold (see Introduction to this Section for definitions). In the case of Taxonomic
Richness, there was also a gentler break in slope of the partial dependence plot around 0.55 mg/LTN, which
was also the location of the break in slope for the partial dependence plot for Intolerant Percent Taxa (Figure
3.27). Thus, for the three Alls, which collectively represent two different assemblages, similar patterns of
response to TN were observed, providing weight of evidence for threshold locations along this gradient.

Partial dependence plots for the TP gradient (Figure 3.28) were reasonably congruent  across All response
variables. There was a precipitous drop in  the  fitted values for  RAWmeanZHR,  D18, and  H20 along the TP
gradient until around 0.05 - 0.1 mg/L TP, beyond which no further decline was evident for RAWmeanZHR,
and minor fluctuations in slope were observed for D18 and H20.
                                                             S2
                                            x       Taxonomic
                                             \      Richness
                              c
                              o
                                 o
                                 o
                                 O  —
                                 o
Intolerant
Percent Taxa
                                         02    04
  06
08
1 o
                                             TN (mg/L)
Figure 3.27. Partial dependence plots of TN from BRT models predicting three ALI response types:
the soft algae IBI, S2; and the BMI ALIs, Taxonomic Richness and Intolerant Percent Taxa. Y-axes
correspond to the standardized, fitted ALI variables.
                                               72

-------
                               10
                               o
                           «   m
                           S   °-
                           «s   o
                           T3
                           1   s
RAWmeanZHR
                                                          I      I
                               u.
                           o   o  -
                               in  H
                               o  -
                                                           D18
                                                           H20
                                   00   02   04    06   08
                                            TP (mg/L)
                10
Figure 3.28. Partial dependence plots of TP from BRT models predicting three ALI response types:
the soft-algae ALI, RAWmeanZHR; and the IBIs, D18 and H20. Y-axes correspond to the standardized,
fitted ALI variables.
3.3.4 Thresholds for Biomass and Nutrient Effects on Biotic Response
We employed a wide variety of analytical methods, ALI response variables from different biotic communities,
and primary producer abundance measures to evaluate potential thresholds of effect of biomass on stream
Alls. The sheer volume of output from this effort practically guaranteed that the results would not all point
to a single biomass or nutrient threshold. However, for many of the gradients  examined,  there was a
reasonable degree of consensus among analytical techniques and ALI response types within each of the four
"ALI categories" (Figures 3.29- 3.30), even between biotic assemblages, thus providing a weight of evidence
for fairly narrow ranges  of threshold values.  Most of the thresholds we observed could be classified as
"exhaustion" thresholds (as defined in Section  3.2.3). In other words, ALI responses, as  inferred through our
approach, were generally saturated at the point along  the  stressor gradient at which we observed most
                                               73

-------
thresholds. Thus most of our thresholds would best be considered "backstops", and this would be important
to keep in mind when considering these results in any policy decisions.

40-


10-

90-


TQ
O

ISLQ-
1.0-
0.5-



BMI




•


•
























1
VDnnnnnnD °D °nn














-
































* D n n ,_, a o n D D P a
D n



1



1
V a

t

1
f















i

1 1

1


,


.



1 "•'• DD •"
3 ° * *
a n D a D DDDD



1 1



i

1 1









,
•i

•
•

•
ADD D
0 n
TD D D D D D ODD
n n 	
diatom

+

w 1
a 1 i


i
] C


i :





c
:
D
: :
] [
























n n
p T n n °
*

f


1




i

i
1
1
1
i
• 1° ° n o o n o :
a

*


1


4

i



I

1
P
* • " • ^
0 t g " D D D D D :












1




hybrid

i

i
i




'

3 c

C
3

1 •
1










ana


1
.

I




.
DDD

<
1


i
• 1

I i




D a a
soft |


•
D D D
n




E



1


1

D DP
•






1
1
tu
n n a 3
i

1

*

.
n •
° • •
n D n n I







1
• i^

D n d D D D D
                                                                                           Analysis.Type

                                                                                            *  BRT_exhaustion

                                                                                            *  BRT_resi5tance

                                                                                            *  CART

                                                                                            A ncpa.bc




                                                                                            B  piecewiae regressic1

                                                                                            D  SiZer

                                                                                            T TITAN-EJeCTEB5ET5
        _

                                   Rlllii
                   ?tg = Sr-

                   '"£ £
                                         ALI response
                                                                     s  <
Figure 3.29. Summary of results across analyses using the chlorophyll a, AFDM, TN, and TP
gradients, stratified by assemblage type. The y-axis corresponds to the threshold that was identified by
the analysis in question for each of the gradients. For the CART results, data are based  on the models using
the full statewide dataset. For the piecewise regression analyses, break points (thresholds) are given only for
those that passed at least the "relaxed" evaluation criteria.  For the BRT thresholds, only those for which the
predictor was significantly correlated with the ALI response in the partial Mantel test are  included. Note that
the values corresponding to SiZer analyses can more appropriately be viewed as indicative of a significant,
dramatic slope change that preceeds a threshold, rather than as a threshold in itself. Analysis-specific
confidence limits, where applicable, are provided in the figures and tables presented previously. See Table
3.1 for definitions of the ALI variables. A tabular version of  this information is provided in Table C.3.
                                                  74

-------


A —
1 A •—
:
0 10 20 30 40 50 60 70 80 90 100 110 120
chlorophyll a (mg/m2)
i ~A
i :
1 -*
.
A A
1
i :
0 0.2 0.4 0.6 08 1 1.2 1.4 1.6 1.8 2 2.2
TN (mg/L)












A J
• A

^ ' *
I
0 5 10 15 20 25 30 35 40
AFDM (g/m2)
I 4,.
I
1 A -
1
1 A •
1
.».

0 0.05 01 0.15 0.2 0.25
TP (mg/L)

•• low nutrients
eutrophication
• integrative
• mean
A median








Figure 3.30. Ranges of thresholds of ALI response by "ALI category" (as described in section 3.2.2)
for two biomass and two nutrient gradients. The same data that are shown  in Figure 3.29, all
assemblages and analyses combined, were used to make these graphs. Circles correspond to the mean of
thresholds within each category, and triangles are the medians. Dashed lines indicate the 75th percentile of
the indicator in question among Reference sites statewide, and dotted lines indicate the 95th.
3.4 Discussion
This study found evidence for a range of thresholds of effect for benthic chlorophyll a, AFDM, and TN and TP
concentrations on BMI and algal community structure. Most of the thresholds observed could be classified as
"exhaustion" thresholds17- a  sharp transition in the stressor gradient at which point the response variable
reaches a natural  limit  (Cuffney et al. 2010). Thus we have generally  characterized  these thresholds as
indicative of "adverse" effects on the ALI responses used. Integrative Alls (such as IBIs) corresponded to
higher thresholds  whereas ALI  measures specific to constrained groups of "sensitive"  taxa  generally
corresponded to lower thresholds, illustrative of the paradigm of the biological condition gradient (Davies
and Jackson 2006; Figure 3.2). In this discussion, we  employ the median range within the ALI categories
(sensitive, low nutrient, eutrophication, integrative) in order to summarize and  compare with the literature.
These ranges do not imply value judgments with respect to rigor of analytical approach nor importance of AL
indicator type and thus should not be construed as policy recommendations.

Most of these thresholds of effect exceeded the 75th percentile of these indicators among Reference stream
reaches statewide, but they were often less than the 95th percentile (Figure 3.30). Statistically significant
relationships between stressors (benthic chlorophyll a concentrations, AFDM, nutrients) and a variety of ALIs
were observed. However, change points in the  response to AFDM and nutrient concentrations were  more
discernible than that for chlorophyll a (as currently measured in California ambient monitoring programs).
These conclusions are based on analytical criteria for assessing the level of confidence in thresholds and the
  Ecologically meaningful resistance thresholds may not always exist (or may be so low as to be undetectable with
  available methods/data), and few were apparent based on our analyses.

                                                 75

-------
degree of consistency of thresholds across All  indicators as multiple lines of evidence, both  within and
among assemblages.

,3.4.1 Statistically         Thresholds of Adverse Effect in California Wadeable Streams
Benthic Chlorophyll a

Benthic chlorophyll a had a  statistically-significant relationship with many California stream Alls.  Overall,
thresholds of adverse effects ranged from 4 to 113 mg/m2 chlorophyll  a, and median thresholds within All
categories ranged from  12 to 43  mg/m2. Most  of our  analysis- and All-specific chlorophyll a thresholds
exceeded the 75th percentile of chlorophyll a values  among Reference stream reaches statewide (14.6
mg/m2; Chapter 2). A review of literature revealed only one study that used statistical methods to detect
thresholds of adverse effects of benthic chlorophyll a on All indicators  in wadeable streams. Miltner (2010)
found a change point at 107 mg/m2 related to changes in the abundance of Ephemeroptera, Plecoptera, and
Trichoptera (EPT)  taxa in Ohio streams.  EPT are benthic macroinvertebrate taxa typically associated with
"clean water" streams.

The range of thresholds found in this study were substantially  lower than the NNE endpoints recommended
for streams by a working group of international  experts, regulatory agencies and stakeholders (Tetra Tech
2006, Appendix A). Two factors should be considered in comparing thresholds in literature that had been
cited in support of the NNE thresholds proposed by  Tetra Tech (2006; e.g., Biggs 2000; Quinn  and Mickey
1990) with those of our study: 1) the temporal breadth of sampling and the temporal statistic that is the basis
for the threshold, and 2) the range of All indicators considered  (benthic invertebrates vs. salmonid fisheries).

Based  on a  study of 31 reaches  in 21  New Zealand streams, Biggs  (2000) observed that chlorophyll a
concentrations exceeding ~13-20 mg/m2 were associated with a 50% reduction in the percentage of EPT taxa.
These chlorophyll  numbers fall within the range of thresholds  found in this study, which is based on a one-
time sample in a spring - summer index period. In contrast, Biggs' (2000) values are based on mean monthly
samples.  Biggs (2000) goes  on to note that mean  monthly sampling over the course of a  year in 16
oligotrophic  streams (defined as those with catchments having <  1% developed land  use) yielded a 90th
percentile of 20 mg/m2 and a  mean peak biomass of 47 mg/m2. It is on the basis of this work that Biggs (2000,
p. 97) stated, "I recommend that the mean monthly biomass not exceed 15 mg/m2 and the peak biomass not
exceed 50 mg/m2 for the protection of benthic  biodiversity in streams".  He goes on to add that the two
measures imply that Alls can  continue  to  thrive when benthic algal abundance is elevated for  a short
duration, but that more substantial adverse effects would occur with chronic algal  blooms. Unfortunately,
repeat sampling that would be helpful to relate  the one-time sample taken  during the  PSA spring-summer
index period  to mean monthly or maximum statistics  has not been  conducted for California. Thus it is
important that any application  of thresholds from this study to policy  development consider the temporal
statistic and the monitoring frequency with which regulatory decisions would be made.

Thresholds  arising from the  present study were  derived  based  on  changes  to  algal  and  benthic
macroinvertebrate community composition,  while NNE endpoints are also supported by literature linked to
salmonid beneficial uses. Biggs (2000)  asserted that  protection of salmonids affords a  slightly higher algal
biomass than is  protective  of benthic  invertebrate "clean water  species".  Quinn  and Mickey  (1990)
demonstrated that trout biomass increased from oligotrophic (< 20 mg/m2) to mesotrophic (20-100  mg/m2)
streams, but then fell three-fold in eutrophic streams (>  100 mg/m2). Biggs (2000) demonstrated that mean

                                               76

-------
monthly benthic algal biomass in New Zealand streams that are "renowned for their trout fisheries" was 23
mg/m2, with average maximum biomass of 171 mg/m2.

Further modeling studies by Quinn and McFarlane (1989) link abundance of macroalgae at 21 °C in excess of
120 mg/m2 chlorophyll a to depressed dissolved oxygen (DO) (i.e., < 5 mg/L). Similarly, Miltner (2010) found
a change point in the 24-h DO concentration range to occur at a benthic chlorophyll a concentration of 182
mg/m2 and suggested that this biomass level not be exceeded in order to maintain DO levels > 4 mg/L, and
protect "existing high-quality waters". For the California streams, algal indicators of oxygen-saturated waters
(RAWDO100) and oxygen-depleted waters (RAWDO50) showed exhaustion thresholds of 45 and 115 mg/m2
chlorophyll a, respectively. Though temperature and other site-specific factors play a role  in determining the
amount of algal  biomass that would result in depression of stream DO, the  scientific basis for establishing
separate biomass endpoints for COLD and WARM wadeable streams remains unclear, and our study does not
further inform this debate.

Statistical  confidence in benthic chlorophyll a thresholds found in this study was not as strong as for AFDM
and nutrient concentration thresholds,  based on analytical criteria for assessing the level of uncertainty in
thresholds. None of the piecewise  regression  analyses for chlorophyll a fulfilled the  "strict" criteria  for
determining confidence in the breakpoint, however some fulfilled the "relaxed" criteria. While a reasonable
degree of  consensus in  thresholds was found among ALIs within assemblages, relatively poor agreement was
found between assemblages—indicating a variable biotic response to chlorophyll a. Furthermore, partial
dependence plots from the BRT analyses exhibited roughly linear relationships between chlorophyll a and
predicted  ALI  response and  suggested only weak thresholds.  Thus  we recommend  use of predictive
regression models to estimate benthic chlorophyll  a concentrations that are quantitatively linked to an ALI
target (such as CSCI, once the index is finalized and a quantitative target is established).

Our limited ability to detect benthic chlorophyll a thresholds may be due to: 1) heterogeneity of the streams
across  more than  100  miles of latitude and 2) low precision  of the  rapid stream assessment  protocol
employed  in ambient surveys. BRT analyses revealed that chlorophyll a had a relatively weak influence on ALI
response variables  within the context of other predictors such  as nutrient concentrations, stream physical
habitat  measures,  meteorological variables, and  land-use.  Fetscher  et al. (2009) found  relatively poor
precision in streams with chlorophyll a values exceeding approximately  50 mg/m2. This is likely due to the
high degree of patchiness of macroalgae,  which is often the primary contributor to high values  of algal
biomass (Sheath et al. 1986, Wehr and Sheath 2003)18. This has led to the suggestion that a higher density of
sampling may be needed in order to overcome some of the sampling error contributed by the patchiness.
18 Meaning that soft algae are more likely to proliferate to nuisance conditions than diatoms, as measured by
  Chlorophyll o.

                                                77

-------
AFDM

AFDM  is an  alternative  measure  of biomass, incorporating  live  as well as dead  autochthonous  and
allochthonous organic matter. As with benthic chlorophyll a, peer-reviewed literature provided little in the
way of examples of wadeable stream studies using quantitative methods to detect AFDM thresholds of effect
on Alls. The only work we found that suggests  AFDM thresholds linked to Alls was Biggs (2000) in which a
50% reduction in the number of EPT taxa was found to correspond to AFDM levels > 5 g/m2, based on a study
of 31 sites across 21 New Zealand streams. This value aligns with the lower range of thresholds found in the
present study (4 to 39 g/m2 overall, with a median values within All categories ranging from 7 to 31 g/m2),
with similar caveats as those stated above regarding mismatch between our two studies in terms of temporal
sampling.

In the  present study,  AFDM  was, overall, the biomass variable with the strongest influence on BMIs and
diatoms in the BRT analyses, and was the second-highest-ranked predictor for the All measure, EPT_Percent.
AFDM exhibited similar thresholds of effect across Alls from different biotic assemblages based on piecewise
regressions, although thresholds tended to be lower for BMI indicators. Furthermore, CIs for these thresholds
were generally narrow, despite the fact that a mix of organic matter sources (labile and refractory) are found
in streams across California, and their modes of action on both algal and BMI communities differ. AFDM may,
in general, be a more suitable predictor of All responses than chlorophyll a, an unsurprising result given that
AFDM  is the most integrative and quantitative  measure of biomass that  we have available. AFDM  is more
quantitative than the percent cover metrics, which either ignore thickness or estimate it into bins of varying
width,  and it is the most integrative biomass indicator because it includes  all forms of stream organic matter
(microbial biomass  and live and dead algal and  vascular plants—in terms  of allochthonous inputs  and
autochthonous production). This is due to the fact that AFDM captures live and dead algal  biomass as well as
fungal  and bacteria  biomass, which are also stimulated by nutrient  overenrichment (Gulis and Suberkropp
2004, Carr et al. 2005). In fact, in their recent review of  stream  nutrient  criteria development approaches,
Evans-White et al. (2013)  asserted that "heterotrophic bases for criteria establishment should be considered
in conjunction with the more traditional autotrophic  bases for criteria  establishment." AFDM has the added
advantage that  it  is  less susceptible to  degradation than  chlorophyll  a, or to variability  in the  algal
C:chlorophyll  a ratio, as noted above.

As an indicator, AFDM  is not without challenges, however. The 75th percentile value of Reference sites (11.9
g/m2) lies squarely within  mid-range of thresholds detected, suggesting that some wadeable streams are
naturally carbon-enriched (e.g., forests with terrestrial carbon inputs). This  would render AFDM an indicator
prone to false positives, without controlling for exogenous factors. It is worth noting that Biggs (2000a) does
not recommend specific criteria for AFDM, because "AFDM is more prone to large measurement error with
low biomass accrual." It may be advisable to move California's PSA program toward piloting a carbon-
enrichment measure that provides information on carbon source as well as biomass. For example, benthic
C:N ratio can be used to indicate algal  (labile) versus terrestrial (refractory)  sources of carbon to sediments
(e.g., Ruttenberg and Goni 1997).  More work may also be needed on detrital-based headwater streams. In
other regions of the country, when nutrients have a disproportionate impact on predator-resistant
consumers, headwater streams have shown long-term declines in organic matter as detritivore activity
increases in response to moderate nutrient enrichment.
                                                78

-------
Other Biomass Gradients

In addition to chlorophyll a and AFDM, we  looked at several other stream primary producer abundance
indicators, including several types of algal and macrophyte percent cover, and soft-algal total biovolume.
Macroalgal percent cover (PCT_MAP) has been suggested as an efficient and informative means of estimating
stream algal biomass (Fetscher and Mclaughlin 2008), because it can be assessed rapidly at a much higher
spatial density than traditional benthic chlorophyll a biomass samples can be quantitatively collected. Results
of our analyses failed to find macroalgal percent cover as a strong predictor of All responses, and no well-
supported thresholds along this gradient were apparent in the analyses we  conducted. However, other types
of analytical approaches  (e.g., those  discussed in the Introduction  to this chapter) that are not based on
thresholds may be useful to incorporate in future work. Also, it is worth noting that nuisance algal mats are of
great concern from an aesthetic standpoint (Biggs 2000, Lembi 2003, Suplee et al. 2009), and  there are
percent cover thresholds in the literature relating to aesthetic (REC-2) beneficial uses. For example, Welch et
al. (1988) and Biggs (2000) have suggested that macroalgal percent cover in  the range of 20-30% and above is
unacceptable from the standpoints of aesthetics and  recreation. Thus macroalgal  percent cover may merit
numeric endpoints on the basis of REC-2 beneficial uses regardless of whether strong relationships between
macroalgal percent cover and Alls can be discerned.

The lack of thresholds of effect of macroalgal percent  cover detected in this study may be a consequence of
the way this  biomass type is  currently measured.  The  rapid,  point-intercept  procedure  that assesses
macroalgal presence/absence along a predetermined grid of 105  points  (Fetscher et al. 2009) takes  into
account  only two-dimensional  (areal) cover,  ignoring thickness, which is  potentially an   important
determinant of biomass. Thus implementing some form of area-weighted biomass, that quantifies algal
biomass at specific points in the stream in addition to recording cover at a high density of observation points,
may be a  means of obtaining higher precision information about stream  algal biomass. However, such an
effort would likely add considerably to field time during sampling.  Nonetheless, percent cover information, as
currently collected,  may  be useful as a screening variable to place a "ceiling"  on the amount of  benthic
chlorophyll a and/or AFDM likely present in a stream (Fetscher et al. 2013). This would require establishing a
relationship  (e.g., via quantile regression) to determine an upper  bound for the "maximum" amount of
chlorophyll a (or AFDM) possible, given a specific percent cover value. Such knowledge would allow the user
to rule out a chlorophyll a or AFDM-based biomass exceedance when  percent cover outcomes are below a
pre-determined value.

Nutrients

TN  and TP  concentration had  strong, statistically-significant  relationships with stream All  indicators;
thresholds detected in this study ranged from 0.13  to  2.1 mg/L for TN  and 0.01 to  0.27 mg/L for TP19
(medians within All categories ranged from of 0.35 to  0.53 mg/L TN and 0.05 to 0.08 mg/L TP among the ALI
categories). These ranges largely fell within the collective ranges of values from  the literature  (0.41 to 1.79
mg/L for TN; 0.0082 to 0.28 mg/L for TP; Table 3.11), thus lending additional support for the numbers we
derived. Most of our analysis- and ALI-specificTN thresholds exceeded the 75th percentile of TN values among
California Reference stream reaches (0.162 mgTN/L), while the 75th percentile of TP
19 Note, however, that many "sensitive" taxa had even lower thresholds, based on TITAN analysis (see Appendix C.2)

                                                79

-------
Table 3.11. Quantitatively determined thresholds of stream (or river) All responses to nutrient
concentrations. "Min." refers to the minimum value from each publication, across all All types and
analytical methods employed. "Max." is the maximum for this value.


Citation
the present study


Baker et al. 2010
Black etal. 2011

Evans-White et
al. 2009

Paul etal. 2007

Qian et al. 2003



Richardson etal.
2007

Smith et al. 2010

Smith et al. 2007

Smucker et al.
2013a
Stevenson et al.
2008


Region
California


Everglades
western United
States
Kansas,
Nebraska,
Missouri
southeastern
Pennsylvania
Everglades



Everglades


New York State

New York State

Connecticut

Mid-Atlantic
Highlands

All
measure(s)
BMI, algae


BMI
diatoms

BMIs


BMIs,
diatoms
BMIs



algal,
macrophyte
and BMI
BMI,
diatom
BMIs

diatoms

diatoms



gradients]
biomass,
nutrients

TP
TN,TP

TN,TP


TP

TP



TP


TN,TP

TP, NO3

TP

TP


threshold
| detection method
TITAN, nCPA, CART,
piecewise regression,
BRT
TITAN and nCPA
piecewise regression

nCPA


nCPA

change point
estimated using the
nonparametric& the
Bayesian methods
Bayesian change
point analysis

nCPA

Hodges-Lehmann
estimation
boosted regression
trees
lowess regression
and regression tree

min.TP
(mg/L)
0.011


0.015
0.03

0.05


0.038

0.011



0.008


0.009

0.065

0.019

0.012

max.
TP min.TN
(mg/L) (mg/L)
0.267 0.13


0.019
0.28 0.59

0.05 1.04


0.064

0.014



0.024


0.07 0.41

0.065 0.98
(N03)
0.082

0.027


max. TN
(mg/L)
2.1


-
1.79

1.04


-

-



-


1.2

0.98
(N03)
-

-

Wang et al. 2007  Wisconsin
Weigel and
Robertson 2007
Wisconsin
                     analysis
fish, BMIs    TN,TP    regression tree        0.06    0.09   0.54    0.61
                     analysis & 2-
                     dimensional
                     Kolmogorov-Smirnov
                     techniques
fish, BMIs    TN,TP    regression tree        0.06    0.06   0.64    0.64
                     analysis
concentrations at Reference sites (0.033 mg TP/L) was within the lower end of the range of TP thresholds we
observed. The agreement in nutrient concentration thresholds between those identified in our study and
what is presented in the literature  is somewhat  surprising, given that  all  but one  of the studies were
conducted in different biogeographic provinces (i.e., east of the Rocky Mountains) and across a diverse array
of stream types. In particular, several studies were conducted in  regions with cooler climates and/or those
with higher  levels of precipitation year-round than that which represents  the bulk of our study region, and
some were conducted in rivers rather than wadeable streams. Black et al. (2011) is the only study from the
                                                 80

-------
western  United States. Their ranges of thresholds of effects on  diatom communities in agriculturally-
dominated to  low-impact wadeable streams in the western U.S were 0.03-0.28 mg/L for TP and 0.59-1.79
mg/Lfor TN.

A recent review by Evans-White et al. (2013) summarized common approaches to stream nutrient criteria
development and thresholds of effect that have been reported for nutrients on BMI and fish Alls. All studies
covered by the review are  included in Table 3.11 of this report. Included in the review were the numeric
criteria for 12  states that were established between 2010 and 2012. Criteria for TP range from 0.01 to 0.49
mg/L and criteria for TN range from 0.13 to 5 mg/L.

Our findings, along with those in  recent studies, suggest that nutrients may be exerting direct effects on ALIs
via means not mediated through pathways typically cited in eutrophication literature (e.g., via increases in
primary production and concomitant reduction  in dissolved oxygen levels; Dodds and Welch 2000). Direct
effects of  stream nutrients can occur  through nutrient toxicity  (Camargo  and  Alonso 2006).  Nutrient
enrichment can also  precipitate changes in instream  food  quality. Under  this  latter  scenario primary
consumers with a high nutrient demand are disproportionately affected by low-quality food relative to those
with lower nutrient demands (Sterner and Elser 2002). This results in altered competitive interactions among
species (Evans-White et al.  2009), which, in turn, decrease diversity and cause shifts in benthic community
structure (Gafner and  Robinson 2007,  Singer and  Battin 2007). More recent studies have  demonstrated
effects of  moderate nutrient loading on headwater streams  as  the  result of effects  on  heterotrophic
production and food web shifts (Davis et al. 2010, Suberkropp et al. 2010).

Statistically, confidence in the nutrient concentration thresholds is high. In BRT models,  nitrogen in one form
or another was the top-ranked predictor for several  BMI and soft algae ALIs and phosphorus was the top-
ranked predictor for two diatom ALIs. This was despite the fact that a wide variety of land use, geographic,
meteorological, geological,  and  local stream physical  habitat variables (as well  as algal  biomass) were
included  as predictors  in the models.  The piecewise  regression  analyses  for  which confidence  in the
breakpoint was highest (i.e., those that passed the "strict" criteria) were  based on ALI responses to nutrient
gradients. TITAN analyses indicated well-supported, community-level change points along nutrient gradients.
BRT  partial dependence plots revealed easily-discernible breaks  in slope across nutrient gradients, not only
for tolerance/sensitivity type metrics, but also for several of the more integrative measures (e.g., IBIs), and a
relatively high  level of consensus in nutrient thresholds from partial dependence plots was observed across
biotic assemblages. Thus, based  on the output of widely different  analytical techniques for multiple biotic
assemblages, narrow ranges of thresholds with high  confidence were realized for both TN and TP.

,3.4.2 Variable          of ALI Types to Biomass and Nutrients: The         Condition Gradient
The  gradient  of thresholds of ALI response to algal abundance  indicators and nutrients illustrates the
paradigm of the biological  condition  gradient (BCG,  Davies and Jackson 2006, Figure 3.2). Integrative ALIs
(such as IBIs) tended to correspond to higher thresholds whereas ALI measures specific to constrained groups
of "sensitive"  taxa generally corresponded to lower thresholds. At the same time, ALI variables that were
based on highly integrative indices (e.g., CSCI, and  the algae IBIs) tended to exhibit threshold responses to
biomass that were not as well-supported as those  for individual  metrics based on sensitive/intolerant taxa
(e.g., EPT_Percent and  lntolerant_PercentTaxa).  This finding conforms to the observation of Baker and King
(2010) that integrative indices may blur taxon-specific change points, relative to information about individual
taxa  or small groups of taxa that share similar autecological characteristics. That notwithstanding, some of
                                                 81

-------
the index-based All variables (e.g., the algae IBI, H20  and the soft algae IBI,  S2) were highly responsive
directly to nutrients, for which clear thresholds were observed. Many of the All metrics also exhibited high
nutrient responsiveness, such as RAWmeanZHR, for which nitrogen was a strong predictor, and for which
marked thresholds were observed. This is not surprising, because this metric incorporates information about
the relative abundance of heterocystous cyanobacteria in the algal community. These organisms are capable
of fixing N2 and are therefore excellent indicators of stream nitrogen limitation (Stancheva et al. 2013). In
general, All variables showed strong responsiveness to  nutrients, but in different ways. Diatom-based Alls
were more influenced by phosphorus (in accordance with the findings of Ponader et al. 2008), whereas BMIs
and soft algae were  more influenced by nitrogen. Thus assessing multiple assemblages concurrently may
provide a broader perspective on stream nutrient status.

Our results did  diverge from the traditional  BCG  gradient paradigm  in  one respect. The  BCG paradigm
suggests that little change will occur in functional level parameters until systems have degraded to levels 4
and 5. However, diatom indicators suggested changes  in  DO regime  may  be occurring at lower levels,
coincident with the loss of sensitive macroinvertebrate taxa.

3.4.3 Study Findings in Context of Policy Applications
The thresholds for algal abundance and nutrients that were derived from this study are based on a data set
that represents an index period  of late spring-early summer. Since nutrient management occurs year-round,
it is important to consider the extent to which our analyses can be applied  outside the index period. We
acknowledge that thresholds may differ for other times of the year and other stream types. For example, our
results are based on  instantaneous  measurement  at low-flow  conditions,  and as such, do not reflect year-
long loads or storm flows.  It is not clear to what degree the types of ALI-stressor relationships we observed
would  hold during rain events. Similarly, although  the target population for the surveys that generated the
data was perennial, wadeable streams, in reality, some  of these streams are  actually intermittent. It is not
always possible to distinguish between perennial and intermittent hydrology unless the site is visited in the
late-summer or fall, prior to the onset of the rainy season  (which is outside of the index period for sampling).

Finally, differences in the type of biotic community that can be supported by different wadeable stream types
(e.g., low-order, high-gradient mountain streams vs. concrete-lined low-gradient streams in developed areas)
may affect the nature of response thresholds. Our statewide data set  included a mix of stream types
spanning multiple  regions  and broad natural and  anthropogenic gradients, including channelized systems.
With the exception of CART analyses of NMS axis breakpoints, we did not explicitly test for differences in
response across regions or stream classes (e.g., natural versus modified  channels). Analyses presented in
Chapter 4 demonstrate that  multiple factors (many correlated  with urban development) may be influencing
and modifying the response  of benthic algal biomass to  nutrients. Thus, it is possible that relationships and
thresholds presented here could be further  refined through stratification of the current data set.
                                                82

-------
                 :: -"-.I
Aboal, M., Puig M.A., Rfos H., and E. Lopez-Jimenez. 2000. Relationship between macroinvertebrate
    diversity and toxicity of (Cyanophyceae Cyanobacteria) in some streams from eastern Spain. Verh.
    Int. Ver. Limnol. 27:555-559.
Baker, M.E. and R.S. King. 2010. A new method for detecting and interpreting biodiversity and ecological
    community thresholds. Methods in Ecology and Evolution, 1:25-37.
Biggs, B.J.F. 2000. New Zealand Periphyton Guideline: Detecting, Monitoring and Managing the
    Enrichment of Streams. Ministry for Environment Publication, Wellington, NZ. 151 pp.
Black, R.W., Moran P.W., and J.D. Frankforter. 2011. Response of algal metrics to nutrients and physical
    factors and identification of nutrient thresholds in agricultural streams. Environmental Monitoring
    and Assessment, 175(1-4): 397-417.
Brenden, T.O., Wang L, and Z. Su 2008. Quantitative identification of disturbance thresholds in support
    of aquatic resource management. Environmental Management, 42:821-832.
Camargo, J.A., and A. Alonso. 2006. Ecological and toxicological effects of inorganic nitrogen pollution in
    aquatic ecosystems: a global assessment. Environmental International, 32:831-849.
Carr, G. M., Morin A., and P.A. Chambers. 2005. Bacteria and algae in stream periphyton along a nutrient
    gradient.  Freshwater Biology, 50:1337-1350.
Chaudhuri, P. and J.S. Marron. 1999. SiZer for exploration of structures in curves. Journal of the American
    Statistical Association, 94:807-823.
Chorus, J., and J. Bartram. (eds.) 1999. Toxic Cyanobacteria in Water.  A Guide to their Public Health
    Consequences, Monitoring and Management. World Health Organization, E & F N Spon., 416pp.
Connolly, N.M. and R.G. Pearson. 2013. Nutrient enrichment of a heterotrophic stream alters leaf litter
    nutritional quality and shredder physiological condition via the microbial pathway.  Hydrobiologia,
    718:85-92.
Cuffney, T.F.,  Brightbill R.A., May J.T., and I.R. Waite. 2010. Responses of benthic macroinvertebrates to
    environmental changes associated with urbanization in nine metropolitan areas. Ecological
    Applications, 20:1384-1401.
Davies, S.P. and S.K. Jackson.  2006. The biological condition gradient: a descriptive model for interpreting
    change in aquatic ecosystems. Ecological Applications, 16:121-1266.
Davies, R.B. 1987. Hypothesis testing when a nuisance parameter is present only under the alternative.
    Biometrika, 74:33-43.
Davis, J.M., Rosemond A.D., Eggert S.L, Cross W.F., and J. Bruce Wallace. 2010. Long-term  nutrient
    enrichment decouples predator and prey production. PNAS, 107(l):121-6.
De'ath, G. and K.E. Fabricius.  2000. Classification and regression trees: a powerful yet simple technique
    for ecological data analysis. Ecology, 81: 3178-3192.
Dodds, W.K. 2007. Trophic state, eutrophication and nutrient criteria in streams. Trends in Ecology and
    Evolution, 22(12):669-76.
Dodds, W. K., and E. B. Welch. 2000. Establishing nutrient criteria in streams. Journal of the North
    American Benthological Society, 19:186-196.
                                                83

-------
Dodds, W.K., Clements W.H., Gido K., Hilderbrand R.H., and R.S. King. 2010. Thresholds, breakpoints, and
    nonlinearity in freshwaters as related to management. Journal of the North American Benthological
    Society, 29:988-997.
Douterelo, I., Perona E., and P. Mateo. 2004. Use of cyanobacteria to assess water quality in running
    waters. Environmental Pollution, 127:377-384.
Dufrene, M. and P.  Legendre. 1997. Species assemblages and indicator species: The need for a flexible
    asymmetrical approach. Ecological Monographs, 67:345-366.
Elith, J., Leathwick J.R., and T. Hastie. 2008. A working guide to boosted regression trees. Journal of
    Animal Ecology, 77:802-813.
Evans-White, M., Dodds W.K., Muggins D.G., and D.S. Baker. 2009. Thresholds in macroinvertebrate
    biodiversity and stoichiometry across water-quality gradients in Central Plains (USA) streams. Journal
    of the North American Benthological Society, 28:855-868.
Evans-White, M.A., Haggard B.E., and J.T. Scott. 2014. A review of stream nutrient criteria development
    in the United States. Journal of Environmental Quality, 42:1002-1014.
Fetscher, A.E., Stancheva R., Kociolek J.P., Sheath R.G., Stein E.D., Mazor R.D., Ode P.R, and L.B. Busse.
    2014. Development and comparison of stream indices of biotic integrity using diatoms vs. non-
    diatom algae vs. a combination. Journal of Applied Phycology, 26:433-450.
Fetscher, A.E., Sutula M.A., Busse L.B., and E.D. Stein. 2013. Condition of California perennial, wadeable
    streams based on algal indicators. Technical Report for the California Surface  Water Ambient
    Monitoring Program (SWAMP), Sacramento, CA.
Fetscher, A.E., Busse L.B., and P.R. Ode 2009. Standard Operating Procedures for Collecting Stream Algae
    Samples and Associated Physical Habitat and Chemical Data for Ambient Bioassessments in
    California. California State Water Resources Control Board Surface Water Ambient Monitoring
    Program (SWAMP) Bioassessment SOP 002. (updated May 2010)
Fetscher, A.E. and K. Mclaughlin. 2008. Incorporating bioassessment using freshwater algae into
    California's Surface Water Ambient Monitoring Program (SWAMP). Technical  Report 563. Southern
    California Coastal Water Research Project. Costa Mesa, CA.
Fovet, O., Belaud G., Litrico X., Charpentier S., Bertrand C., Dollet P., and C. Hugodot.  2012. A model for
    fixed algae management in open channels using flushing flows. River Research and Applications,
    28:960-972.
Gafner, K., and C. T. Robinson. 2007. Nutrient enrichment influences the responses of stream
    macroinvertebrates to disturbance. Journal of the North American Benthological  Society, 26:92-102.
Goslee, S.C. and D.LUrban. 2007. The ecodist package for dissimilarity-based analysis of ecological data.
    Journal of Statistical Software, 22(7):1-19.
Greenwood, J.L., Rosemond A.D., Wallace J.B., Cross W.F. and H.S. Weyers. 2007.  Nutrients stimulate leaf
    breakdown rates and detritivore biomass: Bottom-up effects via heterotrophic pathways. Oecologia,
    151(4): 637-649.
Gulis, V. and K. Suberkropp. 2004. Effects of whole-stream nutrient enrichment on the concentration and
    abundance of aquatic hyphomycete conidia in transport. Mycologia, 96(1): 57-65.
Hijmans, R.J., Phillips S., Leathwick J., and J. Elith. 2013. Dismo: Species distribution modeling. R package
    version 0.8-11.  http://CRAN.R-project.org/package=dismo
                                                84

-------
Kaufmann, P. R., Levine P., Robinson E. G., Seeliger C, and D. V. Peck. 1999. Quantifying Physical Habitat
    in Wadeable Streams. EPA/620/R-99/003, U.S. Environmental Protection Agency, Washington, D.C.
King, R. S., Baker M.E., Whigham D.F., Weller D.E., Jordan I.E., Kazyak P.P., and M.K. Hurd. 2005. Spatial
    considerations for linking watershed land cover to ecological indicators in streams. Ecological
    Applications, 15:137-153.
King, R.S. and C.J. Richardson. 2008. Macroinvertebrate and fish responses to experimental P additions in
    Everglades sloughs. Pp 477-504 In: The Everglades Experiments: Lessons for Ecosystem Restoration
    (Ecological Studies 201), C. J. Richardson. Springer-Verlag, New York.
Lembi, C.A. 2003. Control of nuisance algae. In: Wehr JD, Sheath RG (eds). Freshwater Algae of North
    America  Ecology and Classification. Academic Press, New York pp. 805 - 834.
Mazor, R.D.,  Rehn A., Ode P.R., Engeln M., Schiff K., and C.P. Hawkins. Maximizing the applicability of a
    bioassessment index in environmentally complex regions. Freshwater Science. Under review.
McCune, B. and J.B. Grace, 2002. Analysis of Ecological Communities. MjM Software, Gleneden Beach,
    Oregon,  300 pp.,  ISBN-13: 978-0972129008.
Miltner, R.J.  2010. A method and rationale for deriving nutrient criteria for small rivers and streams in
    Ohio. Environmental Management, 45:842-855.
Muggeo, V.M.R. 2003. Estimating regression models with unknown break-points. Statistics in Medicine,
    22:3055-3071.
Muggeo, V.M.R. 2008. Segmented: An R  Package to Fit Regression Models with Broken-Line
    Relationships. R News, 8/1, 20-25. URL http://cran.r-proiect.org/doc/Rnews/.
Nixon, S. 1995. Coastal eutrophication: A definition, social causes, and future concerns. Ophelia, 41: 199-
    220.
Ode, P. 2007. SWAMP Bioassessment Procedures: Standard operating procedures for collecting benthic
    macroinvertebrate samples and associated physical and chemical data for ambient bioassessment in
    California. Available from http://mpsl.mlml.calstate.edu/phab sopr6.pdf
Ode, P.R., Rehn A.C., and J.T. May.  2005. A quantitative tool for assessing the integrity of southern
    coastal California streams. Environmental  Management, 35:493-504.
Pasek, J. and Tahk A. 2012. Weights: Weighting and Weighted Statistics. R package version 0.75.
    http://CRAN.R-project.org/package=weights
Paul, M. J., and L. Zheng. 2007. Development of Nutrient Endpoints for the Northern Piedmont Ecoregion
    of Pennsylvania: TMDL Application. Tetra Tech, Owing Mills, MD. Prepared for the U.S. Environmental
    Protection Agency Region 3. This report is available on the EPA Region III website at: http://www.
    epa.gov/reg3wapd/tmdl/pa_tmdl/lndianCreekNutrient/NEPStudySEPA.pdf.
Ponader, K.C., Charles D.F., Belton T.J., and D.M. Winter. 2008. Total phosphorus inference models and
    indices for coastal plain streams based on benthic diatom assemblages from artificial substrata.
    Hydrobiologia, 610:139-152.
Potapova, M. and  D.F. Charles. 2007. Diatom metrics for monitoring eutrophication  in rivers of the United
    States. Ecological Indicators, 7: 48-70.
Qian,  S.S., King, R.S., and C.J. Richardson. 2003. Two statistical methods for the detection of
    environmental thresholds. Ecological Modelling, 166:87-97.
Quinn, J.M.,  and B.W. Gilliland. 1989. The Manawatu River cleanup - Has it worked? Transactions of the
    Institution of Professional Engineers, New Zealand, 16:22-26.

                                                85

-------
Quinn, J.M. and C.W. Mickey. 1990. Magnitude of effects of substratum particle size, recent flooding and
    catchment development on benthic invertebrate communities in 88 New Zealand rivers. New
    Zealand Journal of Marine and Freshwater Research, 24: 411-427.
Quinn, J.M. and P.N. McFarlane. 1989. Effects of slaughterhouse and dairy factory wastewaters on
    epilithon: A comparison in laboratory streams. Water Research, 23: 1267-1273.
Richardson, C.J., King R.S., Qian S.S., Vaithiyanathan  P., Quails R.G., and C.A. Stow. 2007 Estimating
    ecological thresholds for phosphorus in the Everglades. Environmental Science and Technology,
    41(23): 8084-8091.
Riseng, CM., Wiley M.J., and R.J. Stevenson. 2004. Hydrologic disturbance and nutrient effects on
    benthic community structure in midwestern US streams: A covariance structure analysis. Journal of
    the North American  Benthological Society, 23:309-326.
Ruttenberg, K.C. and M.A. Goni. 1997. Phosphorous  distributions, C:N:P ratios, and d!3Coc in arctic,
    temperate, and tropical coastal sediments: tools for characterizing bulk sedimentary organic matter.
    Marine Geology, 139:123-145.
Sheath, R.G., Morison M.O., Korch J.E., Kaczmarczyk D., and K.M. Cole. 1986. Distribution of stream
    macroalgae in south-central Alaska.  Hydrobiologia, 135:259-269.
Singer, G.A., and T.J. Battin. 2007. Anthropogenic subsidies alter consumer-resource stoichiometry,
    biodiversity, and food chains. Ecological Applications, 17: 376-389.
Smith, A.J., and C.P. Iran. 2010. A weight-of-evidence approach to define nutrient criteria protective of
    aquatic life in large rivers. Journal of the North American Benthological Society, 29:875-891.
Smith, A. J., Bode R.W., and G.S. Kleppel. 2007. A nutrient biotic index (NBI) for use with benthic
    macroinvertebrate communities. Ecological Indicators, 7(2): 371-386.
Smucker, N. J., Becker M., Detenbeck N.E., and A.C. Morrison. 2013a. Using algal metrics and biomass to
    evaluate multiple ways of defining concentration-based nutrient criteria in streams and their
    ecological relevance. Ecological Indicators, 32:51-61.
Smucker, N.J., Detenbeck N.E., and A.C.  Morrison. 2013b. Diatom responses to watershed development
    and potential moderating effects of  near-stream forest and wetland cover. Freshwater Science,
    32:230-249.
Sonderegger, D. 2011. SiZer: SiZer: Significant Zero Crossings. R package version 0.1-4. http://CRAN.R-
    project.org/package=SiZer
Stancheva, R., Sheath R.G., Read B.A., McArthur K.D., Schroepfer C, Kociolek J.P., and A.E. Fetscher.
    2013. Nitrogen-fixing cyanobacteria  (free-living and diatom endosymbionts): Their use in southern
    California stream bioassessment. Hydrobiologia, 720:111-127.
Steinman, A.D., 1996. Effects of grazers on freshwater benthic algae. In R. G. Stevenson et al. (eds), Algal
    Ecology. Academic Press, San Diego, CA. pp 341-365.
Sterner, R.W. and J. J. Elser. 2002. Ecological Stoichiometry: The Biology of Elements from Molecules to
    the Biosphere.  Princeton University  Press. Princeton, New Jersey.
Stevenson, R.J., Hill B.H., Herlihy AT., Yuan L.L, and  S.B. Norton. 2008. Algae-P relationships, thresholds,
    and frequency distributions guide nutrient criterion development. Journal of the North American
    Benthological Society, 27:783-799.
Stevenson, R.J., Bennett  B.J., Jordan D.N., and R.D. French. 2012. Phosphorus regulates stream injury by
    filamentous green algae, thresholds, DO, and pH. Hydrobiologia, 695:25-42.


                                                86

-------
Suberkropp, K., Gulis V., Rosemond A.D., and J.P. Bernstead. 2010. Ecosystem and physiological scales of
    microbial responses to nutrients in a detritus-based stream: Results of a 5-year continuous
    enrichment. Limnology and Oceanography, 55(1):149-160.
Suplee, M.W., Watson V., Teply M., and H. McKee. 2009. How green is too green? Public opinion of what
    constitutes undesirable algae levels in streams. Journal of the American Water Resources Association,
    45:123-140.
Suplee, M.W., and V. Watson, 2013. Scientific and Technical Basis of the Numeric Nutrient Criteria for
    Montana's Wadeable Streams and Rivers—Update 1. Helena, MT: Montana Dept. of Environmental
    Quality.
Tant, C.J., Rosemond A.D., and M.R. First. 2013. Stream nutrient enrichment has a greater effect on
    coarse than on fine benthic organic matter. Freshwater Science, 32(4):1111-1121.
Tetra Tech. 2006. Technical Approach to Develop Nutrient Numeric Endpoints for California. Tetra Tech,
    Inc. http://rd.tetratech.eom/e pa/Documents/CA_NNE_July_Final.pdf
Therneau, T. M. and B. Atkinson. 2009. mvpart: Multivariate regression trees. R package version 3.1 - 41,
    R port by Brian Ripley, extensions and adaptations of rpart to mvpart  by G.  De'ath. http://CRAN.R-
    proiect.org/package=mvpart
U.S. Environmental Protection Agency. 2000. Nutrient Criteria Technical Guidance Manual, Rivers and
    Streams. U.S. Environmental Protection Agency, Office of Water, Washington, D.C. EPA-822-B-00-
    002.
van Dam, H., Mertens A., and J. Sinkeldam. 1994. A coded checklist and ecological indicator values of
    freshwater diatoms from The Netherlands. Netherlands Journal of Aquatic  Ecology, 28:117-133.
Wang, L, Robertson D.M., and P.J. Garrison. 2007. Linkages between nutrients  and assemblages of
    macroinvertebrates and fish in wadeable streams: Implication to nutrient criteria development.
    Environmental Management, 39:194-212.
Wehr, J.D. and R.G. Sheath. 2003. Freshwater habitats of algae. In: Wehr,  J.D. and R.G. Sheath (eds)
    Freshwater Algae of North America: Ecology and Classification. Academic Press, San Diego, CA.
    pp 11-57.
Weigel, B.M., and  D.M. Robertson. 2007. Identifying biotic integrity and water chemistry relations in
    nonwadeable  rivers of Wisconsin: Toward the development of nutrient criteria. Environmental
    Management 40:691-708.
Welch, E.B., Jacoby J.M.,  Horner R.R., and M.R. Seeley. 1988. Nuisance biomass levels  of periphytic algae
    in streams. Hydrobiologia, 157:161-168.
Wootton J.T., Parker M.S., and M.E. Power. 1996.  Effects of disturbance on river food webs. Science,
    273:1558-1561.
Wright, J.F., Furse M.T., and  P.O. Armitage. 1993. RIVPACS: A technique for evaluating the biological
    quality of rivers in the UK. European Water Pollution Control, 3: 15-25.
                                                87

-------
88

-------
The California State Water Resources Control Board is developing nutrient water quality objectives for the
State's surface waters. Among the approaches that the SWRCB staff is considering is an ecological response
approach, known as the Nutrient Numeric Endpoint  framework (Tetra Tech 2006). The NNE framework,
intended to serve as numeric guidance to translate narrative WQO, consists of two tenets: 1) use of response
indicators to assess the status of waterbody condition  with  respect to eutrophication and other adverse
effects of nutrient-overenrichment and 2) use of models to link regulatory endpoints for response indicators
back to nutrients and other management controls. To facilitate the translation of NNE response indicators to
nutrients, SWRCB  supported the development of  scoping-level nutrient-algal  abundance  models. These
scoping models (e.g., the benthic biomass spreadsheet tool [BBST] for streams) were intended to be used as
a starting point for setting site-specific numeric nutrient targets (Tetra Tech 2006). The intent was that the
BBST helps users determine what the appropriate nutrient concentration targets should be, given other
environmental co-factors at play at a site, based  on the proposed algal abundance endpoints.

The NNE  spreadsheet tools were developed during a period when relatively little California wadeable stream
data was available to optimize the  models. Thus the existing BBST was considered provisional, pending
availability of larger datasets  to aid in  its refinement. Since that time, the Surface Water Ambient Monitoring
Program  (SWAMP) has developed a bioassessment program for perennial, wadeable streams, focused on the
use of benthic macroinvertebrate community composition, water chemistry, measures of physical habitat,
and toxicity to assess ecological condition. SWAMP has supported the development of standardized protocols
for the collection of stream  algal data (Fetscher et al. 2009) with the intent of adding algal community
measures to its suite of indices of biological integrity (Fetscher et al. 2014). Since 2007, data from 1032 sites
have been collected throughout California,  permitting an evaluation of the suitability of the BBST for use in
regulatory application in wadeable streams, which is a goal of the present analysis, and whether additional
refinements are needed (e.g., regionally specific  model coefficients to improve performance).

The objectives of this component of the study are to:

   • Evaluate performance of the BBST for California perennial, wadeable streams,
   • Explore sources of bias  and error in BBST model predictions, in order to recommend potential
      refinements for wadeable stream nutrient-algal abundance models,
   • Understand the relative influence of nutrients and environmental co-factors on stream primary
      producer abundance using Boosted Regression Trees (BRT), and
   • Explore potential regional variation in predictive model coefficients using linearized versions of the
      Dodds and QUAL2K models through a Bayesian Classification and Regression Tree (B-CART)
      analysis.
                                                89

-------
4,2,1                     to
We used existing data to validate the BBST in California wadeable  streams and identify aspects  of its
underlying  models that may  require refinement for particular stream types or regions.  BBST validation
entailed assessment of the accuracy and bias of its underlying models relating nutrients to stream benthic
biomass. This was accomplished by comparing  observed biomass values  (chlorophyll a and AFDM) with
predicted values generated via the BBST models (Objective 1; TetraTech 2006). Furthermore, to facilitate an
exploration of potential sources of model error and inform recommendations for future refinements to the
BBST, the magnitude of  deviation  of BBST model predictions from observed  biomass  was used as the
response variable in random  forest models  with  site-specific and landscape-level factors as explanatory
variables (Objective  2). To build upon this  effort,  we also  used two exploratory approaches to  begin
investigating other ways to model biomass response to nutrients:

    •  BRT (Objective 3) and
    •  B-CART analysis (Objective 4).
4,2,2            on                           Tool                 Testing          1}
The BBST estimates algal density as AFDM (g/m2) and benthic chlorophyll a (mg/m2J using five methods: two
versions of models by Dodds  et al. (1997 and 2002), and three versions of the QUAL2K model: standard,
revised, and revised with accrual (Chapra and Pelletier, 2003).  Model set-up and initial testing are described
in detail Tetra  Tech  (2006)  and summarized here. Table 4.1  summarizes  the models  by type,  input
parameters, and major differences. All five BBST models predict chlorophyll a and AFDM. Total nitrogen (TN)
and phosphorus (TP) are the  base input variables  for both Dodds and QUAL2K models, though additional
variables, such as canopy closure, water temperature, and stream depth, are included as secondary input
variables in the QUAL2K models.

Dodds 1997 and 2002 Models

The Dodds  models (1997, 2002) are statistical log-log regression models of the  mean and maximum20 values
of chlorophyll a as a function of stream TN and TP concentrations from field monitoring data (Eqs. 1 and 2).
The Dodds  (1997) model was developed for wadeable streams in temperate climates, using a compilation of
data from the Clark  Fork River, Montana, and 205 sites throughout North America and New Zealand.  In the
BBST, AFDM is  calculated by  dividing the chlorophyll a values by a constant (2.5; Tetra Tech  2006). The
coefficient  of determination (R2) was 0.43 for the mean seasonal chlorophyll a, and 0.35 for maximum
seasonal chlorophyll a (Chi a).

log(mean Chi a) =  -3.223 + 2.8261og(TN) - 0.431(log [TN])2 + 0.2541og (TP)    Eq(l)
log (max Chi a) = -2.702 + 2.785 log(TN) - 0.433(log [TN])2 + 0.3051og (TP)      Eq(2)
20 In the work of Dodds et al. (2002), "maximum" appears to be intended to represent the spatially-averaged, temporal
  maximum algal growth potential (in response to nutrient and light availability) in the absence of temporary reductions
  in biomass density due to grazing, scour, and other factors. It is thus intended to be a temporal maximum, identified
  via multiple samples taken over the growing season.
                                                90

-------
Table 4.1. Summary of types of models contained in BBST.
Model
Type
Input Parameters
Comment
Dodds1997
Log-log     TN, TP
polynomial
Regression
                        Dodds 1997 is a second order, log-log regression
                        relationship between TN, TP and chlorophyll a. It
                        was developed for wadeable streams in temperate
                        climates, using a compilation of data from the
                        Clark Fork River, Montana, and 205 sites
                        throughout North America and New Zealand (for a
                        total of 300 sites).
Dodds 2002
QUAL2K
(standard)
QUAL2K
(revised)
QUAL2K
(accrual)
Log-log
linear
Regression
Simulation
Model
Simulation
Model
Simulation
Model
TN, TP
Inorganic nutrients,
Stream Depth, Stream
Velocity, Canopy Closure,
Unshaded Solar Radiation
TN, TP, Stream Depth,
Stream Velocity, Canopy
Closure, Unshaded Solar
Radiation
TN, TP, Stream Depth,
Stream Velocity, Canopy
Closure, Unshaded Solar
Radiation, Days of
Accrual
Dodds 2002 is a first order log-log regression
relationship between TN, TP and chlorophyll a. In
addition to the 1997 dataset, the 2002 version
included additional data from the USGS National
Stream Water Quality Monitoring Network stream
data (972 sites from two datasets).
River and Stream Water Quality Model (QUAL2K)
standard version is a parametric representation of
the inorganic nutrient constituents, and physical
parameters such as light, temperature and uses
default model parameters.
In the QUAL2K revised version, the default kinetic
parameters for benthic algae were adjusted to the
Dodds (2002) results for a better fit for application
in California. A nutrient availability fraction was
also added.
Days of accrual, which accounts for the scouring
effect of rain events on algal biomass, was
incorporated using Biggs (2000) regression
coefficients into the revised QUAL2K model.
In the revised model based on Dodds et al. (2002) the regression equation was changed to a first order log-
log  linear relationship and  included additional data from the USGS National  Stream Water Quality
Monitoring Network stream data. In addition to the nutrient concentrations, the effect of stream  gradient,
water temperature,  and latitude was also examined, but not included in  the  linear regression  equation
(Eqs.3-4)
log (mean Chi a) = 0.155 + 0.236 log (TN) + 0.4431og (TP)
log(max Chi a) = 0.714 + 0.372 log(TN) + 0.2231og (TP)
                                                             Eq(3)

                                                             Eq(4)
                                               91

-------
USER INPUTS
Sila:
Analyst:
Dale:
Nutrient Conc«fttrali
Ammonia -H
Nitrite -N
NltrateJt
Organic N
Tolol M (cole)
Inorganic P
Organic P
Total P (c*te)
Hn-.l,.,:l,-.l Solar li.ij
• 1 nti'f manually
•• Estimate



>oa (tngJl )
Average
0.03
0.001
0,14
0.32
0.491
0.025
0.0036
0.02S6
MfnillUMIl
0.02
0.001
0.05
0.1
0.171
0.003
0.002
0.005
Maximum
0.05
0.001
0.2
1
1.251
0.05
001
0-06
far Jon (cat/cm */dJ
Average
572
Minimum
445
Maximum
649
Latitude • Month Range
37.00 II Jun iJI S«P JlJ
Strvain Inputs
Stream Depth (m)
Stream Velocity (m/>)
Water Temperature f*C)
Dow of Accrual {optional)
Canopy Closure
f
CkaunfMt
Light Extinction Coeff. (1/m)
1
OJ
20.0
ja
JJ _
0.9
31
0.5






« Calculate I
Molhvd & Targvt S*J#crion
Select Method:
Oodds '97 max b#nlhic chl a *l
Target Max Benthlc Chi a [mg/rrr) I 10O 1
CoriMgondlng Algal Density fa/m1 AFOWJ
40
C.v/1'ormj ttvnllin: UiiHHjf* 1 mil. vtt* (July 2O1?)


Method
Standard OUAL2K
Revfawd OUAL2K
Revised OUAL2K with accrual jdj
OoootW, nwanCMa
Ocxkto-97, max Chl a
Dooos •02TW, mean Chl a
Dodd> WOS. max Chl a
Laos than or equal to Target
Graatar than Target

Man algal doniity.
a«r condition!
(oym'AFDW)
49
»
35
17
50
12
43
Bonttlic
chlorophyll a
cMlmnto (mgfm1)
12J
97
*r
43
126
31
108
                                                      [Max algal confrlbuUon to DO del
                                                                  Ooddl '97. max benthlc chl i
                                                         0025


                                                         001$

                                                          g v.

                                                         0005
                                                                           100
                                                                         TN (mat)
                                                       Limits not calculated loi Dodd> "37
     a) Input Panel of Dodds 1997 model
b) Output Panel of Dodds 1997 model
Figure 4.1. Example of user interface for the BBST (example highlights output plot for the Dodds
1997 version of the model) with input and output panels. The user inputs nutrient and stream data on
the left panel, and the max algal density and benthic chlorophyll a values are estimated on the output panel
on the right side. The output panel also shows the allowable TN and TP plot for a given site for a user-
selected model. Note that only the TN/TP inputs are required for estimates based on Dodds' models,
whereas the other nutrient types and the environmental data are required for the QUAL2K estimates.

Figure 4.1 shows the user interface screen for the Dodds et al. (1997 and 2002) models in the BBST. The user
enters ammonia, nitrate, nitrite, organic nitrogen, phosphate, and organic phosphorus concentrations in the
input panel  (Figure 41A, and the maximum  (and/or mean) algal density and chlorophyll o are predicted on
the output panel (Figure 4.1b) along with an allowable TN, TP plot. The  plot shows a threshold above which
the combination of TN and TP  is estimated to result in exceedance of a user-stated biomass target. The
observed TN and TP values are plotted on the graph as a triangle to allow the user to visualize whether, and
to what extent, existing nutrient conditions could lead to an exceedance of the  biomass target. Additional
entries on the input panel, such as canopy closure, do not come into play for the Dodds versions of the
model.

QUAL2K Models

Versions of the River and Stream Water Quality Model (QUAL2K) in the BBST are a parametric representation
of the benthic algal component of the  mechanistic steady state model  developed  by Chapra and Pelletier
(2003). This simple parametric  representation was  adapted to provide initial  estimates  of benthic algal
responses to availability of light and nutrients, and can  be adjusted to achieve general agreement with the
empirical relationships developed by Dodds et al. (1997, 2002).
                                                92

-------
The model calculates the steady state algal growth as

 B = *-*»-0"                                                       Eq(5)
Where, Kpmax is the maximum photosynthetic rate at a reference temperature of 20°C, (4>Nb ) is the benthic
algae nutrient attenuation  factor represented by the Michaelis-Menten nutrient  limitation equation for
inorganic nitrogen and  phosphorus.  OLb represents a  light limitation factor with a benthic  algae  light
parameter, Krb is the temperature-dependent benthic algae respiration  rate, and Kdb is the temperature-
dependent benthic algae respiration rate. The prediction of biomass uses only the sum of respiration and
death as a  combined loss  term, and the model is unable to distinguish the  processes independently.
Equations to estimate individual components of equation 5 are not provided in this report but can be found
in the TetraTech 2006 report.

The standard QUAL2K model uses the default model parameters. The user provides ammonia, nitrite, nitrate
as N, total Kjeldahl Nitrogen, phosphate as P, and total phosphorus in addition to hydrology information such
as stream depth and velocity, and site-specific information such as solar radiation (Figure 4.2a).The maximum
algal density and chlorophyll a is predicted on the output panel (Figure 4.2b) along with an allowable TN and
TP plot.

In the revised QUAL2K  model, the default parameters were optimized to achieve a better  agreement
between  the Dodds 2002 equation for maximum chlorophyll a and steady-state QUAL2K predictions based
on the Ecoregion 6 dataset (Tetratech  2006). Thus the standard QUAL2K was optimized for wadeable streams
in temperate climates.

The revised QUAL2K + ACCRUAL model  accounts for the scouring effect of  rain events on algal biomass,
where the  days of accrual is defined as  the average time between flood events greater than 3 times the
median flow in a stream (Biggs 2000). Flow volume is a useful surrogate for velocity, as changes in flow
volume correlate with changes in velocity. Sudden increases in velocity (e.g., by a factor of two to three) can
result in the scour of algae adapted to a constant velocity. A simple statistical representation for the effects
of the hydraulic regime on the biomass was created based on analysis of the mean number of days available
for biomass accrual (Biggs  2000). The best fit regression for  maximum monthly density of benthic  algal
biomass (mg/m2 chlorophyll a) included days  of  accrual and soluble inorganic nitrogen (SIN) concentration,
although  the coefficients for days of accrual are  similar for regressions using accrual only and using accrual
and soluble reactive phosphorus.
                                                93

-------
USER INPUTS
Site:
Analyst:
Date:
Nutrient Connntratf
Ammonia N
HiOite-N
Nitrate N
Organic N
Total N (calcj
Inorganic P
Organic P
Total Pfcafcl
Unshaded 5o«v Rod
>~ Enter manually
<* Estimate



MiAtaflJ
Average
0.03
0.001
0.14
0.32
0.491
0.025
0.0036
0.02W
Minimum
0.02
0.001
0.05
0.1
0.171
0.003
0.002
0.005
Maximum
0.05
0.001
0.2
1
1.251
0.05
0.01
0.06
toij'on real/cm */d)
Average
572
Minimum
445
Maximum
649
Latitude : Montti Rcnf*
37.00 [I J"" dl ^P '
Strum Inputs
SV40M Deptii (ml
Stream Velocity (rn/i)
Water Temperature I°C)
Day* of Accrual (optional)
Canopy Closure
f
Closure 1%)
Light Extinction Cocff. (1/m)
1
0.3
20.0
120
JJ J
=1
0.5


*— Calculate 1
Method & Target Selection
Select Method:
| Standard QUA1.2K benlhic chl » r\
Tatgel Benthic Chi a (mg/m1) I 100
CorrMoondlM Atoal Drnnkv (am»* AFOW) I 40
' v. ... .''"•'.'.,. ' > .... . • ; .,' . M • i. ','/;•. /P;/I



j


RESULTS




M
Method
Standard QUAL2K
Revfeed QUAL2K
Revited OUAL2K with accrual adi
Doddi •97Lme«n CM a
Dodds 57. max Chi a
Dodds W06. mean Chi a
Doddi W06. max Chi a

L M N '

Leu than or equal to Target
Greater than Target

ax algal density, Benthic
ave conditions chlorophyll a
(g/m! AFDW) estimate [mgW]
49 12J
39 97
35 87
17 43
50 126
12 31
43 IDt

Max algal conlfibution to OO deficit (mg'L) ] 1.17


00% •








Standard QUAL2K. benthic cm a
	 «"»•"»'"•»"






•f ' or«™.nn» J








000 010 020 030 040 OSO 060 070
TN imo 1 1
Allowable TN: 0.35
Allowable TP: O.OOSi


      a) Input Panel of Standard QUAL2K model
b) Output Panel of Standard QUAL2K model
Figure 4.2. Example of user interface for BBST Standard QUAL2K model with input and output
panels. The user inputs nutrient and stream data on the left panel, and the maximum algal density and
benthic chlorophyll a values are estimated on the output panel on the right side. The  output panel also
shows the allowable TN and TP plot for a given site. Note that only the TN/TP inputs  are required for
estimates based on Dodds' models, whereas the other nutrient types and the environmental data are
required for the QUAL2k estimates
                                               94

-------
Initial Model Testing

Through the phases of model development, the BBST was tested on two limited datasets from California: 1)
data from  35 sites collected  over 2000-2002  by the Lahontan  Regional  Water Quality Control  Board
(LRWQCB,  N= 93)  and  2) provisional data from the Western  Stream  Environmental  Monitoring  and
Assessment Program (EMAP) dataset from 2000-2002 (n=103). Tetra Tech  (2006) states that both datasets
lacked critical information such as stream hydrology, light availability, and days of accrual at the time of their
availability. Based on comparison of the LRWQCB 6 and EMAP data to Dodds, the report concluded that the
equations are qualitatively reasonable for predicting mean and maximum potential growth of benthic algae
in California streams in the absence of severe light or scour limitation. However, it was noted that the  Dodds
(2002) statistical relationships are quite weak, with R2 values uniformly <0.5. This was attributed to the fact
that light and scour limitation play important roles  in observed chlorophyll a. The report concludes  that
inclusion of average days of accrual and canopy closure might improve the results.

4.2.3 Doto Sources
The California wadeable stream data, as described in Section 3.2.4, were used  in the analyses presented in
this chapter. These data represent one-time sampling events (as opposed to seasonal mean/max data) of
stream algal response,  water chemistry and other in situ variables taken during the time period of April
through October throughout California. These instantaneous snap  shots have an unquantified relationship
with maximum or mean biomass. This variability is compounded by a large latitudinal variability in climate,
rainfall, elevation, hydrology, geology,  land use, and vegetation cover of wadeable streams throughout the
state.

The dataset was used to explain the bias and variance in the predictions. We selected 52 predictor variables
ranging from catchment geological composition to  streambed attributes to meterological data derived from
existing data sources to examine potential factors contributing to model error (Table 3.2).

Estimating Days of Accrual

Biggs (2000) demonstrated that the predictive ability of regression equations could be improved (from an R2
<0.4 to an  R2 >0.7) by inclusion of a measure of average days of accrual. The  revised QUAL2K + ACCRUAL
attempts to account for accrual based on the average time between flood events greater than 3  times the
median flow in a stream. However, no guidance is given for how to estimate the days of accrual from ambient
monitoring  data consisting of one-time site visits. Given that velocity and flow are one-time measures at the
sites in the  wadeable stream data set, we developed a methodology to estimate the average days  of accrual
using the size of storm events as measured by daily  precipitation data, and readily available parameters.

A daily  precipitation database for all the  study  sites  was developed  by matching site data with daily
precipitation from the Daily Global  Historical Climatology Network (GHCND; NOAA National Climatic Data
Center) and was spatially interpolated, using the nearest-neighbor method,  during the period from 2007-
2012 (Figure D.I). Raster files of daily precipitation data by site were generated for the period of 2007-2011.

We established cutoff values for scouring events as a function of storm size (daily precipitation) and extent of
urbanization using best professional judgment (Table 4.2). The days of accrual were estimated for each site
using a recursive algorithm that counts the number of days between the sampling date and an antecedent
"scouring" event with a value equal to or higher than the cutoff. A nominal default of 120 days accrual was

                                                95

-------
used for any sites with no  precipitation data, a time period roughly corresponding to the duration of the
Mediterranean dry season.
Table 4.2. Stream scouring cutoff values for the watersheds developed based on precipitation and
watershed imperviousness
                        Watershed Imperviousness   Cutoff for Scouring Event,
                                   (%)               Precipitation (inches)
                                   <5                         0.5
                                   5-25                        0.4
                                  25-50                       0.3
                                   >50                        0.2

4,2,4                 A/WE      Tool
The five underlying models of the NNE BBST tool: 1)  Dodds 1997, 2)  Dodds 2002, 3) QUAL2K 4)  QUAL2k
Revised, and 5) QUAL2k Revised + Accrual were validated against the observed benthic chlorophyll a data for
1031 sites.

Currently, the spreadsheet models are set up to conduct site-specific assessments. In order to conduct model
runs in a more efficient manner, the BBST was receded using R scripts, enabling batch runs for data from all
sites. Model output from single runs using the original interface was checked against output of the R script to
ensure accuracy of model translation. The models were validated  by comparing predicted versus observed
values, using a linear regression, and the performance was measured in terms of coefficient of determination
(R2) and slope.

4,2,5     Analysis (Objective 2)
We conducted analysis of factors affecting bias using the randomForest package in R (Liaw and Wiener 2002).
A dataset with predicted-minus-observed values for chlorophyll a and AFDM and the 52 selected explanatory
variables was constructed. For the purpose of selecting the top predictor variables, any missing values of the
explanatory variables  were populated in  the dataset using  nearest-neighbor  interpolation.  After  the
preliminary predictor variable selection  process, no interpolated values were used in the subsequent random
forest regression analysis. Bias is the deviation of predicted values from the observed values (chlorophyll a
for this study) resulting from  usage of  poor explanatory variables in the model or just incorrect choice of
models. In  the bias analysis process we try to examine the impact of a  single or a group of explanatory
variables on the prediction abilities of a given model.

Nonlinear multiple regression techniques  in  the  randomForest  package were  used to determine  the
importance of the predictor variables. The strength and correlation  of the predictor variables were estimated
using the out-of-bag error method  (OBEM). The error is estimated internally during the  run.  Each tree  is
constructed using a different bootstrap  sample from the original data. In order to cross-validate, about one-
third of the cases are left out of the bootstrap sample used for tree construction and are used to estimate the
OBEM error. Each regression forest produces a variable importance plot based on the percent increase mean
square error (MSE) for a given  explanatory variable, and the total variance explained by the multivariate
regression.
                                                96

-------
4.2.6                                                          3J             C4I?T
                 4J
Two statistical methods were used to conduct exploratory analyses in order to suggest model refinements:
BRT and B-CART analyses. Boosted regression trees allow nonlinear relationships and variable interactions to
be represented in model predictions. Bayesian CART analyses provide a simplified set of regression models to
predict algal biomass by site class, along with a set of classification rules to define groups.

BRT Analyses
BRT and  partial Mantel tests were  used to  conduct exploratory analysis to investigate the relationship of
nutrients to biomass response variables within a context of a large suite of environmental co-factors.  See
Chapter 3 for a detailed description of BRT and partial Mantel tests.

Bavesian CART

To further explore potential refinement of models explaining benthic algal response to nutrients, we applied
a Bayesian  Classification and Regression Tree (B-CART) analysis to the data matrix containing nutrients  and
potential co-factors influencing the  response of benthic chlorophyll a. Bayesian CART is an approach to the
development  of Regression Trees that is  informed by the  analyst's prior  knowledge of tree form  and
distribution of potential model  coefficients. Unlike the Classification Tree analysis applied in Chapter 3,
Regression  Tree analysis is designed to optimize the fit of regression  models within each final group rather
than the difference in mean values among groups.

Regional  variation in nutrient - Chlorophyll a relationships  for lakes across the  United  States has been
successfully explored using Bayesian Classification and Regression Tree (B-CART) analysis (Freeman et al.
2009). Regression Trees are designed  to  simultaneously  classify observations  and  optimize the  fit of
regression  models within each final  class. Thus they provide a tool for assessing whether regionally specific
model coefficients are appropriate.

We applied the CGMIidCART program for Bayesian CART analysis developed by Chipman et al. (available at:
http://www.rob-mcculloch.org/code/CART/index.html) to data matrices containing logic  chlorophyll a mg/m2
biomass as the dependent variable and multiple explanatory variables. Because the CGMIidCART program
cannot function with missing values, we substituted medians for missing values in the matrices21. Based on
the protocol suggested by Chipman et  al.  (2002), we evaluated  regression trees  using a range of model
parameters, chose  the "most visited tree" among model iterations for each model, and used Aikake's
Information Criterion (Burnham and Anderson 1998) to compare the fit of alternative models. See Appendix
D for a detailed discussion of Chipman's fitting protocol.

Bayesian CART allows the  user to  choose  both potential  classifier  variables and variables to include in
regression  models for end nodes.  We tested  alternative  regression tree  models based on  selection of
classification variable sets and selection of regression variable sets. Two types of regression  trees were fitted
to the data based on regression variables included, one set analogous to the modified Dodds model ("Dodds-
type"), with logi0TP,  (logi0TP)2, logi0TN,  (logi0TN)2, logic (days accrual),  and  logic (days  accrual)2 as
independent variables in the final regression models. The second-order term for total N  allows the model to
incorporate nutrient saturation effects at high levels. Although Dodds only included a second-order term for
21 This corresponded to 5 percent of values for variables retained in the final models.
                                                 97

-------
total N, we added a second-order term for total  P as well to reflect the possibility of P saturation at high
nutrient levels. The second set of regression tree  models were analogous to the QUAL2K model ("QUAL2K-
type"), in which the steady state value for algal biomass, B, is predicted as:

       B = (Kpmax-0Nb-0Lb)/(krb + kdb)

Where B = steady state value for biomass

       Kpmax = maximum photosynthetic rate  in the absence of limiting factors

       0Nb = nutrient limitation term

       0Lb = light limitation term

       krb = loss rate due to respiration

and    kdb = loss  rate due to death (e.g., grazing).

The equation can  be linearized by applying a log transformation to both sides of this expression:

       Log10 B =  Iog10 Kpmax + Iog10 0Nb + Iog10 0Lb - Iog10 (krb + kdb)

Temperature dependence is incorporated into QUAL2K with the Arrenhius relationship:
       K    -K      n
-------
where I0 = light incident on the water surface

        Ke = light attenuation coefficient

        H = water depth, and

        KLb = half-saturation coefficient

The term Ke should be proportional to turbidity, so an interaction term, Turbidity x Depth, was added in the
regression tree. The incident light  term was calculated  as solar radiation x fraction cloud-free sky x (1 -
fraction canopy). To account for the effects of potential light saturation, a second-order available light term
(incident light2) also was incorporated into the regression tree model. Potential algal loss due to scouring was
incorporated into the regression tree with the days of accrual and days of accrual2 terms. Finally, each type of
regression tree model  was estimated with two alternate forms, one using total nutrients (TN, TP) and one
using the dissolved inorganic forms (DIN = NH4-N + NOX).

We identified a reduced set  of potential classification variables from the full potential set described in Table
3.2 supplemented by a few new interaction terms. In addition to the classification variables in Table 3.2, we
considered interaction terms for turbidity x depth, stream power (watershed area x slope), stream power x
antecedent  precipitation,  and stream power x antecedent precipitation x % sands and  fines (an  index of
potential substrate disturbance). Details of the  classification variable  reduction process are provided in
Appendix D.

The  final reduced set of  classification variables  used to develop Regression Trees is listed  in Table 4.3.
Because of the redundancy of PSA ecoregion and geographical coordinates as classifiers, we constructed two
final sets of alternative regression trees, one set  using PSA ecoregions but not geographical coordinates as
classifiers, and  a second set excluding PSA ecoregions as classifiers but using  geographical coordinates as
classifiers to define "empirical  nutrient regions". We also compared trees with and  without geographical
coordinates  as  predictor  variables  in regressions to  allow them to  reflect the effect of smooth climatic
gradients.

Final model  selections were  based  on maximization of log-likelihood values  when comparing models with
equal numbers  of predictor variables (regression coefficients and classifiers)  or minimization of Aikake's
Information  Criteria (AIC)  values for comparison of disparate models (Burnham  and Anderson 1998). Values
of AIC can only  be compared for model runs using equivalent training sets. We also calculated an r2 value for
regressions  relating predicted to observed values to summarize the percent  variation  explained in each
regression tree.
                                                 99

-------
Table 4.3. Final set of classification variables used in B-CART analysis.
 Classification
 Variable	Definition	
 Longitude             Degrees longitude
 Latitude               Degrees latitude
 NH4                  Instream NH4 value (mg NH4-N/L)
                       Percent NLCD "Code 21" land use within a 5-km
 URB21 5K               ,.   .         ..
       ~               radius from sampling site
 JulianDay             Day of year (1-365)

 PSAc                  Perennial Stream Assessment ecoregion (1-6)

 Year                  Year of sample
 Conductivity           Instream conductivity
                       Site disturbance status (Reference, Intermediate,
                       Stressed) as defined  in Chapter 2


Because B-CART forces the computation of regression coefficients for each final node of the tree regardless
of whether or not they are significant, we ran separate regression analyses on each final node (or combined
nodes where final  node size was insufficient for model fitting). We conducted regression analyses  in SAS
using PROC GLMSELECT  with the default  stepwise selection method (SAS version 9.3, Copyright [©] 2002-
2010, SAS Institute, Gary, NC). Final regressions  were tested for model assumptions, i.e.,  normality of
residuals (Wilk-Shapiro test)  and homogeneity of variance (via plots of residuals versus predicted values),
although these assumptions are not required for the original nonparametric B-CART analyses.

4.3
4,3,1 Mode! Performance (Objective 1}
The validation data illustrate that the BBST has very poor model fits for all model types, although QUAL2K
(standard, revised, and  revised with accrual) performs  marginally better  than the Dodd's models for
chlorophyll a  (Tables  4.4-4.5;  Figure 4.3). The models overpredict lower values and underpredict  higher
values for both chlorophyll a and  AFDM, with slopes ranging from 0.1-0.55, and positive intercepts ranging
from 30-122. Results of linear regressions are shown to illustrate the poor match between observed  values
and model predictions. However, because model fits were so poor, model assumptions for linear regressions
were not  met even after multiple standard transformations (log, square root, inverse, power) and  higher
order equations were applied. In general, both variance and residuals tended to decrease with the mean, and
residuals were not normally distributed.
                                                100

-------
Table 4.4. Distribution of residuals of >0.5 standard deviation of mean for chlorophyll a by model type.
Slopes were significantly different from 1 at p-value<0.05.
0.5 Standard Deviation
Chlorophyll a
Dodds97 Mean
Dodds97 Max
Dodds02 Mean
Dodds02 Max
QUAL2K Standard
QUAL2K Revised
QUAL2K Revised
+ ACCRUAL
Number of sites Over
806
806
806
806
868
846
846
Predict
88
301
91
284
234
264
213
Under Predict
96
35
83
26
62
27
38
Overall the model performance is  comparable for all  sites  combined as well as for the Reference and
Intermediate sites (Table 4.6), and it declines for Stressed sites. The model performance is comparable for
the three QUAL2K models, even though the QUAL2K revised with accrual  was expected to perform better
than the other two versions due to customized accrual information. All  models tend to  overpredict the
chlorophyll a and AFDM concentrations.
Table 4.5. Model performance (R2, slope and intercept) for all sites combined.
Chlorophyll a

Dodd 97
Dodd 02
QUAL2k
Standard
QUAL2K
revised
QUAL2K
revised
+Accrual

R2
0.16
0.20
0.15
0.26
0.20
Mean
Slope
0.33
0.32
0.33
0.34
0.43

Intercept
0.93
0.99
1.23
1.48
1.15

R2
0.16
0.18
NA
NA
NA
Max
Slope
0.31
0.22
NA
NA
NA

Intercept
1.43
1.72
NA
NA
NA

R2
0.21
0.21
0.13
0.25
0.21
Mean
Slope
0.44
0.40
0.36
0.40
0.54
AFDM

Intercept
0.43
0.53
0.82
1.04
0.64

R2
0.21
0.20
NA
NA
NA
Max
Slope
0.42
0.27
NA
NA
NA

Intercept
0.93
1.28
NA
NA
NA
                                               101

-------
Table 4.6. Model performance (R2) for all, Reference, Intermediate, and Stressed sites
(see Chapter 2) for predicted mean chlorophyll o.
     10000

  Q
 =  1000
  o
 T3
  E
  0.
      o.oi
       0,01
            D.I

Dodd's97
Dodd's02
QUAL2k Standard
QUAL2K revised
QUAL2K revised
with accrual
R2
(All Sites)
0.16
0.20
0.15
0.26
0.20
R2
(Ref + Inter)
0.11
0.15
0.11
0.20
0.13
R2
(Stressed)
0.04
0.07
0.03
0.11
0.10
all sites
Reference +
Intermediate sites
                          100
                               1000  10000 OBI   0,1
                                                         100   1000   10000 0.01
                                                                               1    10    100    1000  10000
                                         observed chlorophyll a
Figure 4.3. Sample plots of validation data showing measured versus predicted chlorophyll a by
standard QUAL2K model, with 1:1 slope lines.
                                               102

-------
4,3,2         Forest Regression for        Variance Analysis (Objective 2}
    • The random forest regression ranks the explanatory variables that account for the variance and
     bias. These explanatory variables can be divided into three major categories: 1) water chemistry
     variables, 2) other site-specific parameters, such as physical habitat and 3) land use (Table 4.7).
     Recurring key explanatory variables from these categories are observed for both chlorophyll a and
     AFDM. The variance explained by the models ranges from 0-53%, signaling that model fit could be
     improved if refinements are made for how these  variables are currently used in the models.
    • Water chemistry  variables,  such  as nutrients (TN,  TP,  NOX,  NH4, and SRP),  chloride, and
     conductivity were highly ranked in all our random forest models. Nutrients are consistently the
     most important explanatory variables, except for the cases in which Dodd's 97 mean AFDM models
     were being evaluated. Chloride, a reasonable surrogate for urbanization, was a key indicator for the
     chlorophyll a predictions.  Conductivity is also a  strong indicator of water quality associated with
     urbanization. A number of site-specific parameters, such as air and water temperature, canopy
     cover,  solar radiation, reach  slope, mean width of wetted channel, substratum composition (fines
     and particle size  less than sand) are ranked high for  all the models. Coarse particulate organic
     matter is a critical cofactor mostly for the AFDM regression models.
    • Indicators of urbanization ranked highest as predictors in the Random Forest regression analyses
     for both  chlorophyll a and AFDM. For some of the variables, such as road density and urban land
     use, the value measured within the  1  km radius  of the  site was important,  rather than the
     watershed level value. In contrast, the watershed-level values for urban land use and Code 2122, as
     well as values of Wl_Hall (an indicator of human disturbance that is local to the sampling reach)
     were important co-factors for the Dodds models.
    • Note that explanatory variable rankings  are qualitative rather than quantitative. For example, in
     the Dodds 97 mean chlorophyll a regression (Figure 4.4a), the reach slope was ranked as the most
     important variable, with 17% MSB, and water temperature also ranked in the top 15. However, the
     MSB of  water temperature was negative, implying  zero  influence.  Some of the regressions
     performed poorly for all variables. For example, for the QUAL2K with ACCRUAL model (Figure 4.4c),
     regression analysis showed no  significant  relationship  between  the predicted-minus-observed
     chlorophyll  a  and the explanatory variables, and the variable  importance ranks also  showed
     insignificant MSB change. The influence of the explanatory variables was  stronger for the max
     Dodds models compared to the mean Dodds predictions for both chlorophyll a (variance for mean
     = 0.18, and max = 0.53, for Dodds 02) and AFDM  (variance for mean = 0, and max = 0.46, for Dodds
     02). The  QUAL2K with accrual had the lowest variance and the weakest relationships between the
     predicted-observed biomass and the explanatory variables. It should be noted that the % Increase
     MSB values are not  comparable between the models and only  have meaning  when making
     comparisons between variables within a given model.
22 "Code 21" encompasses a wide range of land uses primarily characterized by heavily managed vegetation (e.g., low-
  density residential development, parks, golf courses, highway medians).

                                                103

-------
Table 4.7. Variables ranked according to importance for random forest regression for Chlorophyll o (Chi o) and AFDM by model type.
Standard Qual 2k
Chi a
Urban LU
(1km)
Road Dens
(1km)
TN
TKN
TP
Chloride
NOX
Alkalinity
SRP
NH4
Cond
Substratum
(sand)
Alkalinity
TKN
WetChnl
Width
TP
CPOM
Catchment
Slope
(1km)
Urban LU
(1km)
Cond
Road Dens
(1km)
Area
WetChnl
Width
AFDM
Substratum
(fine)
TN
AirTemp
(same mo)
Geology Cenoz
Wet Chnl
Width
Geology
(Quat)
Road Dens
(WS)
Water
Depth
Turbidity
Precip
(3 mo)
Area
Solar Rad
Geology
(ign/met)
Urban LU (WS)
Dodds 97 mean
Chi a
Reach
Slope
NOX
TN
Code 21
(WS)
TP
Catchment
Slope
(1km)
Chloride
Alkalinity
Solar
Radiation
SRP
Substratu
m (sand)
WetChnl
Width
Canopy Cover
                                                                104

-------
Reach Slope
NOX mg L
TNmgl
CODE 21 WS
TP rng L
Catchment slope 1 km
Chloride rngL
Alkalinity mg L
Solar Radiation same mo
OP04 mg L
Substrate sand see
Conductivity uS cm
MaxWl HALL
Water Temp
Temperature
o
0

0










°
5 10 15
%lncMSE
TNmgl
URBAN 1K
TPmgL
Road Density 1km
NOXmgL
Substrate sand size
URBANWS
OP04 mgL
Chloride mg L
Alkalinity mg L
Conductivity uS cm
Temperature
Water Temp
AREA Km 2
Geolgv Quaternary
0













0 5 10 15 20
KlncMSE
CODE 21 1Km
Chloride mg L
Max Air Temp 3mo
Substrate sand size
Alkalinity mg L
Mid Channel Shade Canopy Cover
TKN mg L
Ag 1 Km
Mean width wet channel
TPmgL
CPC-tf
Catchment slope 1 km
Flow
URBAN 1 K
Monthly Precip 3 mo
Conductivity uS cm
Solar Radiation 3 mo
Max Air Temp same mo
Road Density 1km
ARE A Km 2
Mean Depth wet channel
Measured Water Depth
Fine Substrate
NOX mg L
Ammonia mg L
Road Density WS
MaxWl HALL
Strahler Order
Geolgy Cenoz
Catchment Slope WS
0
0

0

o
0
o
0
o
o
o
o




°


0
I I I I
2468
%lncMSE
                 a) Dodds 97(revised) mean Chi a
                                                       b) Qual2k (revised) Chi a
                                                                                                    c) Qua! 2k with accrual Chi a
TPmgL
OPO4 mg L
Fine Substrate
TN mg I
Conductivity uS cm
Solar Radiation same mo
Reach Slope
Sample Site Elevation
Road Density 1km
NOX mg L
Substrate sand size
CPOM
CODE 21 WS
Mid Channel Shade Canopy Cover
Mean width wet channel
0

0









0
I [ I I I I
5 10 15 20 25 30
%lncMSE
TP mg L
TNmgl
NOX mg L
Fine Substrate
OP04 mg L
Reach Slope
Conductivity uS cm
URBAN 1K
CPOM
Chloride mg L
Solar Radiation same mo
Substrate sand size
Mean width wet channel
Mid Channel Shade Canopy Cover
Ammonia mg L
0

0









o
5 10 15 20 25
%lncMSE
NOX mg L
Fine Substrate
TN mgl
CPOM
Conductivity uS cm
Chloride mg L
OPO4 ma L
TPmgL
Water Temp
Temperature
Mid Channel Shade Canopy Cover
Mean width wet channel
AREAKm2
Flow
Solar Radiation 3 mo



0


0
0








10 20 30 4C
%lncMSE
                   d). Dodds 97 max AFDM
e). Dodds 02 max AFDM
f). QualZK standard AFDM
Figure 4.4. Relative influence of variables for some selected models for chlorophyll a (top row: a, b, c) and AFDM (bottom row: d, e, f). All
other model output is provided in Appendices D3 - D8. Predictor variables ranked on the Y-axis, and the mean squared error values are listed on the
x-axis. Variable names are given in Table 3.2.
                                                                      105

-------
4.3.3                                3)
Results from the BRT analyses examining nutrient  and other environmental co-factor effects on  the  six
biomass response variables are summarized in Table 4.8 and  in the heat map in Figure  4.5. The relative
influence of all predictor variables for the six models are provided in Figures D.3 - D.8. Wide differences were
observed among biomass types in terms of what environmental co-factors most strongly predicted biomass
levels.

The final BRT models had  numbers of predictors ranging from  27 to 33. In no case was a nutrient the top
predictor for any given biomass type, however for AFDM, NH4 was the second highest-ranked predictor, with
a relative influence  of nearly 8%, and  for chlorophyll a, NOX was the fifth highest-ranked  predictor, with a
relative influence of nearly 6%. Partial Mantel tests (Table 4.9) indicated that nitrogen correlated significantly
with both of these  biomass variables when other high-ranking predictors from the BRT models, as  well as
spatial autocorrelation, were accounted for. In the case of chlorophyll a, SRP was also significant, although
the Mantel partial correlation  coefficient was very low. Nutrient predictors collectively had a low relative
influence on the percent-cover biomass types (i.e., PCT_MAP, PCT_MCP, and PCT_MIAT1),  and no nutrients
were significantly correlated with percent cover metrics based on the partial Mantel tests.

Of the six biomass types tested  in this study, chlorophyll a was the most directly responsive  to nutrients,
based on BRT analysis, with a total of >16% of the relative influence on chlorophyll a attributable to nutrient
concentrations, amid the  20 other environmental co-factors  (physical  habitat, meteorological, landscape,
water chemistry; see Table 4.7) included in the models (Figure 4.5). The biomass variable that was least
responsive  to nutrients  was macrophyte percent cover (PCTJV1CP), for which <6% relative influence was
attributed to nutrients (Table 4.8). Among the nutrients, nitrogen species were invariably associated with a
higher degree of influence  on biomass than phosphorus species, but the overall relative influence of nitrogen
vs. phosphorus varied by biomass type. The difference between the two was most dramatic  for chlorophyll a,
with nitrogen  species collectively accounting for 3 times the relative influence  realized for phosphorus
species.

Of the non-nutrient  predictors, stream temperature was the most likely to influence biomass; it was the top-
ranked predictor for both chlorophyll a  and percent presence of  thick (lmm+) microalgae (PCTJV1IAT1;
Appendix Dl), exhibiting almost 10% relative influence in both cases. The top-ranked predictors for the other
biomass response variables included the substratum-specific percent fines and percent sand + fines,  as well
as percent canopy cover and conductivity.
                                                106

-------
 Table 4.8 Relative influence of nutrient species on abundance of stream biomass of six different types,
 from BRT models that included environmental co-factors (see Figure 4.5 for a full list of predictors in each
 final model). "Model cv correlation (se)" refers to the cross-validation correlation coefficient (with
 standard error), indicating reliability of each model (Elith et al. 2008). PCT_MAP is macroalgal percent
 cover; PCTJV1CP is macrophyte percent cover; PCT_MIAT1 is percent presence of thick (lmm+) microalgae.
 Dashes indicate that the predictor in question was not included in the final model for that biomass type.
 Bold values correspond to the highest ranked nutrient predictor for that biomass type.
Biomass
Type
AFDM
(N = 847)
Chlorophyll o
(N = 878)
PCT_MAP
(N = 771)
PCT_MCP
(N = 771)
PCT_MIAT1
(N = 770)
soft algal total
biovolume
(N = 914)
Highest Ranked
Predictor
(relative influence)
fines (%)
(11.91)
stream temperature
(9.85)
canopy cover (%)
(13.4)
sand & fines
(%) (17.43)
stream temperature
(9.23)
conductivity (14.81)
Model CV
Correlation
(se)
0.628
(0.027)
0.503
(0.051)
0.643
(0.022)
0.680
(0.025)
0.458
(0.03)
0.599
(0.024)
Relative Influence of (rank)
Trees Final Model TN
6000* 30 1.38
(24)
6000* 25 3.78
(10)
8250 33 2.73
(11)
7200 33 1.80
(17)
5550 32 2.66
(17)
5000 27 1.51
(26)
NOx
-
5.97
(5)
3.17
(7)
1.16
(25)
2.85
(15)
3.21
(9)
NH4
7.70
(2)
2.83
(14)
2.25
(16)
1.45
(21)
3.07
(14)
2.00
(22)
TP
-
-
1.49
(23)
0.72
(33)
3.40
(12)
3.31
(8)
SRP
3.83
(11)
4.05
(9)
1.46
(24)
0.77
(32)
3.55
(11)
2.08
(21)
*For these models, tree number was not optimized. A fixed number of 6,000 trees was used.
                                                 107

-------
                  URBAN_2000_WS -
                site disturbance class -
                      Ag_2000_5K -
                   URBAN_2000_1K
           total precipitation (3-mo span)
                      stream depth
                      Ag_2000_WS
                 CODE_21_200G_WS
    mean monthly % cloud cover (3-mo span)
                         elevation -
              sedimentary geology (%) -
                      stream width -
    mean monthly solar radiation (3-mo span) -
                     days of accrual -
                    watershed area •
      W1_HALL (riparian disturbance index) -
                 CODE_21_2000_5K-
   •   mean monthly maxtemp (3-mo span) -
                        longitude -
                        ecoregion
                         turbidity
                            PH
                      slope, reach -
                         alkalinity ~
                        discharge •
       coarse paniculate organic matter (%) -
                         latitude •
                        fines (%) -
                  stream temperature •
                       conductivity -
                   sands&fines(%)-
                    canopy cover (%)
                            TP-
                            TN-
                            SRP
                            NOx
                            NH4
                                  chlorophyll_a     AFDI.I    soft_algal_Siovol   PCT_HAP
                                                            biomass type
PCTJJCP
                                                                                          PCT_M1AT1
Figure 4.5. Heat map showing relative influence (%) of predictor variables (nutrients and
environmental co-factors) on biomass response variables, from six independent BRT models. Yellow
= low influence, red = high. The five nutrient-based predictor variables are grouped at the bottom of the
graph. All climate variables are based on data for the month in which the sample in question was collected,
averaged with the prior two months. Grey boxes indicate that the corresponding predictor type was not
included  in the final BRT model for that biomass type.
                                                       108

-------
Table 4.9. Partial Mantel coefficients (95% CIs) for correlation between nutrient predictors and biomass
variables; p values. Grey boxes correspond to explanatory variables that were not included in the partial
Mantel test for the All variables in question. "Space" refers to the geographic distance between sites (for
testing the significance of spatial autocorrelation). "NS" = not significant; dashes correspond to predictors
that were included as explanatory variables in the partial Mantel tests for the indicated biomass
variables, but (because they did not fall under the categories of nutrients or "space") were not the focal
variable in the tests. Values in bold correspond to significant  partial  Mantel tests.
Explanatory
Variable
Chlorophyll a
AFDM
Soft Algal
Biovolume
PCT MAP
PCT MCP
PCT MIAT1
TN



NOX


NH4


TP


SRP


space


CODE_21_2000_5K

stream temperature

alkalinity

mean monthly max temp
 (3-mo span)

pH

Ag_2000_WS

sedimentary geology (%)

canopy cover (%)

latitude

fines  (%)

slope, reach

discharge

turbidity

stream width

sand  & fines (%)

conductivity
     0.156
(0.138 - 0.175);
     0.001
                    0.083
               (0.060 - 0.103);
                    0.001
     0.041           0.005
(0.023 - 0.058);  (-0.008 - 0.018);
     0.019            NS
     0.072           -0.009
(0.054 - 0.090);  (-0.027 - 0.006);
     0.001            NS
                0.014           0.020
            (0.006  - 0.022);  (0.008 -  0.030);
                 NS            NS
                                   0.000
                               (-0.009 - 0.009);
                                    NS
                0.040
                 -0.004
                 -0.036           -0.041
             (-0.057 - -0.011);  (-0.062 - -0.020);
                   NS              NS

                                 0.041
                             (0.019 - 0.055);
                                   NS
                                 0.045
                             (0.001  - 0.071);
                                   NS
                                 -0.052
                             (-0.077 - -0.023);
                                   NS
                                 0.048
                             (0.025 - 0.074);
                                   NS
                 -0.019           -0.064
            (0.029 - 0.050); (-0.015 - 0.004); (-0.037 - -0.006); (-0.082 - -0.046);
                0.001            NS              NS              NS
                                                     109

-------
Table 4.9. (continued)
Explanatory                                       Soft Algal
Variable	Chlorophyll a      AFDM	Biovolume     PCT_MAP      PCT_MCP      PCT_MIAT1
elevation
days of accrual
Ag_2000_5K
watershed area
longitude
coarse particulate
 organic matter (%)
W1_HALL (riparian
 disturbance index)
mean monthly % cloud
 cover (3-mo span)
Figure 4.6 shows  the joint influence of NOX and  a  critical environmental  co-factor, temperature,  on
chlorophyll a levels. The latter had a particularly strong influence, corresponding to an abrupt rise in biomass
response within the range of approximately 26-28°C. Both stream water temperature on the day of sampling
and antecedent ambient air temperature were important determinants of chlorophyll a concentrations based
on BRT models (Figure 4.5 and Appendix D2). Together, these two temperature measures accounted for 15%
of the relative influence in predicting chlorophyll a (Appendix Dl).

Of all predictors in the  BRT  model for AFDM, percent fine substrata had the highest  relative influence (Table
4.8,  Appendix Dl). Percent canopy cover  also exhibited  a fairly  high relative influence, however the
relationship between this co-factor and AFDM was  not monotonic; rather, very high (in particular) and very
low, canopy  cover values were the two states that corresponded to predictions  of higher AFDM  values
(Figure 4.7).

Percent canopy cover was the most important predictor of percent cover of macroalgae (PCT_MAP; Table
4.8), accounting for over 13% relative influence. Unlike the case with  AFDM,  percent canopy cover had
gradual, monotonic relationship with macroalgal cover, in which macroalgal percent cover decreased steadily
with increasing canopy cover (except for levels >80% canopy cover, at which point PCT_MAP dropped  off
precipitously;  Figure 4.8).  The interaction  between canopy cover and  conductivity was significant  in
predicting macroalgal percent cover: high conductivity (>500 u.S, and especially >2500 u.S) combined with low
canopy cover were conditions favoring high macroalgal percent cover.
                                                110

-------
                                                70
                                      1.0
                       0.2
                 30 0.0
Figure 4.6. Three-dimensional plot (two views) of NOX and mean monthly maximum ambient air
temperature (the mean of the month the sample was collected and the two months prior) from BRT
model for chlorophyll a.
 i
     o
     (N -

_^ ;
/

i i i i i i h —
0 20 40 60 80 100 Q

1 I 1 1 I
2Q 40 60 80 100
                    PCT_FN (11.9%)
XCDENMID (4.1%)
Figure 4.7. Partial dependence plots of percent fine substrata (left) and percent canopy cover (right)
from the BRT model for AFDM. The y-axes correspond to the fitted AFDM variable. The values in
parenthese are the relative influence of the variable indicated on the x-axis on the response variable.
PCT_FN is percent fines; XCDENMID is percent canopy cover.
                                             Ill

-------
Figure 4.8. Three-dimensional plot of percent canopy cover (XCDENMID) and conductivity from a
BRT model for PCT_MAP (percent macroalgal cover). The two predictors exhibited a significant
interaction in their relationship to PCT_MAP.

A substantially different set of environmental co-factors came into play as key predictors for macrophyte
percent cover (PCT_MCP) relative to what was observed for the other biomass variables (Figure 4.5). The
highest-ranked predictor of PCT_MCP (with a relative influence of >17%) was percent sand + fines (Table 4.7),
which had a significant interaction with several other predictors, such as days of accrual (Figure 4.9). Sites
with percent sand  + fines >40% (and especially >80%), with days of  accrual exceeding approximately 100
days, had particularly high macrophyte percent cover.
                                               112

-------
                     20
                                                                       100
                                          200
Figure 4.9. Three-dimensional plot of percent sand + fine substrata and days of accrual from BRT
model for PCT_MCP (percent macrophyte cover).
4.3.4 Results of Bayesian CART Analyses (Objective 4)
Bavesian CART Trees

Bayesian CART analyses were run with  a training set to fit the model, with  independent model validation
using a test set (10% of observations). The Bayesian CART analysis with a full set of classification variables
yielded a relatively high explanatory power for the training set (r2 = 0.84), with only a slightly lower value for
the validation test set (r2 = 0.80). Model fit based on AIC was even better for the Dodds-type DINDIP model
(training r2 = 0.91, test r2 = 0.88; Table D.2). For the Bayesian CART analyses with reduced classification sets,
and within the PSA ecoregion regression trees, the Dodds-type TNTP models outperformed the Dodds DINDIP
models. In both cases, model fit improved significantly and substantially (from test set r2 values of 0.23 or
0.44 to 0.63  or  0.81, respectively) when geographic coordinate predictors were  included as continuous
variables.  For PSA ecoregion DINDIP regression trees, QUAL2K-type models performed slightly worse than the
simpler Dodds-type models (Table D.4). For  the best Dodds-type TNTP model, Julian Day, NH4, and local
urbanization were  included as final classifiers; PSA ecoregions were not. For the best QUAL2K-type DINDIP
model, Julian Day and local urbanization were included as final classifiers.

Overall, based on AIC values, the empirical nutrient region regression trees performed better than the PSA
ecoregion regression trees (Tables D.4, D.5). Again, for Dodds-type models, TNTP models outperformed
DINDIP models,  and model  fits were significantly improved by the  addition of  geographical coordinate
covariates. Some reversals of these trends are apparent in the test set r2 values due to the presence of a few
                                               113

-------
outliers (Table D.5, Figure D.9). Again, the simpler Dodds-type models yielded a better fit than the QUAL2K-
type models and explained about 10% more variation, although even the QUAL2K-type models performed
much better than the original BBST models with the California data set. For the best Dodds-type TNTP model,
latitude, longitude, Julian Day, and NH4 were included as final classifiers, while for the best QUAL2K-type
DINDIP  model, latitude, longitude, Julian Day and local urbanization were included as final classifiers (Table
4.10).

Table 4.10. Results of B-CART analyses based on reduced sets of classification variables. Models used
reduced set of four or five potential classification variables (PSA ecoregion (PSAci) OR Latitude and
Longitude, Julian Day, NH4, and URB21_5K). Training set used is 1. Predictor variables for Dodds-type
models  included TN, TP, TN2, TP2, days accrual, and days accrua!2 (TNTP models) or the same variables
with DIN and DIP substituted for TN and TP (DINDIP models). Predictor variables for the QUAL2K-type
models  also included stream temperature, incident light, and a turbidity x water depth interaction). Models
were also run with or without latitude and longitude as predictors. Model numbers are provided for a
subset for ease of reference in the text. Model fits are assessed based on the Aikake's Information Criteria
(AIC) values and by r2 values associated with plots of predicted versus observed values for loglO chlorophyll
a (mg/m2) for training and test sets.
Independent Lat/ i
Model Model Variable Long
No. Type Type included
Model 1 Dodds TNTP Yes
Model 2 QUAL2K DINDIP Yes
Models Dodds TNTP Yes
Model 4 QUAL2K DINDIP Yes
. Predicted vs.
Number Most 2
ndependent Visited

regression Log Like- Tree Train- Final
variables lihood Size AIC ing Test Classification Variables
8 736.17 7 -1346.33 0.79 0.81 JulDay NH4
11 635.90 4 -1175.79 0.69 0.66 JulDay URB21_5K
8 1249.05 23 -2084.1 0.92 0.57 [~at' JulDay
Long
11 1033.65 18 -1635.3 0.9 0.72 Lat/ JulDay
Long
URB21_5K

NH4
URB21_5K
The Regression Trees generated by the final B-CART analyses are illustrated in Figures 4.10-4.13. Figures 4.10
and  4.11  show  the sequential  splits of the  original training set  into different nodes, along  with  the
classification variable and rule associated with each split. For example, Tree 1 generates five classes of sites
after two are collapsed due to insufficient  size. The first split separates off a small early spring class (n = 8)
with Julian Day of sample less than 124.  Nodes B and C have very  low NH4, with Node B representing spring
values and C representing summer values. Node D and E represent sites with higher NH4 and with low versus
higher localized URB21_5K values (Figure 4.10). Tree 2 describes four classes of sites, a late summer set, a
spring/midsummer non-Reference set, and two Reference sets (low localized URB21_5K development), one
from the spring sampling period and the second from the summer sampling period (Figure 4.11).
                                                114

-------
           JDay<124
           ^,- +--',
                               Jday<166
                                   N = 7
                                                     --___ Jday >166
                  /     N = 2
Figure 4.10. Bayesian regression tree for Model 1. Ovals represent nodes in tree and arrows correspond
to classification rules. Filled-in ovals represent final nodes in tree. Dashed lines and borders are added to
indicate nodes that should be collapsed due to inadequate final node size.
                      JDay<171
     Urban5K< 1.2
    Spring ref
Summer ref
Figure 4.11. Bayesian regression tree for Model 2. Ovals represent nodes in tree and arrows correspond
to classification rules. Filled-in ovals represent final nodes in tree. Dashed lines and borders are added to
indicate nodes that should be collapsed due to inadequate final node size.
                                                115

-------
Structures for trees 3 and 4 were more complex but similar (Figures 4.12, 4.13). Because of the complex tree
structure with a large number of end nodes defined by geographic coordinates, the nodes for Trees 3 and 4
are illustrated in map form, with the legend indicating nodes with classification rules based on nongeographic
variables. The initial split in both trees represented separation of northern from southern sites. In Tree 3,
southern sites were then distinguished on the basis of NH4 levels, and then by a combination of season and
geographic region. Northern nodes were further classified only by geographic region. In Tree 4, the southern
nodes were differentiated by season (spring versus summer, spatial region, and then level of localized urban
development (URB21_5K), while northern nodes were classified by geographic region.
                                    Legend
                                    Tree 3 Nodes
                                         LoNH4-2_3
                                         LoNH4-1
                                         LoNH4-4_5
                                         LoNH4-6_7
                                         HiNH4spr-8_9
                                         HiNH4spr-10
                                         HiNH4smr-11
                                          -12
                                          -13
O
•
O
O
•
O
O
O
O
-14
-15
-16
-17
-18
-19
-20
-21
-22
-23
                                     210
                                           105
                                                             210 Kilometers
Figure 4.12. Location of sampling sites corresponding to nodes in Bayesian CART Model 3. Nodes
not classified according to NH4 level or season (solid circles) were classified solely on the basis of latitude
and longitude. HiNH4 = High NH4, LoNH4 = Low NH4, spr = spring, smr = summer.
                                                116

-------
                                                          Season UrbSK  Node
Figure 4.13. Sampling station locations corresponding to final nodes in Bayesian CART Model 4.
Nodes were classified basd on lat/long coordinates, Julian day (season), and level of NLCD Code 21
urbanization within 5K radius, spr = spring, smr = summer, Lo = Low URB21_5K, Hi = High URB21_5K.
                                            117

-------
Regression Analysis of Bavesian CART Nodes

Table 4.11 presents the results of the stepwise regressions performed on observations from the end nodes of
the Regression Trees (after collapsing nodes of inadequate size). Final regressions were performed using both
training and test set observations combined. Model assumptions for linear regressions (i.e., normality of
residuals,  homogeneity  of variance) were not always  met. These assumptions are not  required for the
original Bayesian CART analyses, but final regression models will need to be  refined. For both Trees 1 and 2
(the Dodds-type and QUAL2K model types without lat/long predictors), days of accrual was retained as the
best predictor of benthic algal biomass for samples collected in early spring, spring with low NH4 values, or
spring with low urban21_5k values (Table 4.11). However, the sign of the regression coefficient was negative,
opposite of that predicted by Dodds models in which biomass is expected to accrue over time following a
spate. For the two smaller nodes this effect is probably due to outliers. For nodes representing samples
collected in mid to late summer either total or dissolved inorganic N were retained in regression models.

For both Trees 3 and 4 (the Dodds-type and QUAL2K model types with lat/long predictors), either latitude or
longitude was retained as the best predictor in regression models for most end nodes. Longitude was always
associated with a positive effect, while latitude effects varied within region. Total N or total P were selected
as the primary explanatory variable in regressions for only a few regions.

Table 4.11. Variables retained in regression analyses to predict benthic biomass (loglO chlorophyll a)
based on Dodds-type models for nodes in B-CART models 1 and 3, and based on QUAL2K-type models for
nodes in B-CART models 2 and 4. Nodes are numbered from left to right in B-CART trees in corresponding
Figures 4.12 and 4.13. Regressions were fit using stepwise regression with a y-intercept (Int). Sign of
regression coefficients is given inside parentheses following parameter. Node  characteristics describe the
classification rules producing each final group. - No regression results due to low class size.
Tree
1
1
1
1
1
1
2
2
2
2
Node
I23
2 24, 24
3
4_5
6
725
I25
225
-,24, 25
424, 25
Size
9
70
125
10
123
236
34
35
191
313
Regression variables
Int
Int
Int
Int
Int
Int
Int
Int
Int
Int
loglOdays (-)
loglOdays (-)



loglOTN (+)
loglOdays (-)

Iogl0shadsolr(+)
loglODIN (+)
adj r2
0.59
0.19



0.11
0.58

0.06
0.14
Node Characteristics
early spring
low NH4 spring
low NH4 summer
high NH4 late spring low
urban21_5K
high NH4 late spring high
urban21_5K
mid to late summer
low urban21_5K, early spring
low urban21_5K, late spring
spring, hi urban21_5K
summer
  Residuals demonstrated heterogeneity of variance
24 Residuals not distributed normally according to Wilk-Shapiro test (p <0.01)

                                                118

-------
Table 4.11 (continued)
Tree
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
4
4
Node
1
2_3
4_5
6_7
8_9
10
11
12
13
14
15
16
17
18
19
20
2125
22
23
-i 24, 25
2
3_4
5
6
7_8_924
10
11
12_1324
14
15_1624'25
17_18
Size
9
9
43
48
21
13
129
38
4
3
5
13
16
15
45
55
55
15
1
58
30
60
2
17
69
99
22
40
36
129
11
Regression variables
Int
Int
Int
Int
Int
Int
Int
Int
Int
Int
Int
Int
Int
Int
Int
Int
Int
Int
-
Int
Int
Int
-
Int

Int
Int
Int
Int
Int
Int


longitude (+)
longitude (+)
loglOTP(-)
logTN(+)

latitude (-)
loglOTN (+)
Iogl0days(+)
latitude (-)
latitude (-)

latitude (+)
latitude (-)
latitude (+)
longitude (+)
loglOTP(+)

latitude (-)
longitude (+)
longitude (+)

latitude(-)
longitude (-)
temperature (+)
latitude (-)
longitude (+)
latitude (-)
longitude (+)
longitude (+)
adj r2


0.7
6
0.24
0.34

0.72
0.74
1
0.76
0.86

0.79
0.71
0.48
0.65
0.23

0.27
0.49
0.47

0.45
0.11
0.07
0.52
0.26
0.82
0.52
0.44
Node Characteristics
lowNH4
lowNH4
lowNH4
lowNH4
high NH4
high NH4
high NH4












spring
spring
spring
summer
summer
summer
summer
summer
summer
summer
summer
summer




spring
spring
summer
















low Urb21_5K
high Urb21_5K
high Urb21_5K
high Urb21_5K
high Urb21_5K
high Urb21_5K
high Urb21_5K
high Urb21_5K
                                               119

-------
4,4
4,4.1 Validation Exercise Shows Considerable Room for Improvement in BBST
The BBST Dodds and QUAL2K models showed a very poor fit when validated against a statewide dataset of
1031 wadeable stream sites in California over 2007-2011. The perceived poor performance of the underlying
BBST  models is understandable  for a variety of  reasons, including difference in climate  and hydrology
between the data set supporting Dodds model development and the California wadeable stream dataset and
differences in the spatial and temporal representativeness of modeled output versus the validation data set.
In addition, algal primary production may be affected by other factors in the California wadeable stream
dataset than those considered in the models.  Finally, precision of observed  benthic algal biomass data, as
currently measured in  California ambient monitoring programs,  is uncertain. Data used  to develop the
original Dodds models were compiled from various sources reflecting different temporal intensity of sampling
and collection or analytical methods, which are not necessarily comparable to the California stream data.

At the time of BBST development, California wadeable streams data were  scarce and thus models  were
optimized for available national or international datasets. Fundamental differences in the factors controlling
primary production between these national and  California wadeable streams is an obvious  reason for poor
model performance.  The California wadeable  dataset is  comprised  largely of sites from a  Mediterranean
climate and perennial to intermittent flow regimes, while  Dodds et al. (1997, 2002) models are derived from
largely temperate, wadeable stream data. The BBST QUAL2K  models (Tetra Tech 2006) were optimized to
Dodds et al. (1997, 2002). In the application of their empirical model to a USGS data set, Dodds et al. (2002)
report best fit at R2 of 0.18, comparable to our findings. Though the Tetra Tech report (2006) suggests that
the equations  proposed by Dodds et  al. (2002) are qualitatively reasonable for predicting  mean  and
maximum potential  growth of benthic algae in California streams in the absence of severe light or scour
limitation, they also report low R2 (~0.20) values for model validation of the RWQCB 6 data.

Another reason for poor model performance is a fundamental inequality in predicted biomass versus what is
measured on both temporal and spatial scales. Conceptually, the BBST models are predicting algal abundance
as spatially and seasonally averaged means or maximum values. A true validation of this model output is
difficult, as large, geographically expansive wadeable stream datasets  rarely have  both good temporal and
spatial (within streams) resolution. The California wadeable stream dataset reflects a one-time sampling of
both explanatory variables and biomass responses,  integrated over 150-m stream reaches, sampled over the
growing season25, across a 1000-mile range of latitude. Thus, these data are not likely to  be representative of
a spatially and temporally averaged "mean", nor maximum, values. Furthermore, it is likely that site-specific
factors acting on the expression of primary producer biomass include ones beyond  those which are typically
considered  in eutrophication models. As such, recommended future work includes time-course sampling in
streams to understand seasonal means and maxima, as well work toward better understanding the potential
role of a wider-array of site-specific factors.

In addition, factors associated with urbanization had strong explanatory power for models' lack of fit and we
observed the poorest model performance  in the Stressed sites. Approximately one-third of our data are from
25 The index period for stream sampling for the validated data used here starts in May for drier parts of the state and
  June or July in colder/wetter parts of the state (depending upon stream flow conditions), and lasts for two to three
  months.
                                               120

-------
Stressed  sites. Urbanization impacts a  stream in numerous ways (Walsh et al. 2005, Booth et al. 2004),
including increased scouring incidents. Stream channelization can also lead to increased water temperatures.
Urban-derived toxicants have the potential to lower the biomass and might also explain the discrepancy in
the model predictions. Application of the BBST may improve if applied on smaller spatial scales, where site-
specific and landscape factors controlling eutrophication are more homogeneous (see Chapter 5). Biggs
(1995) observed significant variation in the benthic algae population in streams  based on land use and
underlying geology. Other studies report stronger relationships (based on R2 values) between the nutrients
and biomass, along with secondary co-factors (such as days of accrual and  watershed area) when applied to a
homogeneous set of sites at a watershed scale (Biggs 2000,  Van Nieuwenhuyse  and Jones, 1996, Dodds
2006). The NNE-BBST performance improved when applied to the 270 Reference sites in  our study (R2 ~0.2)
compared to aggregation of 1031 more heterogeneous sites. Finally, previous work by Fetscher (unpublished)
suggests that, at least in streams supporting macroalgae and relatively high benthic algal biomass, precision
of biomass estimates, based on data as currently collected for California ambient  monitoring programs, is
uncertain.

4.4.2 Inclusion of Landscape and Site-Specific Factors Provide Avenue for       Refinement
Preliminary BRT and Bayesian CART analyses indicate that inclusion of landscape and site-specific factors into
statistical stress-response models appeared to improve model fit over existing BBST Dodds et al. (1997, 2002)
and QUAL2K models. Several landscape- and site-scale explanatory variables were high (relative to nutrients)
in their relative influence in the  variance analysis of the difference between observed and BBST-predicted
biomass, and in the preliminary BRT  models. This finding validates the fundamental NNE approach (i.e., that
site-specific  co-factors that vary across the California landscape can control algal response to nutrients, thus
overriding a simple nutrient limitation on algal abundance). Some  of  these variables, such as a water
temperature, canopy cover,  and solar radiation  are already  included in the BBST QUAL2K, providing
validation that the fundamental factors considered in the TetraTech  (2006) modeling approach are relevant.
Other explanatory variables not  previously available for BBST modeling,  such as ambient  air temperature,
"CODE 21" land use, alkalinity, sedimentary geology, solar radiation, and sediment percent  fines, had a high
level of relative influence in preliminary BRT models, though the importance of these variables varied among
models predicting benthic chlorophyll a, AFDM and macroalgal percent cover (PCT_MAP).

Bayesian CART, a modeling approach that directly incorporates  nutrients and/or mechanistic relationships
into the model, also found improved  fits with inclusion of variables representing geographic position (latitude
and longitude). This suggests that model fit could  be substantially improved by regionalizing coefficients.
Both modeling  approaches suggest  that the strong influence of environmental gradients  associated  with
latitude and longitude  are not well-represented by the  PSA ecoregions.  Ecoregion was  also among the
predictors that exhibited somewhat surprisingly low relative influence  on  the  biomass measures in BRT
models (<5%).  Bayesian CART models incorporating PSA  ecoregion underperformed relative  to those
incorporating geographic position. This makes  intuitive sense, as latitude and longitude  can capture multi-
factor gradients  in temperature, precipitation  regime, slope, and cloudiness/fog. A single ecoregion could
contain both east-  and west-facing slopes,  such that east-west direction would  be a poor predictor  of
monotonic   environmental conditions  within  a  single  PSA ecoregion.  For  example,   elevation  (and
temperature) could both increase and decrease with distance east.
                                                121

-------
Inclusion of explanatory variables that are integrative  over time and space,  in addition to instantaneous
"snap-shots" taken in the field, may help to improve model performance. Examples of this are proxies for air
temperature and available light, developed through GIS  analysis. Elevation and  distance from the coast (e.g.,
cover and fog) produce strong gradients in  air temperature and available light  that are only imperfectly
captured in  the QUAL2K model (which uses  only latitude and sampling month range  to calculate available
solar radiation). Water temperature at time of sampling is typically employed  in the model, yet water (and
air) temperatures show a strong diurnal variation, so  the time of field sampling could  confound model
application for regulatory  purposes. In preliminary BRT models, both antecedent ambient  air  temperature
and stream temperature on the day of sampling exhibited strong correlations with chlorophyll a, the levels of
which  increased  dramatically over a  relatively narrow range  of  antecedent ambient  air  temperature
(approximately 26-28°C). Thus inclusion of landscape-scale monthly averaged air temperature variables may
help to improve the prediction of temperature  influence on primary production over use of a one-time
measure of stream water temperature alone.

Other explanatory variables, such as "Code 21" land use and Wl_Hall, indicators of development and riparian
disturbance, respectively, are not typically included in mechanistic models of eutrophication (e.g., QUAL2k),
yet they were  identified as having a high relative influence in both BRT and  B-CART models.  These represent
indirect effects and may actually represent a suite of stressors.  A number of recent studies  have  reported a
positive correlation between increase in urbanization and benthic biomass (Catford et al. 2007, Walsh  et al.
2005).  Cuffney et al.  (2005) observed that basin-scale land use changes were the most important variables
influencing the benthic response to urbanization. Rather than a single metric of urbanization, it is often the
interaction among multiple impacts of urbanization that  has the most significant  influence on the benthic
algal biomass (Taylor et al. 2004). A number of stressors such as hydromodification effects on hydroperiod
and stream channel morphology and habitat  type, as well as chemical contaminants, such as herbicides and
heavy and trace metals, can affect algal abundance. These factors, and their interactions related  to "Code 21"
land use and Wl_Hall, are difficult to model mechanistically.

Nutrient concentrations were important predictors in variance analysis of the difference between observed
and BBST-predicted biomass, and in BRT models, albeit occupying less prominent roles than other factors.
However,  Bayesian CART models illustrated that inclusion of season when modeling the role of nutrients is
important. Model results showed that total or  dissolved inorganic N (and occasionally total P)  was a better
predictor of benthic algal biomass measured  in the summer than in the spring. This could be because in the
summer,  when  nutrient  levels are  associated  with  baseflow, grab  samples  are more  likely to  be
representative of available nutrients than in the spring.

The Bayesian CART results suggest there are seasonal shifts in controlling factors, with days of accrual being a
better  predictor of benthic algal biomass for spring  samples,  and total  or  dissolved  inorganic N  (and
occasionally total P) being  a better predictor of benthic algal biomass measured in the  summer. Biggs (2000)
reported an improvement in the regression  model from a R2 of 0.40 to 0.74  with the inclusion of accrual
information. The negative  sign of regression  coefficients we observed to be associated with days of accrual
could have resulted either from an inaccurate specification of threshold  discharges associated with scouring
events or from the effect of scouring events on macroinvertebrate grazer populations. Low or intermediate-
level events in spring could be associated with pulses  of nutrient inputs but without sufficient power to
remove existing algal biomass. Predicting site-specific scour based on land-use and historic meteorology data
is  challenging, and it is possible that we cannot currently estimate it using  the available PSA data with

                                                122

-------
sufficient accuracy.  Better information  is needed on  levels  of discharge associated with scouring of algal
biomass in California systems. In addition, numerous other  factors could be at play that are unrelated to
precipitation. For instance, anthropogenically originating controlled releases could be at play, but are difficult
or impossible to model on a large scale. Other factors that are difficult to account for based on the existing
dataset can (like scour) lead to the removal of algae and macrophytes from streams, thus reducing biomass.
These include herbicides and algaecides, but we do not currently have these data on a large scale in order to
include them in our models.

Among biomass types, chlorophyll a was the most responsive to nutrients in the BRT models. The higher
responsiveness of chlorophyll  a than AFDM to stream nutrient concentrations could be due to the fact that
algal chlorophyll a production is by necessity stimulated by stream water-column nutrients, whereas stream
AFDM  can be subsidized by allochthonous material, thus weakening its connection to stream nutrient levels.
However, as noted in Chapter  3, several findings suggested that AFDM may, in general, be a more meaningful
predictor of All responses than chlorophyll a. Thus there  may be value in assessing chlorophyll a and AFDM
jointly  in order to determine nutrient impacts to Alls,  as they represent two important components in the
linkage of nutrient concentrations to Alls: the former being  more directly responsive to  nutrients, and the
latter apparently having more  direct influence on Alls.

In determining what factors belong in predictive models for biomass levels in response to nutrients and other
environmental co-factors, it is  important to consider the possible mechanisms behind observed relationships,
such as those presented earlier in this  section. For instance, percent fine substrata was the top-ranked
"predictor" for AFDM biomass, and yet,  it is  possible that the strong relationship between this predictor and
the response variable is not causative in nature, but rather the result of the fact that the organic component
of fine bed material in a stream is, in  itself, AFDM in the form of fine particulate organic matter (FROM).
Another observation that will  be useful in further model development is the  non-monotonic relationship
between canopy cover and AFDM. At the low  end of the canopy cover gradient, the somewhat elevated
predicted AFDM is likely the result of increased sun exposure supporting instream primary production. But
the high AFDM predicted at  the high  end of the  canopy  cover  gradient cannot be  due to the same
phenomenon; rather it is  most likely the result of allochthonous organic matter input to the stream from the
canopy itself, which, following breakdown by shredders and weathering, would be included in the FROM pool
analyzed as part of AFDM. In both of these cases, anthropogenic nutrient loading would not be responsible
for some fractions of the AFDM, and any predictive models  must take this into account.  It is important to
keep in mind, however, that, while the results of the preliminary BRT modeling presented here offer insights
into nutrient-biomass relationships, a more thorough approach to nutrient-algal  abundance modeling is still
needed in order to better refine and optimize predictive models for wadeable streams.

4.4.3 Summary of Validation     Recommendations for Refining          Stream Nutrient Algal
       Abundance Models
Our analyses indicate that the  existing BBST models could  be improved substantially, and existing data can be
used to pursue refinements. The compiled dataset now includes a variety of explanatory variables that are
available to begin  a more thorough set of analyses. If algal abundance is among numeric endpoints utilized,
then we recommend revising scoping models for wadeable  streams, considering a full range of predictive
statistical models.
                                                123

-------
US Environmental Protection Agency (2009) recommends using a variety of modeling approaches including
regression  (e.g., linear, logistic,  quantile,  and piecewise), change  point analysis, and  structural equation
modeling to explore relationships between nutrients and algal abundance. More than one model categorized
by classes, such as regions within the state,  may be necessary in  order to capture the range of nutrient-
response relationships statewide. More complex mechanistic  models could be considered over the long-term
if the need to offer greater flexibility and  applications to site-specific waterbody assessment are warranted.

Although the Bayesian CART trees incorporating latitude and longitude as  both  classification and predictor
variables were the most accurate predictors of benthic algal biomass, in practice they may be too complex to
be useful to managers. It is likely that these  could  be  simplified by incorporating degree days (cumulative
temperature effect) and distance from the coast as classification variables in place of latitude/longitude. The
significance of temperature in predicting potential peak algal biomass is apparent in  both BRT and Bayesian
CART results. Temperature effects appear to be captured in classifier variables (e.g., season, latitude and
longitude)  but not as a continuous  variable in final linear regressions. This could reflect the temperature
optimum of filamentous green algal taxa  that are responsible for the larger biomass accruals, as illustrated by
the nonlinear interaction plots for NOX and 3-month antecedent air temperature (Figure 4.6).
                                                124

-------
.  .•;  "•-::.' -.-  , •-,-  ; -' •  J
Biggs, B.J.F. 1995. The contribution of disturbance, catchment geology and land use to the habitat
    template of periphyton in stream ecosystems. Freshwater Biology, 33:419-438.
Biggs, B.J.F. 2000. Eutrophication of streams and rivers: Dissolved nutrient-chlorophyll relationships for
    benthic algae. Journal of the North American  Benthological Society,  19(1): 17-31.
Booth, D.B., Karr J.R., Schauman S., Konrad C.P., Morley S.A., Larson M.G., and S.J. Burges. 2004. Reviving
    urban streams: Land use, hydrology, biology, and human behavior. Journal of the American Water
    Resources Association, 40:1351-1364.
Burnham, K.P. and D.R. Anderson. 1998. Model Selection and MultiModel Inference (2nd ed.) Springer,
    New York, NY.
Catford, J.A., Walsh C.J., and J. Beardall. 2007. Catchment urbanization increases benthic microalgal
    biomass in streams under controlled light conditions. Aquatic Sciences, 69(4), 511-522.
Chapra, S. and G. Pelletier. 2003. QUAL2K: A Modeling Framework for Simulation River and Stream Water
    Quality: Documentation and User's Manual. Civil and Environmental Engineering Dept., Tufts
    University, Medford, MA.
Chipman, H.A., George E.I., and R.E. McCulloch. 1998. Bayesian CART model search. Journal of the
    American Statistical Association, 93:935-948.
Chipman, H., George E.I., and R.E. McCulloch. 2002. Bayesian treed models. Machine Learning, 48:299-
    320.
Cuffney, T.F., Zappia H., Giddings E.M.P., and J.F. Coles. 2005. Effects of  urbanization on benthic
    macroinvertebrate assemblages in contrasting environmental settings: Boston, Massachusetts;
    Birmingham, Alabama; and Salt Lake City, Utah. Pages 361-407 in L. R. Brown, R. H. Gray, R. M.
    Hughes, and M. R. Meador, editors. Effects of urbanization on stream ecosystems. American Fisheries
    Society, Symposium 47, Bethesda, Maryland.
Dodds, W.K., Smith V.H., and K. Lohman. 2002. Nitrogen and phosphorus relationships to benthic algal
    biomass in temperate streams. Canadian Journal of Fisheries and Aquatic Sciences, 59: 865-874.
Dodds, W.K., Smith V.H. and K. Lohman. 2006. Nitrogen and phosphorus relationships to benthic algal
    biomass in temperate streams (Erratum). Canadian Journal of Fisheries and Aquatic Sciences,
    63:1190-1191.
Dodds, W.K., Smith V.H., and B. Zander. 1997. Developing nutrient targets to control benthic chlorophyll
    levels in streams: A case study of the Clark Fork River. Water Research, 31(7): 1738-1750.
Dodds, W.K. 2006. Nutrients and the  "Dead Zone": Ecological stoichiometry and depressed dissolved
    oxygen in the northern Gulf of Mexico. Frontiers in Ecology and the  Environment, 4:211-217.
Elith, J., Leathwick J.R. and T. Hastie. 2008. A working guide to boosted regression trees. Journal of
    Animal Ecology, 77: 802-813.
Fetscher A.E., Busse L.B., and P.R. Ode. 2009. Standard Operating Procedures for Collecting Stream Algae
    Samples and Associated Physical  Habitat and Chemical Data for Ambient Bioassessments in
    California. California State Water Resources Control Board Surface Water Ambient Monitoring
    Program (SWAMP) Bioassessment SOP 002. (updated May 2010)
Fetscher, A.E., Stancheva R., Kociolek J.P., Sheath  R.G., Mazor R.D., Stein E.D., Ode P.R.,  and L.B. Busse
    2013. Development and comparison of stream indices of biotic integrity using diatoms vs. non-
    diatom algae vs. a combination. Journal of Applied Phycology, 26:433-450.
                                                125

-------
Freeman, A.M., Lamon III B.C., and C.A. Stow. 2009. Nutrient criteria for lakes, ponds, and reservoirs: A
    Bayesian TREED model approach. Ecological Modelling, 220:630-639.
Liaw, A and M. Wiener. 2002. Classification and Regression by Random Forest. R News 2(3): 18-22.
Taylor, S. L, Roberts S.C., Walsh C.J., and B.E. Hatt. 2004. Catchment urbanization and increased benthic
    algal biomass in streams: Linking mechanisms to management. Freshwater Biology, 49: 835-851.
Tetra Tech. 2006. Technical Approach to Develop Nutrient Numeric Endpoints for California. Tetra Tech,
    Inc. http://rd.tetratech.com/epa/Documents/CA_NNE_July_Final.pdf
U.S. Environmental Protection Agency. 2009. Guidance on the Development, Evaluation, and Application
    of Environmental Models. Office of the Science Advisor, Council for Regulatory Environmental
    Modeling. Washington, DC. EPA/100/K-09/003.
Van Nieuwenhuyse, E.E. and J.R. Jones. 1996, Phosphorus-chlorophyll relationship in temperate streams
    and its variation with stream catchment  size, Canadian Journal of Fisheries and Aquatic Science
    53:99-105.
Walsh C.J, Roy A.M., Feminella J.W., Cottingham P.O., Groffman P.M., and R.P. Morgan II. 2005. The urban
    stream syndrome: Current knowledge and the search for a cure. Journal of the North American
    Benthological Society, 24(3): 706-723.
                                               126

-------
Important Definitions
For those outside the regulatory world, distinction between terms like "criteria," "standards", "objectives,"
and "endpoints" can be confusing. The purpose of this section is to provide definitions of the terms that are
used in this document within the context of California water quality regulations.

Eutrophication: Eutrophication  is defined as the acceleration of the delivery, in situ production of organic
matter,  and accumulation of organic matter (Nixon 1995). One main cause of eutrophication  in estuaries is
nutrient overenrichment  (nitrogen,  phosphorus and  silica).  However,  other  factors influence primary
producer growth and the build-up of nutrient concentrations, and hence modify (or buffer) the response of a
system  to increased nutrient loads  (hereto referred to as co-factors). These co-factors include hydrologic
residence times, mixing characteristics, water temperature, light climate, and grazing pressure.

Indicator: A characteristic of an ecosystem that is related to, or derived from, a measure of biotic or abiotic
variable, that can provide quantitative information on ecological condition, structure and/or function. With
respect  to the water quality objectives, indicators are the ecological  parameters for which narrative  or
numeric objectives are developed.

Numeric Endpoint; Within the context of the NNE framework, numeric endpoints are thresholds that define
the magnitude of  a  response indicator  that is considered  protective  of  ecological health. These numeric
endpoints serve as guidance to  Regional  Boards in translating narrative nutrient or biostimulatory substance
water quality objectives. They are called "numeric endpoints" rather than "numeric objectives" to distinguish
the difference with  respect to  SWRCB  policy. Objectives are promulgated through a  public process and
incorporated into basin plans. Numeric endpoints are guidance that can evolve over time without the need  to
go through a formal standards development process.

Water Quality Criteria: Section 303 of the Clean Water Act gives the States  and authorized Tribes power
to adopt water quality criteria with sufficient coverage of parameters and of adequate stringency to
protect  designated uses. In adopting criteria, States and Tribes may:

    •    Adopt the criteria that US EPA publishes under §304(a) of the Clean Water Act;

    •    Modify the §304(a) criteria to reflect site-specific conditions; or

    •    Adopt criteria based on other scientifically-defensible methods.
The State of California's water criteria are implemented as "water quality objectives," as defined in the Water
Code (of the Porter Cologne Act; for further explanation, see below).
                                                127

-------
States and Tribes typically adopt both numeric and narrative criteria. Numeric criteria are quantitative.
Narrative criteria lack specific numeric targets but define a targeted condition that must be achieved.

Section  303(c)(2)(B) of the Clean Water Act requires States and authorized Tribes to adopt numeric criteria
for priority toxic pollutants for which the Agency has published §304(a) criteria. In addition to narrative and
numeric (chemical-specific) criteria, other types of water quality criteria include:

    •   Biological criteria: a description of the desired biological condition of the aquatic community, for
        example, based on the numbers and kinds of organisms expected to be present in a water body.
    •   Nutrient  criteria:  a  means  to protect  against  nutrient  over-enrichment  and  cultural
        eutrophication.
    •   Sediment criteria: a description of conditions that will avoid adverse effects of contaminated and
        uncontaminated sediments.

Water Quality Objectives; The Water Code (Porter-Cologne Act) provides that each Regional Water Quality
Control  Board shall establish water quality objectives for the waters of the state (i.e., ground and  surface
waters)  which, in the Regional Board's judgment, are necessary for the reasonable protection of beneficial
uses and for the prevention of nuisance. The State of California typically adopts both numeric and narrative
objectives.  Numeric objectives are quantitative.  Narrative objectives present  general descriptions of water
quality that must be attained through pollutant control  measures. Narrative objectives  are also often a basis
for the development of numerical objectives.

Water Quality Standards: Water quality standards  are the foundation of the water quality-based  control
program mandated by the Clean Water Act. Water  quality wtandards define  the goals for a waterbody by
designating its uses, setting criteria to protect those uses, and establishing provisions to protect water quality
from pollutants. A water quality standard  consists of three basic elements:

    1.   Designated uses of the water body (e.g., recreation, water supply, aquatic life, agriculture;
        Table A.I),

    2.   Water quality criteria to protect designated uses (numeric pollutant concentrations and
        narrative requirements), and

    3.   Antidegradation policy to maintain and protect  existing uses and high quality waters.
                                                128

-------
Table A.I. Definition of beneficial uses applicable to freshwater habitat.
 Cold Freshwater Habitat (COLD) - Uses of water that support cold water ecosystems including,
 but not limited to, preservation or enhancement of aquatic habitats, vegetation, fish or wildlife,
 including invertebrates.
 Commercial and Sport Fishing (COMM) - Uses of water for commercial or recreational
 collection of fish, shellfish, or other organisms including, but not limited to, uses involving
 organisms intended for human consumption or bait purposes.
 Contact Water Recreation (REC-1) - Uses of water for recreational activities involving body
 contact with water, where ingestion of water is reasonably possible. These uses include, but are
 not limited to, swimming, wading, water-skiing, skin and SCUBA diving, surfing, white water
 activities, fishing, or use of natural hot springs.
 Migration of Aquatic Organisms (MIGR) - Uses of water that support habitats necessary for
 migration, acclimatization between fresh and salt water, or other temporary activities by
 aquatic organisms, such as anadromous fish
 Non-contact Water Recreation (REC-2) - Uses of water for recreational activities involving
 proximity to water, but not normally involving body contact with water, where ingestion of
 water is reasonably possible. These uses include, but are not limited to, picnicking, sunbathing,
 hiking, beachcombing, camping, boating, tide pool and marine life study, hunting, sightseeing,
 or aesthetic  enjoyment in conjunction with the above activities.
 Rare, Threatened, or Endangered Species (RARE) - Uses of water that support habitats
 necessary, at least in  part, for the survival and successful maintenance of plant or animal
 species established under state or federal law as rare, threatened or endangered.
 Spawning, Reproduction, and/or Early Development (SPWN) - Uses of water that support high
 quality aquatic habitats suitable for reproduction and early development of fish. This use is
 applicable only for the protection of anadromous fish.
 Warm Freshwater Habitat (WARM) - Uses of water that support warm water ecosystems
 including, but not limited to, preservation or enhancement of aquatic habitats, vegetation, fish
 or wildlife, including invertebrates.
 Wildlife Habitat (WILD) - Uses of water that support terrestrial ecosystems including, but not
 limited to, preservation and enhancement of terrestrial habitats, vegetation, wildlife (e.g.,
 mammals, birds, reptiles, amphibians, invertebrates), or wildlife water and food sources.
                                                129

-------
Appendix B. Graphics and Tables Supporting Analyses of
   Reference and Ambient Concentrations of Stream
   Eutrophication Indicator

B.I. Histograms of Biomass and Algal/Macrophyte Cover Data

Figure B.I Histograms of biomass and algal/macrophyte cover data, all California probability
data combined. Y-axes indicate number of sites (N).
350-

300-

250-

^200-
c
§150-
100-
50-
0-






















mrw^


300-

250-

200-
-t—l

0150-
0
100-
50-
o-



















IL
rn-r>^__

         200 400 600 800 1000 1200 1400 1600
        benthic chlorophyll a (mg rrf2)
     100
                                                  200
                                                       300
                                                           400
500
-2\
benthic ash-free dry mass (g m )
                                   130

-------
Figure B.I (continued)
     200-
     150-
   C
   2 100-
      50-
       0-
                                                        200-
                                                         150-
                                                       8
                             "i	1	1	1	1	r
                                                         100-
                                                         50-
           0  10  20  30 40  50  60  70 80  90  100
        i Presence of Macroalgae (Attached and/or Unattached)
 0   10 20  30  40  50 60  70  80  90 100
     % Presence of Macroalgae (Attached)
     400-
     300-
    !200-
     100-
           0  10  20  30  40  50  60 70  80  90  100
              % Presence of Macroalgae (Unattached)
    150-
    100-
     50-
                        I   f    I    I   I    I
                                                       300-
                                                       250-
                                                       200-
                                                      '150-
                                                       100-
                                                        50-
                                                         o-
         0   10 20  30  40 50  60  70 80  90 100
                % Presence of 'Nuisance Algae'
0  10  20  30  40  50  60  70  80  90  100
        % Presence of Macrophytes
                                                          131

-------
Figure B.I (continued)
    400-
    300-
   S200-
     100-
      o-
                                                            100-
                                                             75-
                                                             50-
                                                             25-
          0   10  20  30  40  50  60  70  80  90 100
             % Presence of Thick Microalgae (1mm+)
0  10  20  30  40  50  60  70  80 90  100
         % Presence of Microalgae
     450-
     400-
     350-
     300-
   7=250-
   0200-
     150-
     100-
      50-
           i	1	1	1	1	1	r
                                                            400-
                                                            300-
                                                            100-
                                                              0-
           01   2   3456789   10
                Mean Microalgae Thickness (mm)
  0       5        10      15       20
Mean Microalgae Thickness (mm,) where Microalgae Present
                                                          132

-------
Figure B.2 Boxplots of Biomass, Ash-Free Dry Mass, and Macroalgal Percent Cover.


Boxplots (with "jitter" data points) of biomass, ash-free dry mass, and macroalgal percent cover, for all
statewide data combined (i.e., probability plus target sites), stratified by site disturbance class.
   1500-
   1400-
<
-------
Figure B.2 (continued)
   J100-
    75-
    50-
    25-
     0-
100-
75-
50-

25-

0-





/j .

\





<•.'.;•
:,,.:.




. )
1

: ;'
'»!' •
fc W Jj
•
• •
- 1' •
• •
•

' v- ' ' ' •

r.*.^_--S_ JTES^rSr Si.-lL
                                                                                         100-
                                                                                          75-
                                                                                       S

                                                                                       a
                                                                                       9
                                                                                           0
           Stressed    Intermediate   Reference
                                                      Stressed    Intermediate   Reference
                                                                                                Stressed    Intermediate   Reference
100-


g
|= 75-
0
c
03
in
? 50-
o
o
i

rT 25 ~

•s^


0



















' !



* , x
' " •'.'





1
"
* I . '



V

"^










-,.-,


















i
«















i








. .


























<
'. J
1










......





1 '
!•











• "<_•;


















. .
*
.,1 ' -
.75- ,
I * '.";
. :i ••• ;
CO
Eso- .-• •'
o •
o ' »
C •.'••'
0)
in
3? •'-.'• '. ' ' • '

^25- ; •
3^ (' . ' • — 7" — 71 'J-* •
. '.* I* . . " ' ,
...."•_ •:•. . •.•• ;
".+*,.*••""*"*,. , :. .-.V.
f 	 ' 	 *°


1
»
#_


: .-' 1
. • • . 1


_ ~ _


i / • .> .« • •
I'.v--.1 ]• :? •/."•''•.• -4

                Stressed      Intermediate     Reference
Stressed       Intermediate     Reference
                                                              134

-------
Figure B.2 (continued)
  c-75~

  OJ
    50-
    25-
     o-
            Stressed     Intermediate    Reference

                  Site Reference Status
    Stressed     Intermediate    Reference

         Site Reference Status
100-
90-
80-
CD
& 70-
•2 60-
"o 50-
8
£ 40-
w
CD
Q.30-
^p
20-
10-
o-














.;•. :


•. .









.-. :.'•
"•.'.-••












'-.-:Ji-















,-•:••

••'. •'
' • •
• A !






» '
: •;'.'
.- •-.-.-•
. _ •»
.•^7
.. •



/. ".

i




:'













'
*•* .-"
* rt'**
*,,

'




*I




" .? _.

*. **


.





;
















             Stressed     Intermediate    Reference

                   Site Reference Status
                                                         14-
                                                        ,



                                                        "
                                                       -21  6
                                                       ra  u
                                                       e
                                                       CO
                                                       o>
                                                          4-
                                                          2-
                                                          0-
   I             i              l
Stressed     Intermediate    Reference

      Site Reference Status
                                                       135

-------
Figure B.3. Cumulative distribution functions of biomass, ash-free dry mass, and macroalgal percent
cover, by region, for all probability sites. Shaded areas delineate 95% confidence intervals. Dashed lines
                   -th
correspond to the 75  percentile of values for the indicator in question among Reference sites statewide.
 100
  90
  80
  70
  60
  50
  40
  30
  20
[ 10
!>  °
i
J100-
i
*  90
  80
  70
  60
  50
  40
  30
  20
  10
   0
                   Chaparral
                  North Coast
                                      100-
                                       90
                                       80-
                                       70-
                                       60-
                                       50-
                                       40
                                       30-
                                       20-
                                       10-
                                       o-
                                      100-
                                      90-
                                      80
                                      70-
                                      60-
                                      50-
                                      40-
                                      30
                                      20-
                                      10-
                                       0-
Central Valley
                                                   South Coast
                    100
                     90
                     80
                     70
                     60
                     50
                     40
                     30
                     20
                     10
                      0

                    100
                     90
                     80
                     70
                     60
                     50
                     40
                     30
                     20
                     10
                      0
                                                                                    Deserts-Mod oc
                                                                                    Sierra Nevada
        0   50   100  150  200  250  300
                                              50  100  150  200 250  300
                                             benthic chlorophyll o (mg/m^J
                                                                                50   100  150 200  250 300
                                                   136

-------
Figure B.3 (continued)
100-
90-
80-
70-
60-
50-
40-
30-
20-
S" 10-
f, °-
Q)
§100-
ffi
S 90-
80-
70-
60-
50-

20-
10-
o-
Chaparral


f

























< 	










North Coast


f


































100
90
80
70
60
50
40
30
20
10
0
100
90
80
70
60
50
40
30
20
10
0
         0     40    80   120   160  200
                                                 40    80   120   160
                                                  benthic AFDM (g/m2)
                                                                      200
                                                                                   40    80   120   160   200
                                           0  10 20 30 40 50 60 70 80 90100
                                                macroalgal percent cover
                                                                                 10 20 30 40 50 60 70 80 90100
                                                        137

-------
B.2. Chlorophyll a Distributions within South Coast
To investigate possible differences in chlorophyll a distributions within the PSAS South Coast ecoregion, we
conducted a set of analyses complementary to  that which is presented in the main body of the report, in
which this ecoregion was further divided into "xeric" and "mountain" zones. This subdivision was based on
the Level III classification scheme of Omernik (1987).  Multiple "reference" sites were sampled for chlorophyll
a within both regions (Table B.I), however they were nearly three times as abundant in the mountain zone as
in the xeric zone.
Table B.I. Number of sites within each Level
by site disturbance class.
ecoregion (Omernik, 1987) in the South Coast,
Ecoregion
South Coast Mountain
South Coast Xeric
Reference
27
11
Intermediate
33
80
Stressed
1
144
For each of the 3 NNE endpoints for chlorophyll a, higher proportions of stream length exceeded endpoints
within the xeric ecoregion than in the mountain ecoregion (Figures B.4). The same tendency was observed
within each site disturbance class (where data were available; Figure B.5).
                                                                    Subpopulation
                                                                       South Coast Mountain
                                                                    — South Coast Xeric
                 0     50    100    150   200   250    300   350
                          benthic chlorophyll a (mg/m2)

Figure B.4. CDFs for benthic chlorophyll o, for the "xeric" and "mountain" Level III ecoregions (Omernik
1987) within the South Coast. The graph shows the estimated probability distributions of chlorophyll a
relative to the cumulative proportion of stream length. The dashed grey line on the graph denotes the
75th percentile of chlorophyll a values among Reference sites in the South Coast (24.4 mg/m2).
Highlighted areas delineate the 95% confidence intervals for each estimate.
                                               138

-------
           -1100
               90-
               80-
O)
o
o
V
£  70-
$

I  60

ro
£  50-
                        South Coast Mountain
                                                South Coast Xeric
                                                                  I
                     Stressed  Intermediate Reference     Stressed  Intermediate  Reference
                                        Site Reference Status
Figure B.5. Within-ecoregion estimated percent of stream kilometers lower than the lowest proposed
NNE endpoint for chlorophyll o (100 mg m2), by site disturbance class. Bars indicate 95% confidence
intervals. Note that y-axis scale begins at 50% mark. Due to insufficient sample size, no estimate is
available for the "Stressed" site disturbance class within the South Coast Mountain ecoregion (Omernik
JM. 1987. Ecoregions of the conterminous United States. Map [scale 1:7,500,000]. Annals of the
Association of American Geographers 77:118-125).
                                               139

-------
140

-------
Appendix C. Graphics and Tables Supporting Analyses of
Thresholds of Adverse Effects of Primary Producer Biomass
and Nutrient  on Wadeable Stream Aquatic Life

C.I. Sample output from TITAN and SiZer analyses.
   g
   i

   8
             easily discerned maximum
             sum(i) scores (also similar
               fordecreasersand
                  mcreasersi
plots of summed z scores
for
"decreaser" taxa (•) and
"increaser" taxa ;*j for the
candidate change points
along the gradient
                       *
          2    4    6    6   10  12

              gradient of interest
                                 1.;
                              cumulative frequency distributions of change
                               points among 500 bootstrap replicates for
                               sum(z-) (black) and sum(z+) (red, dashed)
                                            LJ
                                            Q
                                            CN
M
?S
                                                    no clear maximum
                                                     sum(z) score for
                                                      decreasers or
                                                       increasers
                                                                 *
                                                              o  5
                                                                 cr
                                                                 1
                               0   10   20  30  40   50  60  70   80

                                        gradient of interest
Figure C.I. Examples of plots of TITAN sum(z) scores for taxa that decrease in frequency along the
gradient of interest (in black) and those that increase (in red). The graph on the left shows an example
of a clear community-level change point, or threshold. The graph on the right indicates no clear
community-level change point associated with the gradient in question.
                                          141

-------
taxon
r
Nwntl
'•'- ••! '_ .
F sir -•
r,, i M
•.,-,|-,f.'
• IIF, ,
ri'iln-i
MI*»'J
.
• •( .1 1 <
' .-.i !
! 1 it .> •
I. ...l
1 :;:: •
as
Pum«
.-, '," 'ii

— •
-»•
*

codes


	 •—


-r. . . : "*
	 	 .. jt .-
r

\.

•• 	 VJ*"
:£:$£•,




....
....



'








Taxon-specific change
points, al gned in order
of increasing values, for
"decreaser" taxa (•) and
"increaser" taxa (•}
(symbols are sized in
proportion to z-scores)
i
"ij«i
. lanh
.Uti
r r i ;
'. -f
\-vmn
h,l
••, II
.i1 '
i ,|,
Lurm ii
1 i
1 i. .
l i ,i
' . |. 1
1 1 il •:•
'. ,j ,.-,
               50   100   150  200   250  300
                    gradient of interest
                                                          5th to 95& percentiles
                                                          in change points from
                                                          bootstrap restates
                                                                                            £ ,
0   5  10  15 20  25  30  35 40
        gradient of interest
Figure C.2. Examples of plots of TITAN change points for individual taxa. The graph on the left shows an
example of a well-supported, community-level change point, in that several of the "decreaser" taxa are
aligned at essentially the same point along the gradient (see shaded box), and the 5th/95th percentile
ranges from the bootstrap replicates for the estimated changepoints are narrow for most of these taxa.
The graph on the right provides an example of the opposite case (see shaded box), in which no
community-level change point is clear and 5th/95th percentile ranges are generally broad.
                                                142

-------
                     red areas: the 1st
                       derivative is
                   significantly negative
     blue areas: the 1st derivative is significantly
   positive (in this case, this signal does not carry
         through to broader bandwidths)
                                                                             the distance
                                                                             between the
                                                                              white lines
                                                                              provides a
                                                                                visual
                                                                             indication of
                                                                                 the
                                                                              bandwidth
                                                                              at a given
                                                                              point along
                                                                              the y-axis;
                                                                              the lower
                                                                              the y value,
                                                                             the narrower
                                                                                 the
                                                                              bandwidth
                                            200           300
                                       gradient of interest
                           400
                   purple areas: the 1st
                   derivative is neither
                   significantly positive
                       nor negative
 grey areas: data density is insufficient at these
  gradient-bandwidth-combinations (e.g., high
gradient value/narrow bandwidth, in this case) to
     determine whether the 1st derivative is
        significantly positive or negative
Figure C.3 Example of a SiZer map. This map indicates that there is an overall downward trend  in the
relationship between the response variable and the gradient, as evidenced by the red color spanning the
entire length of the gradient at the top portion of the map (where the bandwidth is broadest). Moving
downward along the  y-axis, which represents the bandwidths (i.e., as the bandwidths for assessing the
first derivative of the locally fitted polynomial relating response to gradient  become narrower), red
transitions to purple,  for most of the gradient, indicating that as bandwidth narrows, the local derivative
(i.e., slope) associated with most points along the gradient is no longer significantly positive (or negative).
However, at one point along the gradient (i.e., at ~40, as indicated  at the yellow arrow), a significant
negative first derivative continues to be evident by virtue of the red coloration, even at relatively narrow
bandwidths (i.e., extending far downward along the  y-axis), providing  compelling evidence for a well-
supported, steep, negative slope in that narrow region of the gradient. Such a signature characteristically
immediately precedes a threshold for this type of gradient-response relationship.
                                                  143

-------
Table C.I. TITAN change point values for BMI and diatom taxa ("pure" and "reliable")
Also shown is the order (increasing) of diatom taxa in terms of their change points, within the
"decreaser" and "increaser" groups, for the AFDM and TP gradients. This information is
supplemental to Figure 3.10, where the taxon codes are not legible due to the number of taxa
involved. Number of "decreaser" diatom taxa is 65 for AFDM and 68 forTP. Number of
"increaser" diatom taxa is 100 for AFDM and 98 for TP.
Table C.I
Direction
Of
Response Assemblage Taxon
Acari
Acentrella
Agapetus
Ambrysus
Ameletus
Amiocentrus
Ampumixis
Anagapetus
Antocha
Apatania
decrease BMI Arctopsyche
Atherix
Atrichopogon
Attenella
Baetis
Berosus
Bezzia, Palpomyia
Blephariceridae
Brachycentrus
Caenis
Calineuria
Caloparyphus,
Euparyphus
Capniidae
Caudatella
Centroptilum
Ceratopsyche,
Hydro psyche
decrease BMI
Chelifera,
Metachela
Cheumatopsyche
Chimarra
Cinygma
Cinygmula
Chloro
phyll a
(mg/m2)
48.59
48.70
14.05
-
6.27
22.04
18.31
4.09
46.94
12.98
4.20
2.47
5.35
10.50
11.12
-
84.70
-
5.39
-
20.55
-
15.11
4.23
77.87
49.89
3.88
-
-
8.00
13.37
AFDM
Change
Point

AFDM Order TN
(g/m2) (diatoms) PCT_MAP PCT_MCP (mg/L)
14.42
18.30
5.97
-
6.20
9.44
6.03
-
8.86
3.97
6.01
4.54
-
6.01
6.33
-
-
3.23
1.87
-
6.50
-
9.77
6.01
18.72
11.77
12.25
-
-
9.73
6.47
41.37 0.00 0.51
0.22
19.00 27.50 0.16
-
5.00 0.00 0.22
9.00 0.23
7.00 26.50 0.07
4.00 0.12
45.32 27.50 0.17
3.40 0.00 0.16
45.32 2.00 0.14
0.21
36.60 - 0.50
8.07 - 0.02
16.17 19.02 0.19
.
0.50
6.00 - 0.01
0.02
-
8.00 1.00 0.20
-
0.17
20.57 4.00 0.13
0.28
37.07 - 0.54
4.38 0.00 0.27
-
0.01
3.96 - 0.10
3.87 0.48 0.15

TP
Change
Point
TP Order
(mg/L) (diatoms)
0.02
0.04
0.06
0.00
0.02
-
0.06
-
0.05
0.02
0.03
0.02
-
0.06
0.03
0.01
0.07
0.02
-
0.13
0.02
0.01
0.04
0.03
0.04
0.14
0.13
-
0.08
0.08
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
                                           144

-------
Table C.I
Direction
Of
Response Assemblage Taxon
Cleptelmis
Clinocera
Cloeodes
Cordulegaster
Cultus
Despaxia
Deuterophlebia
Diamesinae
Dicosmoecus
Dicranota
Diphetor
decrease BMI
Dixa
Dolophilodes
Doroneuria
Drunella
Ecclisomyia
Ecdyonurus
Epeorus
Ephemerella
Eubrianax
Frisonia
Glossosoma
Glutops
Gumaga
Helico psyche
Helodon, Prosimulium
Hesperoconopa
Hesperoperla
Heterlimnius
Heteroplectron
Hexatoma
Hydraena
Ironodes
Isoperla
decrease BMI
Juga
Lara
Lepidostoma
Limnophila
Malenka
Manila
Chloro
phyll a
(mg/m2)
8.09
6.35
12.81
9.60
4.99
3.86
6.23
17.40
19.08
17.37
4.33
15.11
9.27
12.96
21.15
10.80
9.78
26.07
6.35
11.57
17.47
18.36
-
4.93
-
46.60
17.15
12.98
21.93
18.73
6.06
48.37
6.27
20.79
-
18.36
-
AFDM
Change
Point
AFDM Order
(g/m2) (diatoms) PCT_MAP
1.07
8.62
14.83
5.79
3.17
18.72
4.60
9.04
9.73
4.32
6.01
6.01
7.42
4.11
6.30
3.78
9.49
4.76
6.17
6.50
9.70
-
5.24
-
10.24
-
5.86
6.45
-
5.34
; ;
7.36
9.49
1.80
6.02
-
-
6.33
-
19.28
17.00
7.31
31.00
6.00
-
0.00
16.17
0.00
0.00
5.03
20.10
8.00
2.00
8.44
4.76
41.00
-
1.00
-
33.50
4.00
2.01
2.09
0.00
2.01
;
-
18.55
-
10.00
-
PCT_MCP
-
-
-
2.00
-
0.00
0.00
-
4.00
2.00
-
0.00
2.00
1.00
5.71
1.00
27.50
-
4.00
-
2.00
-
6.67
-
5.36
0.00
-
13.00
;
-
0.00
0.00
12.00
-
TN
(mg/L)
0.17
0.11
0.27
0.06
0.12
0.10
0.17
0.03
0.22
0.24
0.17
0.10
0.21
0.13
0.29
0.17
0.30
0.30
0.30
0.21
0.16
0.31
0.29
0.13
-
0.06
0.30
0.17
0.28
0.60
0.11
0.31
0.29
0.15
0.32
0.17
0.60
0.25
TP
Change
Point
TP Order
(mg/L) (diatoms)
0.02
0.02
0.02
0.06
0.03
0.08
0.06
0.00
0.06
0.02
0.00
0.02
0.02
0.01
0.08
0.01
0.02
0.04
0.02
0.09
-
0.02
0.02
0.07
0.02
0.02
0.03
0.02
0.04
-
0.05
0.04
0.03
0.02
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
;
-
-
-
-
.
Maruina
                              4.91
                                                          14.00
                                                                   0.17
                                                                           0.01
                                145

-------
Table C.I
Direction
Of
Response Assemblage Taxon
Matriella, Serratella
Micrasema
Microcylloepus
Moselia
Mystacides
Narpus
Neohermes
Neophylax
Neoplasta
Neotrichia
Nothotrichia
Ochrotrichia
Octogomphus,
Specularis
decrease BMI
Oecetis
Ophiogomphus
Optioservus
Ordobrevia
Oreodytes
Oreogeton
Orohermes
Oroperla
Orthodadiinae
Paraleptophlebia
Paraleuctra
Parapsyche
Parthina
Pedomoecus
Pericoma,
Telmatoscopus
Perlinodes
Polycentropus
decrease BMI Probezzia
Procloeon
Prodiamesinae
Protoptila
Psephenus
Psychoglypha
Pteronarcys
Ptychopteridae
Rhithrogena
Rhizelmis
Chloro
phyll a
(mg/m2)
10.13
46.22
-
4.15
-
20.99
-
12.96
8.09
12.96
-
-
39.16
-
-
20.55
17.37
12.96
3.04
15.68
-
-
19.96
-
9.34
17.62
2.36
-
17.31
17.37
-
-
-
-
-
-
15.98
-
6.93
-
AFDM
Change
Point
AFDM Order
(g/m2) (diatoms) PCT_MAP
4.25
7.29
-
7.36
-
12.50
-
5.97
5.19
3.10
-
36.48
1.37
-
-
12.46
5.97
7.59
1.31
4.50
4.09
-
8.95
-
5.55
7.14
1.50
-
6.97
6.33
-
-
-
-
-
-
6.63
-
7.07
-
41.00
2.00
-
2.00
-
18.05
-
12.00
18.00
-
-
-
7.62
-
-
38.55
3.00
15.50
-
0.00
-
-
19.00
-
7.00
7.31
13.50
30.50
8.00
16.17
6.00
-
-
-
-
7.00
26.33
7.31
5.00
-
PCT_MCP
-
-
-
-
-
7.31
-
-
-
-
-
-
-
-
-
7.31
0.00
-
0.00
-
-
2.00
0.00
-
7.00
-
-
-
0.00
0.00
-
-
-
-
-
-
0.00
-
0.00
-
TN
(mg/L)
0.25
0.30
-
0.12
0.29
0.30
0.08
0.20
-
0.02
0.11
0.03
0.13
0.49
0.09
0.31
0.30
0.10
0.09
0.13
0.12
-
0.36
0.08
0.29
0.26
0.10
-
0.21
0.23
0.03
0.44
0.05
0.22
0.29
0.30
0.31
0.05
0.11
0.09
TP
Change
Point
TP Order
(mg/L) (diatoms)
0.04
0.26
0.04
0.03
0.09
0.07
-
0.02
-
0.02
0.01
0.01
0.09
0.08
0.00
0.09
0.02
0.02
0.00
0.00
0.02
-
0.09
-
0.03
-
0.03
-
0.02
0.01
-
-
-
0.04
0.02
-
0.01
-
0.07
0.00
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
146

-------
Table C.I
Direction
Of
Response










decrease





decrease











Assemblage Taxon
Rhyacophila
Rickera
Sialis
Sierraperla
Simulium
Skwala
Sphaeriidae
Stictotarsus
Stilobezzia
Suwallia
Sweltsa
Tab anus,
Atylotus
BMI Timpanoga
Tinodes
Visoka
Wiedemannia
Wormaldia
Yoraperla
Yphria
Zaitzevia
Zapada
Diatom Achnanthidium
biasolettianum
Achnanthidium
deflexum
Achnanthidium
minutissimum
Achnanthidium
minutissimum var
gracillima
Achnanthidium sp 1
SWAMP KB
Adlafia bryophila
Adlafia minuscula
Amphipleura pellucida
Amphipleura sp 1
SCCWRPJPK
Amphipleura sp A
SWAMP JPK
Aulacoseira alpigena
Aulacoseia ambigua
Aulacoseira crenulata
Chloro
phyll a
(mg/m2)
19.96
3.88
23.14
-
7.46
12.58
-
-
10.55
4.86
19.96
-
10.28
-
5.79
-
-
5.80
4.05
20.55
9.20
-
18.72
26.65
2.64
3.85
-
12.41
-
-
-
3.83
4.02
8.89
AFDM
(g/m2)
6.03
-
8.95
-
-
4.23
-
-
2.21
1.33
7.25
5.03
5.88
-
2.31
1.37
-
4.97
2.90
12.39
5.19
-
4.28
12.83
1.28
2.58
-
18.52
-
-
9.24
3.40
-
_
AFDM
Change
Point
Order TN
(diatoms) PCT_MAP PCT_MCP (mg/L)
19.28 6.00 0.17
-
35.27 - 0.26
0.05
-
0.12
3.00
0.62
0.20
1.00 0.01
5.00 0.00 0.15
-
0.02
0.50
0.00 - 0.09
0.06
0.22
4.00 7.00 0.19
0.08
45.00 0.00 0.15
2.00 18.18 0.10
-
22 16.00 - 0.67
48 9.00 8.57 0.56
3 - 2.93 0.02
11 - - 0.05
0.00 - 0.05
56 - - 0.04
5.71 0.45
0.13
37
16 - - 0.12
-
16.00 - 0.04
TP
(mg/L)
0.02
-
-
-
-
0.00
-
0.03
0.03
0.01
0.02
0.02
0.02
0.01
0.06
0.05
0.03
0.02
0.02
0.09
0.06
0.01
0.02
0.07
-
0.01
0.04
-
0.04
0.01
-
-
-
_
TP
Change
Point
Order
(diatoms)
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
16
23
59
-
9
41
-
48
17
-
-
-
_
147

-------
Table C.I
Direction
Of
Response










decrease













Chloro
phyll a
Assemblage Taxon (mg/m2)
Aulacoseira distans 6.08
Aulacoseira italica 30.51
Aulacoseira subarctica 1.87
Brachysim vitrea
Colonels bacillum 76.75
Cocconeis disculus
Cocconeis placentula
var euglypta
Cocconeis placentula
var lineata
Cymbella affinis 0.69
Cymbella cistula 37.64
Cymbella tumida
Cymbella turgidula
Cymbopleura 13 gl
naviculiformis
Diatom Diatoma hiemale
Diatoma mesodon 3.82
Diatoma moniliforme
Diatoma tenuis 35.73
Diatoma vulgaris
Diatoma vulgaris var
linearis
Didymosphenia
geminata
Diploneis oblongella
Encyonema elginense 2.60
Encyonema minutum 5.29
Encyonema muelleri
Encyonema silesiacum 0.90
Encyonopsis
falaisensis
Encyonopsis
microcephala
Epithemia adnata
Epithemia sorex
Epithemia turgida
AFDM
(g/m2)
9.69
-
3.95
-
-
-
-
2.65
1.28
12.25
12.93
-
-
-
13.15
-

-
-
9.10
-
3.42
17.96
-
-
-
-
34.78
31.74
17.15
AFDM
Change
Point
Order
(diatoms) PCT_MAP PCT_MCP
39
.
20
-
-
4.00
25.00
12
2 - 0.00
43
49
-
-
.
50 2.00
-
-
-
0.00
35
-
17
54 2.00 19.00
-
-
-
-
64
61
53
TN
(mg/L)
0.03
1.33
-
0.03
1.03
0.37
0.09
-
0.67
0.28
0.01
0.22
-
0.02
0.09
0.74
-
0.02
0.21
0.20
0.18
0.22
0.19
0.08
0.21
-
0.45
0.44
0.48
0.05
TP
(mg/L)
-
-
-
0.01
0.05
-
0.14
-
0.01
0.03
-
0.02
-
0.02
0.06
0.02
0.01
0.04
0.00
0.01
0.00
0.01
-
0.01
0.02
0.01
0.02
0.03
0.04
0.04
TP
Change
Point
Order
(diatoms)
-
-
-
12
51
-
67
-
13
37
-
21
-
24
54
22
4
40
2
5
1
15
-
18
31
14
25
34
44
47
decrease    Diatom
Epithemia turgida var
  westermannii
Eunotia bilunaris
Eunotia incisa
Fragilaria capucina
                                               49.40    18.30
                                                2.71
                                                                   55
0.04
0.09

0.01    0.01
                                                                                                                  10
                                                          148

-------
Table C.I
Direction
Of
Response










decrease







decrease



Chloro
phyll a
Assemblage Taxon (mg/m2)
Fragilaria cap uc in a
var gracilis
Fragilaria capucina
0.67
var rumpens
Fragilaria vaucheriae 3.86
Fragilaria vaucheriae
var capitellata
Frustulia
amphipleuroides
Frustulia krammeri
Frustulia vulgaris
Geissleria acceptata 6.23
Geissleria ignota
Gomphoneis geitleri
Gomphoneis minuta 49.36
Gomphoneis
olivaceoides
Gomphoneis
Diatom ,.F 48.94
olivaceum
Gomphoneis
pseudokunoi
Gomphoneis rhombica
Gomphonema
acuminatum
Gomphonema
angustatum
Gomphonema
bohemicum
Gomphonema
clavatum
Gomphonema clevei 22.29
Gomphonema 1Q gQ
kobayasii
Gomphonema
micro pus
Gomphonema
Diatom 7.41
minutum
Gomphonema
montanum
Gomphonema
., 13.86
pumilum
Gomphonema sp B
SWAMP JPK
Gomphonema sp C
SWAMP JPK
AFDM
(g/m2)
-
2.45
4.37
-
-
-
-
12.46
4.50
1.31
9.58
-
1.30
1.82
-
-
-
2.22
2.89
4.50
4.47
-
6.80
5.30
27.03
9.14
12.32
AFDM
Change
Point
Order
(diatoms) PCT_MAP PCT_MCP
-
9
23 34.64
-
-
1.90
21.67
47 7.00
26
5
38
-
4
6
-
-
-
8 - 15.24
13
27 0.00
25 9.00 8.57
-
34 0.00 1.00
32
59 13.50 0.00
36 7.62 4.00
45 27.72 0.00
TN
(mg/L)
0.24
0.25
0.31
0.02
0.04
-
-
0.06
0.10
0.03
0.15
0.21
0.03
-
0.04
0.47
-
0.02
-
0.21
0.22
-
0.01
0.05
0.30
0.06
0.07
TP
(mg/L)
0.03
-
0.06
-
-
-
-
-
-
-
0.05
-
0.04
0.02
-
-
0.01
0.03
-
-
0.06
0.09
0.02
0.00
0.11
0.07
0.08
TP
Change
Point
Order
(diatoms)
36
-
55
-
-
-
-
-
-
-
52
-
42
19
-
-
8
33
-
-
53
63
28
3
65
60
61
149

-------
Table C.I
Direction
Of
Response









decrease








decrease







Assemblage Taxon
Gomphonema
stoermeri
Gomphonema
subclavatum
Gomphonema
truncatum
Gomphosphenia sp A
SWAMP EWT
Gomphosphenia sp A
SWAMP JPK
Gomphosphenia sp B
SWAMP EWT
Halamphora normanii
Hannaea arcus
Hantzschia amphioxys
Karayevia clevei
Karayevia oblongella
„. . Karayevia suchlandtii
Diatom
Kolbesia suchlandtii
Luticola mutica
Mastogloia smithii
Meridian circulare
Meridian circulare var
constrictum
Navicula angusta
Navicula
capitatoradiata
Navicula
cryptocephala
Navicula cryptotenella
Navicula
cryptotenelloides
Navicula digitoradiata
Diatom Navicula globulifera
Navicula margalithii
Navicula menisculus
Navicula radiosa
Navicula
rhynchocephala
Navicula tenelloides
Nitzschia dissipata
Nitzschia dissipata var
media
Nitzschia dubia
Chloro
phyll a
(mg/m2)
-
-
-
2.13
2.70
-
-
2.51
5.80
21.90
17.22
7.09
16.18
-
-
3.73
1.20
5.35
63.93
9.79
15.28
5.78
-
-
-
22.84
-
-
-
62.68
23.30
_
AFDM
(g/m2)
-
-
-
5.03
1.88
-
-
4.77
-
33.78
-
4.10
9.77
-
-
0.81
-
32.83
-
28.84
10.01
2.58
-
-
-
20.57
-
-
-
20.12
3.65
_
AFDM
Change
Point
Order
(diatoms) PCT_MAP PCT_MCP
-
-
-
31
7
-
0.00
30 0.00 0.00
11.59
63 39.00
3.00 0.00
21 4.00 1.00
41 - 0.00
24.76
-
1
2.09
62
-
60
42 40.60 0.00
10
-
-
0.00
58 24.00 5.71
.
1.98
0.00
57 32.38
18 21.21 2.86
3.00
TN
(mg/L)
-
-
0.74
0.05
-
0.05
-
0.06
-
0.89
0.13
0.18
0.17
-
-
0.39
0.48
-
1.14
1.16
0.24
0.03
-
0.04
-
0.30
0.74
-
-
0.48
0.25
_
TP
(mg/L)
0.02
0.03
0.04
-
-
-
-
0.03
-
-
-
0.02
0.05
-
0.01
0.04
-
-
-
-
0.06
0.01
0.01
-
-
0.11
0.04
-
-
0.09
-
_
TP
Change
Point
Order
(diatoms)
26
32
43
-
-
-
-
38
-
-
-
30
50
-
6
46
-
-
-
-
58
7
11
-
-
66
39
-
-
64
-
_
150

-------
Table C.I


Direction
Of
Response

























decrease






















AFDM
Change
Chloro Point
phyllo AFDM Order TN TP
Assemblage Taxon (mg/m2) (g/m2) (diatoms) PCT_MAP PCT_MCP (mg/L) (mg/L)
Nitzschia innominata - 4.50 28 - - 0.11
Nitzschia linearis ... 21.67 ...
Nitzschia nana ... i QO ...
Nitzschia paleacea - 0.05
Nitzschia perminuta ----- o.04
Nitzschia recta 9.67 12.32 46 - -
Nitzschia tenuirostris 3.82 -
Pinnularia borealis 2.44 - - - -
Pinnularia
12 13 - - - - - -
microstauron
Placoneis elginensis 10.26 - - - -
Planothidium dubium 34.08 16.10 52 3.00
Planothidium
, .... ----- o.60
haynaldn
Planothidium
15 67
lanceolatum
Planothidium „ „„
4.93 6.40 33 - -
rostratum
Psammothidium
bioretii
Psammothidium „ „ „ _ „ _ „ _
. . 8.18 3.17 15 7.00
margmulatum
Diatom Psammothidium
, ., ----- o.l6 0.02
subatomoides
Reimeria sinuata 18.25 14.90 51 18.07 3.00 1.14 0.18
Reimeria uniseriata 10.04 3.95 19 0.02
Rhoicosphenia
abbreviata
Rhoicosphenia so B
SWAMP EWT 16'18 12'25 44 - - °'09 -
Rhoicosphenia so C
48 94 - - - _ _ _
SWAMP EWT
Rhopalodia gibba - 35.13 65 - - 0.37 0.06
Rossithidium nodosum - - - - - 0.10 -
Rossithidium pusillum 12.63 9.70 40 19.42 2.00 0.29 0.04
Sellaphora bacillum - 0.29 0.08
Sellaphora hustedtii - - - 1.98
Sellaphora seminulum 48.94 - - 12.38
Sellaphora stroemii - - - 3.94
Stauroneis smithii - 9.71 -
Staurosirella
. . . . . o.SO
leptostauron
Surirella angusta - - - 1.00
Synedra delicatissima - - - - 0.03

TP
Change
Point
Order
(diatoms)
-
-
-
-
-
-
-
-

_

-
-






_


45

"

27
68
20


-



56
-
49
62
-
-
-
-

"
-
35
151

-------
Table C.I
Direction
Of
Response Assemblage Taxon
Synedra mazamaensis
Synedra ulna
Tabellaria fenestrata
decrease Diatom
Tabellaria flocculosa
increase BMI Agabus
Ambrysus
Argia
Caenis
Callibaetis
Caloparyphus,
Euparyphus
Calopterygidae
Ceratopogon
Chironominae
Crangonyx
Culicidae
Culicoides
Dasyhelea
Dixella
Dolichopodidae
Dubiraphia
Enochrus
Ephydridae
Fallceon
Ferrissia
Gammarus
Glossiphoniidae
Gymulus
Hedriodiscus,
Odontomyia
Helisoma
Hemerodromia
Hyalella
increase BMI
Hydra
Hydrobiidae
Hydroptila
Ischnura
Laccobius
Lymnaea
Menetus
Muscidae
Nectopsyche
Chloro
phyll a
(mg/m2)
-
-
4.01
2.58
-
3.58
-
-
8.18
22.29
18.41
-
5.63
-
-
-
18.25
-
155.68
-
-
-
22.29
-
-
-
15.37
-
5.02
24.12
18.41
-
22.86
-
-
7.60
-
34.08
12.00
AFDM
Change
Point
AFDM Order
(g/m2) (diatoms)
-
-
4.50 29
3.17 14
-
-
12.32
6.01
6.25
14.42
6.02
-
2.89
-
-
60.87
11.42

171.39
7.97
12.15
169.11
6.51
-
-
12.35
6.17
-
45.68
6.25
10.29
6.30
26.26
12.32
17.17
-
12.25
-
19.00
-
TN
PCT_MAP PCT_MCP (mg/L)
0.04
1.30
0.08
0.00 0.00 0.11
-
17.50
31.21 0.19
-
7.00 39.52
35.00 - 0.21
60.98 12.38
31.21 13.00
-
0.30
0.11
0.52
18.00 - 8.14
-
0.31
36.00
45.00
2.00 67.83 0.94
38.05 - 0.48
0.17
0.55
42.43 77.02 0.48
60.02 0.24
4.00
28.50 0.17
35.00
20.00 12.00 0.32
18.00 6.46
-
19.28 - 0.21
11.00 50.24 0.24
41.50
37.50 80.02
0.22
0.61
10.00 13.17 4.59
TP
(mg/L)
-
0.06
0.02
0.02
-
-
-
-
-
-
-
-
-
0.12
0.08
-
0.03
0.07
-
-
0.39
0.12
0.04
0.14
0.58
0.10
-
0.07
-
0.05
0.04
-
-
-
-
-
0.03
-
TP
Change
Point
Order
(diatoms)
-
57
29
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
;
-
-
-
-
-
-
-
-
152

-------
Table C.I
Direction
Of
Response














increase
increase












increase
Assemblage Taxon
Ochthebius
Oligochaeta
Orthodadiinae
Ostmcoda
Oxyethira
Peltodytes
Pericoma,
Telmatoscopus
Petrophila
Physa, Physella
Prostoma
Psycho da
Sphaeriidae
Tanypodinae
Tipula
Trichocorixa
Tricorythodes
Tropisternus
BMI Turbellaria
Achnanthidium
Diatom exiguum
Achnanthidium
exiguum var
hetemvalvum
Amphora copulata
Amphora libyca
Amphora ovalis
Amphora pediculus
Amphora perpusilla
Amphora sp 1
SCCWRP BSL
Amphora sp 1 SWAMP
JPK
Amphora sp 5 SWAMP
JPK
Amphora stoermerii
Aulacoseira granulata
Bacillaria paradoxa
Biremis sp 1 SCCWRP
JPK
Colonels amphisbaena
Colonels silicula
Diatom Cocconeis pediculus
Chloro
phyll a
(mg/m2)
-
17.37
159.69
29.67
-
-
-
352.84
17.15
-
-
-
-
-
-
9.34
-
18.41
-
-
-
-
37.82
12.88
10.35
20.55
203.60
747.25
-
-
37.82
-
46.94
-
22.67
AFDM
(g/m2)
-
12.46
-
12.35
5.32
-
-
7.29
20.29
7.87
-
-
3.17
20.61
26.75
6.47
-
-
-
289.93
71.95
-
95.37
2.60
28.55
-
227.52
153.88
34.68
-
28.27
168.60
55.91
9.24
7.80
AFDM
Change
Point
Order
(diatoms)
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
98
63
-
75
2
37
-
94
87
43
-
36
89
55
8
6
PCT_MAP PCT_MCP
26.00
0.95 5.36
-
26.83 9.31
50.00
28.71 57.62
-
9.71
39.00 18.00
-
-
36.50
18.00
-
-
19.78
66.00
-
-
-
-
-
57.07
-
67.63 4.76
-
42.34 35.62
-
-
26.11
20.99
-
-
17.07
7.77
TN
(mg/L)
0.28
0.29
1.80
0.41
-
-
8.00
-
0.21
-
-
-
-
8.46
0.60
-
0.23
0.46
7.02
-
0.74
0.66
1.05
0.18
0.21
-
-
1.91
1.01
1.89
0.54
-
3.18
-
4.28
TP
(mg/L)
-
0.05
-
0.08
-
-
-
-
0.03
-
0.19
-
-
-
0.06
-
-
0.03
0.03
-
0.05
0.04
-
-
0.02
-
0.13
0.52
0.30
0.25
0.08
-
-
-
-
TP
Change
Point
Order
(diatoms)
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
12
-
21
16
-
-
5
-
64
90
82
79
44
-
-
-
-
153

-------
Table C.I
Direction
Of
Response   Assemblage
                               AFDM
                              Change
              Chloro            Point
              phyll a   AFDM    Order
                           TN
Taxon
(mg/m2)  (g/m2)  (diatoms)  PCT_MAP PCT_MCP (mg/L)
                                  TP
                                 (mg/L)
                          TP
                        Change
                         Point
                         Order
                       (diatoms)
increase    Diatom
                       Craticula accomoda
                       Craticula cuspidata
                       Craticula halophila
                                        54.14
              48.94    18.02
                                26
         0.40
         0.89
37.88     1.28
                                                                  0.17
                                                                  0.10
                       Cyclostephanos
                         invisitatus
                                                                  0.23
                       Gomphonema affine

                       Gomphonema augur
                       Gomphonema
                         lagenula
              10.55

              121.56   60.75

              137.52   66.22
57

61
        47.82
                 70.48

                 15.24
         3.86

         2.48
                       Gomphonema
                         mexicanum
                       Gomphonema
                         parvulum
              46.86   27.37
                                33
                                        25.00
                                        45.86
                                                  70.95
                                                  15.24
                                                          0.17
                                                          0.62
                                                                  0.62

                                                                  0.11
                                            73
                                            56
                                                                            76
Cyclostephanos
tholiformis
Cyclotella 2g_gg
meneghmiana
Cymatopleura solea
Denticula kuetzingii 20.85
Diadesmis ^ ^
confervacea
Diatoma vulgaris
Diploneis elliptica
Diploneis oblongella
Diploneis smithii
Discostella
pseudostelligera
Entomoneis alata 63.93
Entomoneis paludosa
Eolimna subadnata
Eolimna subminuscula 148.16
Eolimna tantula
Epithemia sorex
Epithemia turgida
Fallacia monoculata
Fallacia pygmaea
Fallacia tenera
Fragilaria capucina
Frustulia
creuzburgensis
Frustulia vulgaris
Gomphonema
8.09
acummatum
-
33.40
68.86
18.66
137.14
-
-
304.33
-
-

49.58
201.11
45.78
304.33
-
-
6.20
209.19
36.48
-
-
150.31
-
-
42
62
27
84
-
-
99
-
-
-
52
91
49
100
-
-
3
93
45
-
-
86
-
0.24
36.00 19.05 0.44
25.36
35.27 - 2.11
89.00 - 0.39
15.00
70.48
12.10
16.51
0.24
39.00 0.94
6.64
2.42
1.91
39.00
8.00
5.00
28.29 0.57
0.57
0.70
0.95
1.14
8.72
25.86
-
0.06
-
-
0.08
-
-
-
-
0.08
0.05
0.08
-
0.20
-
-
-
0.05
0.14
0.14
-
0.20
0.02
-
-
30
-
-
37
-
-
-
-
42
28
39
-
74
-
-
-
24
69
68
-
75
8
-
                                            94

                                            58
                                                      154

-------
Table C.I
Direction Chloro
Of phyll a
Response Assemblage Taxon (mg/m2)
Gomphonema
pseudoaugur
Gomphonema
truncatum
Gyrosigma
acuminatum
Gyrosigma nodiferum 28.73
Halamphora ^Q 3Q
coffeaeformis
increase Diatom Halamphora normanii
Halamphora veneta 47.23
Hantzschia amphioxys
Hippodonta capitata
Hippodonta hungarica
Hippodonta pumila
Karayevia ploenensis
Lemnicola hungarica
Luticola cohnii
Luticola goeppertiana
Luticola mutica
Mayamaea agrestis 31.65
Mayamaea atomus
Melosira varians
Navicula angusta
Navicula arenaria
Navicula can
Navicula cincta
Navicula
cryptocephala
Navicula digitoradiata
increase Diatom Navicula erifuga
Navicula graciloides
Navicula gregaria 26.73
Navicula libonensis
Navicula normaloides 171.70
Navicula peregrina
Navicula phyllepta 28.04
Navicula radiosa
Navicula radiosa var
tenella
Navicula recens
Navicula
rhynchocephala
AFDM
Change
Point
AFDM Order TN
(g/m2) (diatoms) PCT_MAP PCT_MCP (mg/L)
80.51
-
77.65
87.18
14.30
-
14.05
-
19.01
27.83
-
-
-
87.70
-
72.61
100.83
11.42
-
-
62.50
-
6.48
-
-
-
109.30
13.90
23.18
72.93
124.41
47.96
-
27.37
13.97
-
67
-
66
68
21
-
19
-
28
35
-
-
-
69
-
64
78
14
-
-
60
-
5
-
-
-
79
17
32
65
82
50
-
34
18
-
42.00 - 1.54
31.00
0.66
30.48 0.61
9.93
-
41.75 - 0.76
46.33 0.31
57.07 0.62
79.52 0.98
0.26
0.46
51.21
.
-
64.76 0.81
5.84
6.00
2.96
-
-
50.24 1.33
31.43 0.60
80.99
61.45
3.00 0.94
73.73 0.36
7.00 0.57
13.00
0.67
0.58
3.81
25.86
-
9.74
7.00
TP
(mg/L)
-
-
-
0.05
-
0.08
0.09
0.08
0.06
0.10
-
0.04
0.13
-
0.39
0.07
-
-
0.02
0.04
-
-
0.13
-
-
1.87
0.24
0.08
-
-
-
-
-
-
-
-
TP
Change
Point
Order
(diatoms)
-
-
-
25
-
41
49
40
32
55
-
19
66
-
86
36
-
-
2
15
-
-
65
-
-
98
78
45
-
-
-
-
-
-
-
-
155

-------
Table C.I
Direction
Of
Response Assemblage Taxon
Navicula rostellata
Navicula salinarum
Navicula schroeteri
Navicula sp 3 SWAMP
JPK
Navicula subrotundata
Navicula tenelloides
Navicula trivialis
Navicula veneta
Navicula viridula
Navicula viridula var
linearis
Nitzschia acicularis
Nitzschia amphibia
increase Diatom Nitzschia amphiboides
Nitzschia angustatula
Nitzschia aurariae
Nitzschia bacillum
Nitzschia bryophila
Nitzschia
bulnheimiana
Nitzschia capitellata
Nitzschia communis
Nitzschia commutata
Nitzschia compressa
var vexans
Nitzschia desertorum
Nitzschia dubia
Nitzschia elegantula
Nitzschia filiformis
Nitzschia fan ticola
Nitzschia frustulum
Nitzschia inconspicua
Nitzschia intermedia
Nitzschia lacuum
Nitzschia liebethruthii
Nitzschia linearis
Nitzschia
microcephala
Nitzschia minuta
Nitzschia palea
Nitzschia paleaeformis
Chloro
phyll a
(mg/m2)
-
65.28
-
747.25
-
-
-
13.40
-
-
-
26.43
94.19
-
50.71
117.41
-
25.97
126.93
-
57.63
-
77.89
-
13.70
23.95
-
314.91
63.87
-
9.38
10.30
-
15.28
-
49.81
-
AFDM
Change
Point
AFDM Order
(g/m2) (diatoms)
-
111.63
90.46
-
-
100.08
12.71
11.40
-
-
-
17.20
201.11
271.41
14.95
-
-
11.27
-
93.23
48.78
50.88
14.13
-
60.87
-
-
13.09
2.26
-
8.38
32.83
16.63
29.77
-
44.72
28.57
-
80
72
-
-
77
15
13
-
-
-
25
92
97
22
-
-
12
-
73
51
53
20
-
58
-
-
16
1
-
7
41
24
40
-
48
39
PCT_MAP PCT_MCP
-
15.24
-
-
-
-
72.45
8.00
2.00
6.92
2.00
31.38 59.04
-
-
16.00
-
-
90.50
-
37.07
10.48
63.33
29.00
-
-
-
7.69
53.67
70.74
-
-
-
29.26
18.02 12.00
-
6.81
-
TN
(mg/L)
-
0.97
1.12
1.91
0.62
-
0.21
0.16
0.24
-
-
0.76
3.12
0.57
1.33
-
1.81
1.47
8.63
2.03
-
-
1.05
0.54
-
16.42
-
-
0.22
-
-
0.66
-
0.29
4.45
0.73
-
TP
(mg/L)
0.30
-
0.09
0.38
0.46
-
0.12
-
0.09
-
-
0.06
-
0.05
1.01
-
0.44
0.12
0.09
0.08
-
0.02
0.69
0.05
-
-
0.03
0.10
0.03
0.92
-
-
-
0.08
-
0.09
-
TP
Change
Point
Order
(diatoms)
83
-
50
85
89
-
61
-
51
-
-
33
-
29
97
-
87
60
48
46
-
7
95
27
-
-
10
54
9
96
-
-
-
47
-
52
-
Nitzschia perminuta
39.02
                                 156

-------
Table C.I
Direction
Of
Response

increase







increase













increase



Chloro
phyll a
Assemblage Taxon (mg/m2)
Nitzschia perspicua
Diatom Nitzschia rosenstockii
Nitzschia sigma
Nitzschia solita 28.04
Nitzschia umbonata
Nitzschia valdecostata
Nitzschia vitrea
Parlibellus protract a
Placoneis elginensis
Planothidium
delicatulum
Planothidium
engelbrechtii
Planothidium
frequentissimum
Planothidium
Diatom lanceolatum
Pleurosira laevis 58.35
Psammodictyon
+ • + 121.56
constnctum
Pseudostaurosira
elliptica
Pseudostaurosira
parasitica
Pseudostaurosira
subsalina
Rhoicosphenia sp 1
SCCWRPJPK
Rhoicosphenia sp B
SWAMP EWT
Rhopalodia gibba
Rhopalodia musculus
Rhopalodia operculata
Sellaphora hustedtii
Sellaphora laevissima
Sellaphora nyassensis
Sellaphora pupula
Sellaphora seminulum
Sellaphora sp 2
Diatom SWAMP JPK
Simonsenia delognei
Stauroneis smithii
Staurosira construens
AFDM
(g/m2)
-
98.26
-
-
-
90.46
162.77
6.26
-
10.92
137.14
10.27
39.47
20.57
61.55
34.69
117.98
-
-
-
-
-
138.66
229.99
-
-
21.87
-
-
187.78
227.52
-
AFDM
Change
Point
Order
(diatoms) PCT_MAP PCT_MCP
-
76 - 58.55
-
-
.
71
88
4
-
11 36.00 17.00
83 - 53.67
10 88.50
46 - 42.00
29 83.33 31.21
59 - 21.00
44 24.22 5.86
81 - 39.00
90.50
20.95
-
3.00
10.00
85
96
-
39.00
30 - 2.00
-
6.03
90
95
3.98 62.86
TN TP
(mg/L) (mg/L)
4.58 0.28
0.98 0.13
0.62 0.44
0.03
3.75 0.15
0.03
-
0.38 0.02
0.23
0.61 0.10
0.26 0.02
0.41 0.06
0.02
6.46 0.08
0.94
0.54 0.13
0.05
4.15
-
0.04
-
-
0.20
-
0.30
-
0.46 0.26
0.02
-
-
-
0.58
TP
Change
Point
Order
(diatoms)
81
67
88
14
70
13
-
6
77
57
4
35
3
38
-
63
26
-
-
17
-
-
-
-
84
-
80
1
-
-
-
93
157

-------
Table C.I
Direction Chloro
Of phyll a AFDM
Response Assemblage Taxon (mg/m2) (g/m2)
Staurosim construens
var binodis
Staurosim construens

var venter
Staurosirella pinnata
Stephanodiscus
medius
Surirella angusta
Surirella brebissonii 28.73 22.00
Surirella brebissonii
. ^ . .. - 15.98
var kuetzmgn
Surirella brightwellii
Surirella ovalis - 52.31
Surirella ovata
Synedra ulna
Tabulariafasciculata 28.79 28.57
Tabularia tabulata
Thalassionema „„,-.,„
increase Diatom . .... 185.30 43.73
nitzschioides
Thalassiosira ^ ^
weissflogn
Tryblionella calida - 57.08
Tryblionella constricta 63.87 9.49
Tryblionella hungarica 136.63 89.37
Tryblionella levidensis
Tryblionella littoralis
AFDM
Change
Point
Order
(diatoms) PCT_MAP PCT_MCP

16.27

32.00

-


15.74
31 - 57.62
23 - 10.48
10.00
54 - 20.95
-
2.86 0.00
38 66.81 12.38
-
47 - 39.05
74

56
9 - 20.80
70 - 25.00
-
-


TN
(mg/L)

"

0.17

-


0.35
1.07
0.58
9.93
0.49
5.59
-
0.36
0.38
-
0.76

1.26
0.97
0.74
0.69
0.70


TP
(mg/L)

"

0.05

0.03
0.05

0.12
0.15
0.04
-
0.09
0.53
-
0.08
0.06
-
0.15

-
0.06
0.11
0.05
0.53
TP
Change
Point
Order
(diatoms)

"

20

11
22

62
71
18
-
53
92
-
43
31
-
72

-
34
59
23
91
158

-------
Table C.2. Results of piecewise regressions for all analyses in which "relaxed"
criteria were met.

Gradient
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll a
(mg/m2)
Chlorophyll o
(mg/m2)
Chlorophyll o
(mg/m2)
Chlorophyll o
(mg/m2)
Chlorophyll o
(mg/m2)
Chlorophyll o
(mg/m2)
Chlorophyll o
(mg/m2)
AFDM (g/m2)
AFDM (g/m2)

Response
D18
D18
EPT_Taxa
EPT_Taxa
H20
H20
H21
H21
H23
H23
lntolerant_
Percent
lntolerant_
Percent
lntolerant_
PercentTaxa
lntolerant_
PercentTaxa
lntolerant_
Taxa
lntolerant_
Taxa
RAWpropGree
nCRUS
RAWpropGree
nCRUS
S2
S2
CSCI
CSCI

Analysis
Type
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted

Breakpoint (SE),
95% Confidence
Interval Width
98.82 (12.11),
47.55
119.29(20.48),
80.40
68.77(8.72),
34.25
79.77 (14.57),
57.24
104.54(10.91),
42.88
118.31 (20.40),
80.14
93.13 (9.46),
37.14
110.66(17.42),
68.44
102.55 (9.86),
38.71
119.02 (19.66),
77.18
23.29(2.91),
11.41
55.73 (10.64),
41.79
31.40(3.34),
13.11
74.91 (12.58),
49.42
31.43(3.56),
13.99
60.93 (10.03),
39.39
103.37(12.60),
49.47
117.59(28.46),
111.78
113.46(13.00),
51.05
133.22(43.18),
169.53
15.02 (1.87),
7.35
16.75 (2.33),
9.14

Slope 1
(95% Confidence
Interval)
-0.32
(-0.39- -0.24)
-0.32
(-0.38- -0.25)
-0.18
(-0.23- -0.14)
-0.21
(-0.26- -0.16)
-0.31
(-0.37- -0.25)
-0.27
(-0.33- -0.21)
-0.36
(-0.43- -0.29)
-0.34
(-0.40- -0.27)
-0.35
(-0.41- -0.29)
-0.30
(-0.36- -0.24)
-0.01
(-0.01- -0.01)
0.00
(-0.01-0.00)
-0.01
(-0.01- -0.01)
0.00
(-0.01-0.00)
-0.30
(-0.37- -0.23)
-0.22
(-0.28- -0.17)
0.01
(0.00-0.01)
0.00
(0.00-0.01)
-0.32
(-0.39- -0.26)
-0.19
(-0.26- -0.12)
-0.02
(-0.03- -0.01)
-0.02
(-0.03- -0.01)


Slope 2
(95% Confidence Adjusted
Interval) R2
0.00
(-0.02-0.02)
-0.01
(-0.07-0.06)
0.00
(-0.01-0.01)
0.00
(-0.03-0.02)
-0.01
(-0.02-0.01)
-0.01
(-0.07-0.05)
-0.01
(-0.02-0.01)
-0.01
(-0.07-0.05)
-0.01 (
-0.02-0.01)
-0.01
(-0.07-0.05)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
0.00(0.00-0.00)
0.00
(0.00-0.00)
0.00
(-0.01-0.00)
0.00
(-0.03-0.02)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
-0.01
(-0.03-0.01)
-0.01
(-0.09-0.07)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
0.19
0.14
0.25
0.18
0.28
0.16
0.28
0.19
0.31
0.17
0.21
0.14
0.29
0.20
0.26
0.18
0.22
0.09
0.25
0.06
0.23
0.19
all 4
criteria
(relaxed)
fulfilled?
yes
no
yes
no
yes
no
yes
no
yes
no
yes
no
yes
no
yes
no
yes
no
yes
no
yes
no
all 4
criteria
(strict)
fulfilled?
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
                                      159

-------
Table C.2 (continued)



Gradient
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)



Response
D18
D18
EPT_Percent
EPT_Percent
EPT_Percent
Taxa
EPT_Percent
Taxa
EPT_Taxa
EPT_Taxa
H20
H20
H21
H21
H23
H23
lntolerant_
Percent
lntolerant_
Percent
lntolerant_
PercentTaxa
lntolerant_
PercentTaxa
lntolerant_
Taxa
lntolerant_
Taxa
propAchMin
propAchMin
RAWD0100
RAWD0100


Analysis
Type
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted

Breakpoint (SE),
95% Confidence
Interval Width
31.72(3.70),
14.55
33.91(4.73),
18.59
16.10(1.62),
6.38
15.42 (1.41),
5.52
18.93 (1.84),
7.23
22.19(2.96),
11.61
17.47 (1.64),
6.43
19.71(2.72),
10.67
36.06(3.68),
14.44
43.02 (6.79),
26.66
34.75 (3.40),
13.36
34.65 (4.74),
18.62
35.79 (3.40),
13.35
35.05 (5.79),
22.74
12.99(1.10),
4.34
13.74 (1.45),
5.69
15.41 (1.26),
4.94
16.20 (2.01),
7.90
15.29(1.27),
4.98
16.52 (2.04),
8.01
11.17(1.41),
5.52
14.52 (2.17),
8.52
7.10(0.88),
3.45
14.39 (2.65),
10.40

Slope 1
(95% Confidence
Interval)
-0.93
(-1.17- -0.70)
-0.84
(-1.04- -0.64)
-0.02
(-0.03- -0.02)
-0.03
(-0.03- -0.02)
-0.02
(-0.02- -0.01)
-0.01
(-0.01- -0.01)
-0.78
(-0.95- -0.61)
-0.65
(-0.83- -0.48)
-0.81
(-0.98- -0.63)
-0.56
(-0.69- -0.43)
-0.91
(-1.10- -0.72)
-0.78
(-0.96- -0.60)
-0.91
(-1.09- -0.73)
-0.66
(-0.83- -0.49)
-0.02
(-0.02- -0.02)
-0.02
(-0.02- -0.01)
-0.02
(-0.02- -0.02)
-0.02
(-0.02- -0.01)
-0.78
(-0.94- -0.62)
-0.71
(-0.89- -0.53)
-0.02
(-0.03- -0.01)
-0.01
(-0.02- -0.01)
-0.04
(-0.06- -0.02)
-0.01 (
-0.02- -0.01)

Slope 2
(95% Confidence
Interval)
-0.05
(-0.09- -0.01)
-0.06
(-0.12-0.00)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
-0.02
(-0.04 - 0.00)
-0.02
(-0.06-0.03)
-0.03
(-0.06-0.01)
-0.04
(-0.09-0.02)
-0.02
(-0.06-0.02)
-0.03
(-0.09-0.02)
-0.04
(-0.08-0.00)
-0.06
(-0.12- -0.01)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
-0.01
(-0.03-0.00)
-0.01
(-0.05-0.03)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
0.00
(0.00-0.00)


Adjusted
R2
0.24
0.20
0.29
0.30
0.35
0.23
0.35
0.20
0.27
0.18
0.28
0.20
0.31
0.18
0.31
0.22
0.38
0.21
0.36
0.20
0.13
0.10
0.10
0.06
all 4
criteria
(relaxed)
fulfilled?
yes
no
yes
yes
yes
no
yes
no
yes
no
yes
no
yes
no
yes
yes
yes
yes
yes
yes
yes
no
yes
no
all 4
criteria
(strict)
fulfilled?
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
                                                   160

-------
Table C.2 (continued)



Gradient
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
AFDM (g/m2)
NOX (mg/L)
NOX (mg/L)
NOX (mg/L)
NOX (mg/L)
NOX (mg/L)
NOX (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)



Response
S2
S2
Tolerant_
PercentTaxa
Tolerant_
PercentTaxa
H20
H20
H23
H23
S2
S2
D18
D18
EPT_Percent
Taxa
EPT_Percent
Taxa
EPT_Taxa
EPT_Taxa
H20
H20
H21
H21
H23
H23
lntolerant_
PercentTaxa
lntolerant_
PercentTaxa


Analysis
Type
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted

Breakpoint (SE),
95% Confidence
Interval Width
39.33 (4.90),
19.24
94.63 (31.41),
123.32
28.40 (3.46),
13.61
29.18 (4.86),
19.09
0.61 (0.07),
0.28
0.63 (0.09),
0.34
0.38(0.05),
0.18
0.60 (0.09),
0.34
0.29(0.03),
0.13
0.29(0.05),
0.18
0.13(0.01),
0.04
0.13(0.01),
0.04
0.13(0.01),
0.05
0.14(0.02),
0.07
0.13(0.01),
0.06
0.13 (0.02),
0.07
0.13(0.01),
0.04
0.14(0.01),
0.05
0.14(0.01),
0.04
0.14(0.01),
0.05
0.14(0.01),
0.04
0.14(0.01),
0.05
0.12(0.01),
0.05
0.12 (0.02),
0.07

Slope 1
(95% Confidence
Interval)
-0.76
(-0.96- -0.56)
-0.25
(-0.36- -0.15)
0.01
(0.01-0.01)
0.01
(0.00-0.01)
-52.67
(-64.23 --41. 12)
-60.65
(-73. 24 --48.06)
-82.75
(-102.20 --63. 34)
-63.72
(-77.25 --50.19)
-129.30
(-158.40 --100.30)
-114.80
(-147.20 --82. 37)
-329.50
(-378.60 --280.40)
-347.40
(-401. 10 --293.80)
-2.37
(-2. 84- -1.89)
-2.21
(-2. 75- -1.68)
-103.80
(-125.30 --82.41)
-115.70
(-145.00 --86.37)
-315.70
(-360.00 --271.40)
-284.30
(-330.80 --237.80)
-299.60
(-347.30 --251.80)
-300.20
(-352.40 --248.00)
-312.40
(-359.60 --265.20)
-284.60
(-334.80 --234.40)
-2.36
(-2. 87- -1.85)
-2.40
(-2. 98- -1.81)

Slope 2
(95% Confidence
Interval)
-0.01
(-0.06-0.04)
0.01
(-0.11-0.12)
0.00
(0.00-0.00)
0.00
(0.00-0.00)
-0.10
(-0.81-0.62)
-0.70
(-2.53-1.14)
-0.46
(-1.17-0.24)
-0.64
(-2.60-1.33)
-0.92
(-1.67- -0.18)
-0.98
(-2.88-0.93)
1.78
(-5.03-8.60)
7.36
(-6.93-21.64)
-0.01
(-0.06-0.04)
-0.04
(-0.14-0.06)
-0.32
(-2.52-1.89)
-2.49
(-7.91-2.94)
-1.12
(-7.26-5.02)
4.09
(-8.30-16.48)
1.04
(-5.71-7.79)
5.03
(-8.87-18.93)
-0.05
-6.72-6.61)
2.78
(-10.60-16.15)
-0.01
(-0.06-0.03)
-0.05
(-0.16-0.06)


Adjusted
R2
0.18
0.05
0.30
0.19
0.33
0.26
0.33
0.24
0.39
0.21
0.33
0.28
0.26
0.22
0.25
0.20
0.37
0.29
0.31
0.26
0.34
0.25
0.24
0.20
all 4
criteria
(relaxed)
fulfilled?
yes
no
yes
no
yes
no
yes
no
yes
no
yes
yes
yes
no
yes
no
yes
yes
yes
yes
yes
yes
yes
no
all 4
criteria
(strict)
fulfilled?
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
yes
no
no
no
no
no
no
no
                                                   161

-------
Table C.2 (continued)



Gradient
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)
SRP (mg/L)



Response
lntolerant_
Taxa
lntolerant_
Taxa
propAchMin
propAchMin
RAWD0100
RAWD0100
RAWD050
RAWD050
RAWeutro
RAWeutro
RAWIowN
RAWIowN
RAWIowP
RAWIowP
RAWNhet
RAWNhet
RAWsapro
RAWsapro
S2
S2
Taxonomic_
Richness
Taxonomic_
Richness
Tolerant_
PercentTaxa
Tolerant_
PercentTaxa


Analysis
Type
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted

Breakpoint (SE),
95% Confidence
Interval Width
0.12(0.01),
0.06
0.12 (0.02),
0.07
0.09(0.01),
0.04
0.09(0.01),
0.05
0.08(0.01),
0.03
0.05 (0.00),
0.02
0.23 (0.02),
0.09
0.17(0.02),
0.07
0.09(0.01),
0.04
0.08(0.01),
0.04
0.10(0.01),
0.03
0.10(0.01),
0.05
0.09(0.01),
0.03
0.09(0.01),
0.04
0.14(0.02),
0.07
0.14(0.02),
0.07
0.14(0.01),
0.06
0.13 (0.02),
0.06
0.13(0.01),
0.06
0.14(0.03),
0.13
0.14(0.02),
0.07
0.14(0.02),
0.08
0.13 (0.02),
0.06
0.15 (0.02),
0.08

Slope 1
(95% Confidence
Interval)
-85.69
(-105.50 --65. 85)
-104.10
(-131.40 --76.87)
-2.92
(-3. 59 --2.24)
-3.27
(-4.09 --2.45)
-4.36
(-5.45 --3.27)
-6.94
(-8.58- -5.31)
-1.08
(-1.26- -0.90)
-1.36
(-1.60- -1.13)
3.99
(3.13-4.86)
4.14
(3.09-5.20)
-4.74
(-5. 57 --3.92)
-4.59
(-5. 56- -3.61)
-5.28
(-6.19 --4.38)
-5.59
(-6.66 --4.52)
1.65
(1.26-2.05)
1.95
(1.53-2.37)
-2.73
(-3. 28 --2. 17)
-2.85
(-3.48 --2.22)
-277.70
(-338.10 --217.40)
-164.40
(-228.90 --99.92)
-128.10
(-157.80 --98.42)
-135.40
(-172.50 --98.38)
1.65
(1.24-2.05)
1.55
(1.15-1.95)

Slope 2
(95% Confidence
Interval)
-0.38
(-2.22-1.46)
-1.59
(-6.60-3.41)
-0.01
(-0.07-0.05)
-0.01
(-0.17-0.15)
0.04
(-0.04-0.12)
0.16
(-0.02-0.34)
0.08
(0.03-0.13)
0.09
(0.01-0.17)
-0.05
(-0.13-0.03)
-0.20
(-0.41-0.00)
-0.02
(-0.11-0.06)
0.03
(-0.19-0.26)
-0.04
(-0.12-0.05)
-0.02
(-0.25-0.20)
0.01
(-0.04-0.07)
-0.02
(-0.13-0.09)
0.06
(-0.02-0.14)
0.17
(0.00-0.33)
-4.84
(-13.02-3.34)
-9.55
(-26.76-7.66)
-0.10
(-3.20-3.01)
-3.08
(-9.96-3.79)
-0.01
(-0.05-0.03)
0.03
(-0.05-0.10)


Adjusted
R2
0.21
0.17
0.18
0.13
0.16
0.13
0.22
0.22
0.19
0.11
0.28
0.16
0.29
0.19
0.18
0.18
0.21
0.15
0.20
0.07
0.22
0.19
0.19
0.21
all 4
criteria
(relaxed)
fulfilled?
yes
no
yes
no
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
no
yes
no
yes
no
all 4
criteria
(strict)
fulfilled?
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
                                                   162

-------
Table C.2 (continued)



Gradient
RAWpropGre
enCRUS
RAWpropGre
enCRUS
RAWpropGre
enCRUS
RAWpropGre
enCRUS
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)



Response
propTaxaZHR
propTaxaZHR
S2
S2
CSCI
CSCI
D18
D18
EPT_Percent
EPT_Percent
EPT_Percent
Taxa
EPT_Percent
Taxa
EPT_Taxa
EPT_Taxa
H20
H20
H21
H21
H23
H23
lntolerant_
Percent
lntolerant_
Percent
lntolerant_
PercentTaxa
lntolerant_
PercentTaxa


Analysis
Type
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted

Breakpoint (SE),
95% Confidence
Interval Width
0.86 (0.08),
0.32
0.92 (0.10),
0.41
0.99(0.01),
0.02
0.97(0.03),
0.11
0.74 (0.06),
0.25
0.72 (0.07),
0.27
0.88 (0.07),
0.26
1.29(0.13),
0.50
0.59(0.07),
0.26
0.64 (0.08),
0.32
0.68 (0.04),
0.16
0.72 (0.06),
0.22
0.63 (0.04),
0.14
0.62 (0.05),
0.21
1.06 (0.06),
0.25
1.29(0.12),
0.46
0.68(0.05),
0.18
1.19(0.12),
0.47
0.77 (0.04),
0.18
1.21(0.11),
0.45
0.54(0.05),
0.19
0.55 (0.06),
0.25
0.62 (0.04),
0.15
0.58(0.05),
0.19

Slope 1
(95% Confidence
Interval)
-0.07
(-0.13- -0.02)
-0.01
(-0.07-0.05)
-42.71
(-46.16 --39.27)
-33.86
(-39.20 --28.52)
-0.58
(-0.68- -0.47)
-0.73
(-0.85- -0.60)
-45.66 (
-53. 16 --38.16)
-34.89
(-39.88 --29.91)
-0.60
(-0.75- -0.45)
-0.60
(-0.74- -0.47)
-0.56
(-0.63- -0.48)
-0.55
(-0.62- -0.47)
-27.25
(-30.69 --23.81)
-31.09
(-35. 72 --26.46)
-40.18
(-45.02 --35.33)
-32.65
(-36.85 --28.46)
-58.63
(-67.41 --49.85)
-35.13
(-40.14 --30.12)
-56.32
(-63.40 --49.25)
-34.86
(-39.56 --30.16)
-0.46
(-0.56- -0.37)
-0.50
(-0.61- -0.40)
-0.57
(-0.65- -0.50)
-0.65
(-0.75- -0.55)

Slope 2
(95% Confidence
Interval)
-0.57
(-1.14-0.00)
-0.72
(-2.77-1.32)
1290
(3768-1188)
329.8
(896.6-237)
0.01
(0.00-0.01)
0.00
(-0.02-0.02)
0.38
(-0.22-0.97)
0.25
(-1.54-2.04)
0.00
(0.00-0.01)
0.00
(-0.02-0.03)
0.00
(0.00-0.01)
0.00
(-0.01-0.01)
0.01
(-0.22-0.23)
-0.07
(-0.81-0.67)
0.29
(-0.22-0.80)
-0.14
(-1.64-1.37)
-0.19
(-0.72-0.34)
-0.33 (-2.02-
1.37)
-0.18 (-0.69-
0.33)
-0.31
(-1.90-1.27)
0.00
(-0.01-0.00)
0.00
(-0.02-0.01)
0.00
(-0.01-0.00)
0.00
(-0.02-0.01)


Adjusted
R2
0.13
0.01
0.58
0.36
0.40
0.37
0.37
0.31
0.27
0.23
0.59
0.46
0.60
0.41
0.53
0.39
0.46
0.34
0.53
0.36
0.37
0.26
0.57
0.41
all 4
criteria
(relaxed)
fulfilled?
yes
no
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
all 4
criteria
(strict)
fulfilled?
no
no
yes
yes
no
no
yes
no
no
no
yes
no
yes
no
yes
no
yes
no
yes
no
no
no
yes
no
                                                   163

-------
Table C.2 (continued)



Gradient
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)

TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)



Response
lntolerant_
Taxa
lntolerant_
Taxa
0/E
0/E
propTaxaZHR
propTaxaZHR
RAWD050
RAWD050
RAWIowN
RAWIowN
RAWIowP
RAWIowP
RAWmeanZHR
RAWmeanZHR
RAWNhet
RAWNhet
RAWprop
GreenCRUS
RAWprop
GreenCRUS

RAWsapro
RAWsapro
S2
S2
Shannon_
Diversity
Shannon_
Diversity


Analysis
Type
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted

unweighted
weighted
unweighted
weighted
unweighted
weighted

Breakpoint (SE),
95% Confidence
Interval Width
0.52 (0.03),
0.13
0.51 (0.04),
0.18
0.79 (0.09),
0.36
0.78 (0.09),
0.35
0.84(0.10),
0.37
0.70 (0.20),
0.77
1.95(0.19),
0.76
1.89(0.16),
0.62
0.94 (0.09),
0.37
1.14(0.23),
0.91
0.78 (0.08),
0.30
0.85(0.15),
.59
0.67 (0.07),
0.29
0.90(0.25),
0.97
1.95(0.22),
0.85
1.58(0.20),
0.80
0.60 (0.07),
0.29
0.58(0.12),
0.46

1.04(0.11), 0.44
1.29(0.19),
0.73
0.83 (0.06),
0.24
0.93 (0.14),
0.53
0.76 (0.08),
0.30
0.75(0.11),
0.43

Slope 1
(95% Confidence
Interval)
-25.78
(-29.45 --22. 10)
-31.35
(-36.28 --26.43)
-0.46
(-0.57- -0.35)
-0.64
(-0.77- -0.51)
-0.24
(-0.29- -0.18)
-0.18
(-0.26- -0.09)
-0.11
(-0.13- -0.09)
-0.16
(-0.18- -0.13)
-0.44
(-0.54- -0.35)
-0.32
(-0.41- -0.22)
-0.53
(-0.65- -0.42)
-0.43
(-0.57- -0.29)
-0.44
(-0.55- -0.34)
-0.25
(-0.37- -0.13)
0.13
(0.11-0.16)
0.19
(0.15-0.22)
0.85
(0.61-1.08)
0.54
(0.35-0.74)
-0.33 (-0.40 --
0.25)
-0.28 (
-0.34- -0.22)
-52.74
(-60.75 --44.72)
-34.70
(-43. 39 --26.01)
-1.29
(-1.56- -1.03)
-1.35
(-1.71- -1.00)

Slope 2
(95% Confidence
Interval)
-0.06
(-0.25-0.14)
-0.13
(-0.81-0.55)
0.01
(0.00-0.02)
0.00
(-0.02-0.03)
0.00
(-0.01-0.00)
-0.01
(-0.02-0.01)
0.01
(0.00-0.01)
0.02
(0.01-0.03)
0.01
(0.00-0.02)
0.00
(-0.03-0.03)
0.01
(0.00-0.01)
0.00
(-0.02-0.03)
0.00
(-0.01-0.00)
-0.01
(-0.03-0.02)
0.00
(-0.01-0.00)
0.00
(-0.02-0.01)
0.02
(0.01-0.03)
0.02
(0.00-0.05)
0.01
(0.00-0.01)
0.01
(-0.01-0.03)
-0.80
(-1.43- -0.17)
-1.38
(-3.23-0.46)
-0.02
(-0.04-0.01)
-0.01
(-0.07-0.05)


Adjusted
R2
0.53
0.37
0.25
0.29
0.25
0.06
0.23
0.33
0.24
0.09
0.24
0.10
0.25
0.06
0.23
0.21
0.26
0.11

0.22
0.17
0.46
0.20
0.36
0.21
all 4
criteria
(relaxed)
fulfilled?
yes
yes
yes
yes
yes
no
yes
yes
yes
no
yes
no
yes
no
yes
no
yes
no

yes
no
yes
no
yes
no
all 4
criteria
(strict)
fulfilled?
yes
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no

no
no
yes
no
no
no
                                                   164

-------
Table C.2 (continued)



Gradient
N (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TN (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)



Response
Taxonomic_
Richness
Taxonomic_
Richness
Tolerant_
Percent
Tolerant_
Percent
Tolerant_
PercentTaxa
Tolerant_
PercentTaxa
CSCI
CSCI
D18
D18
EPT_Percent
Taxa
EPT_Percent
Taxa
EPT_Taxa
EPT_Taxa
H20
H20
H21
H21
H23
H23
lntolerant_
PercentTaxa
lntolerant_
PercentTaxa
lntolerant_
Taxa
lntolerant_
Taxa


Analysis
Type
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted

Breakpoint (SE),
95% Confidence
Interval Width
0.71 (0.05),
0.19
0.71 (0.07),
0.27
0.61 (0.07),
0.28
0.74 (0.09),
0.37
0.67(0.05),
0.18
0.73 (0.06),
0.23
0.15 (0.02),
0.06
0.15 (0.03),
0.10
0.12(0.01),
0.03
0.14(0.01),
0.05
0.10(0.01),
0.04
0.12(0.01),
0.06
0.11(0.01),
0.04
0.14(0.02),
0.08
0.11(0.01),
0.03
0.13(0.01),
0.05
0.11(0.01),
0.03
0.14(0.01),
0.05
0.11(0.01),
0.03
0.14(0.01),
0.05
0.11(0.01),
0.05
0.12 (0.02),
0.07
0.11(0.01),
0.05
0.12 (0.02),
0.08

Slope 1
(95% Confidence
Interval)
-31.26
(-35. 72 --26.80)
-33.33
(-38.93 --27.72)
0.48
(0.36-0.61)
0.38
(0.29-0.47)
0.42
(0.35-0.48)
0.41 (
0.35-0.46)
-2.70
(-3.29--2.il)
-2.49
(-3. 21- -1.76)
-352.10
(-406.00 --298.20)
-290.00
(-335.00 --245.00)
-3.10
(-3. 73 --2.48)
-2.43
(-3.03- -1.83)
-129.00
(-157.40 --100.50)
-105.90
(-132.40 --79.39)
-369.00
(-420.80 --317.20)
-275.10
(-315. 90 --234.30)
-372.10
(-434.90 --309.30)
-272.00
(-318.00 --226.10)
-371.90
(-426.40 --317.30)
-266.10
(-309.70 --222.60)
-2.60
(-3. 22- -1.97)
-2.24
(-2. 92- -1.57)
-96.90
(-121.90 --71. 91)
-93.62
(-123.20 --64.03)

Slope 2
(95% Confidence
Interval)
-0.07
(-0.40-0.27)
-0.11
(-1.08-0.86)
-0.01
(-0.01-0.00)
0.00
(-0.02-0.01)
0.00
(-0.01-0.00)
0.00
(-0.01-0.01)
0.00
(-0.06-0.06)
-0.12
(-0.24-0.01)
-4.04
(-9.01-0.93)
-9.16
(-19.92-1.60)
-0.02
(-0.06-0.02)
-0.10
(-0.18- -0.02)
-0.74 (
-2.49-1.02)
-3.30
(-7.84-1.25)
-3.79
(-7.96-0.38)
-6.31
(-15.21-2.59)
-2.74
(-7.41-1.93)
-7.26
(-17.59-3.08)
-2.34
(-6.92-2.24)
-5.41
(-15.22-4.40)
-0.02
(-0.06-0.02)
-0.08
(-0.17-0.01)
-0.57
(-2.11-0.98)
-2.62
(-6.70-1.46)


Adjusted
R2
0.54
0.37
0.24
0.25
0.50
0.45
0.28
0.23
0.41
0.35
0.37
0.30
0.34
0.25
0.50
0.38
0.42
0.32
0.46
0.34
0.30
0.22
0.27
0.19
all 4
criteria
(relaxed)
fulfilled?
yes
yes
yes
no
yes
yes
yes
no
yes
yes
yes
no
yes
no
yes
yes
yes
yes
yes
yes
yes
no
yes
no
all 4
criteria
(strict)
fulfilled?
yes
no
no
no
yes
no
no
no
yes
no
no
no
no
no
yes
no
no
no
yes
no
no
no
no
no
                                                   165

-------
Table C.2 (continued)



Gradient
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)
TP (mg/L)



Response
propAchMin
propAchMin
RAWD0100
RAWD0100
RAWD050
RAWD050
RAWeutro
RAWeutro
RAWIowN
RAWIowN
RAWIowP
RAWIowP
RAWIowTPsp
RAWIowTPsp
RAWsapro
RAWsapro
S2
S2
Taxonomic_
Richness
Taxonomic_
Richness
Tolerant_
PercentTaxa
Tolerant_
PercentTaxa


Analysis
Type
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted
unweighted
weighted

Breakpoint (SE),
95% Confidence
Interval Width
0.04 (0.00),
0.01
0.08(0.01),
0.04
0.06(0.01),
0.02
0.03 (0.00),
0.01
0.27(0.03),
0.10
0.29(0.03),
0.11
0.08(0.01),
0.03
0.08(0.01),
0.04
0.09(0.01),
0.03
0.09(0.01),
0.04
0.08(0.01),
0.02
0.08(0.01),
0.03
0.07(0.01),
0.03
0.06(0.01),
0.03
0.15(0.01),
0.06
0.15 (0.02),
0.07
0.11(0.01),
0.04
0.11(0.02),
0.07
0.14(0.02),
0.06
0.15 (0.03),
0.10
0.11(0.01),
0.04
0.16(0.02),
0.08

Slope 1
(95% Confidence
Interval)
-7.42
(-9.19 --5.66)
-3.83 (-4.81- -
2.85)
-5.75
(-7.43- -4.08)
-10.88
(-13.43 --8.33)
-0.86 (-1.01- -
0.72)
-0.89
(-1.03- -0.75)
4.47
(3.33-5.60)
3.96
(2.70-5.22)
-5.53
(-6.58 --4.48)
-4.73
(-5. 90 --3.55)
-6.78
(-7.97- -5.60)
-6.02
(-7.24- -4.81)
-1.99
(-2.49- -1.49)
-2.01
(-2. 71- -1.32)
-2.56
(-3.05- -2.06)
-2.46
(-2. 98- -1.94)
-374.80
(-453. 90 --295.70)
-249.10
(-319.60 --178.60)
-129.10
(-157.00 --101.20)
-109.90
(-143.40 --76.43)
2.17
(1.67-2.67)
1.43
(1.08-1.78)

Slope 2
(95% Confidence
Interval)
-0.05
(-0.10- -0.01)
-0.01
(-0.13-0.11)
0.00 (-0.06-
0.07)
0.01
(-0.12-0.13)
0.04
(0.00-0.07)
0.04
(-0.03-0.11)
-0.02
(-0.09-0.04)
-0.08
(-0.24-0.07)
-0.02
(-0.09-0.04)
-0.05
(-0.21-0.12)
-0.03
(-0.10-0.03)
-0.03
(-0.20-0.13)
0.00
(-0.03-0.02)
0.01
(-0.05-0.07)
0.02
(-0.05-0.08)
-0.01
(-0.14-0.12)
-3.11
(-8.62-2.40)
-0.72
(-12.58-11.15)
-0.73
(-3.34-1.87)
-6.32
(-12. 15- -0.49)
0.00
(-0.03-0.03)
0.06
(0.00-0.12)


Adjusted
R2
0.21
0.16
0.15
0.15
0.27
0.32
0.19
0.09
0.32
0.17
0.35
0.22
0.20
0.09
0.25
0.21
0.30
0.14
0.29
0.22
0.30
0.29
all 4
criteria
(relaxed)
fulfilled?
yes
yes
yes
yes
yes
yes
yes
no
yes
yes
yes
yes
yes
no
yes
yes
yes
no
yes
no
yes
no
all 4
criteria
(strict)
fulfilled?
no
no
no
no
no
no
no
no
no
no
yes
no
no
no
no
no
no
no
no
no
no
no
                                                   166

-------
Table C.3. Shown are 1) a summary of thresholds, from all analyses, for the chlorophyll o,
AFDM, TN, and TP gradients, and 2) mean and distributions of All values among sites that had
gradient values below and above the indicated threshold. In addition, interquartile ranges
(IQRs) of the All distributions are provided (see Figure 3.21 for example graphical depictions
based on piecewise-regression-derived thresholds). To provide perspective on where each
threshold lies relative to the full range of corresponding gradient values across the data set as a
whole, maximum values are as follows: Chlorophyll a = 1504 mg/m2, AFDM = 405 g/m2, TN =
26.4 mg/L, TP = 5.4 mg/L.  Note that in most cases, the threshold is far below the maximum for
the gradient in question. As such, when normalized for the proportion of the range of gradient
values represented by each bin (i.e., "below" vs. "above" the threshold), the IQR is, relatively
speaking, greater for the sites with gradient values below the threshold than for those above.
Table C.3.
0)
Sf
_Q
CD
Gradient jj All
BMIcommunity
BMIcommunity
BMIcommunity
BMIcommunity
BMIcommunity
BMIcommunity
BMIcommunity
BMIcommunity
CSCI
CSCI
CSCI
CSCI
AFDM (g/m2) BMI EPT_Percent
EPT_Percent
EPT_Percent
EPT_Percent
EPT_Percent
EPT_Percent
EPT_PercentTaxa
EPT_PercentTaxa
EPT_PercentTaxa
EPT_PercentTaxa
EPT_Taxa
EPT_Taxa
EPT_Taxa
EPT_Taxa
lntolerant_Percent
, Intolerant Percent
AFDM (g/m ) BMI
lntolerant_Percent
lntolerant_Percent
lntolerant_PercentTaxa
analysis type
CART
CART
ncpa.bc
ncpa.bc
ncpa.euc
ncpa.euc
TITAN. decreasers
TITAN. decreasers
piecewiseregression
piecewiseregression
SiZer
SiZer
BRT_exhaustion
BRT_exhaustion
piecewiseregression
piecewiseregression
SiZer
SiZer
piecewiseregression
piecewiseregression
SiZer
SiZer
piecewiseregression
piecewiseregression
SiZer
SiZer
piecewiseregression
piecewiseregression
SiZer
SiZer
BRT_exhaustion
threshold
30.84
30.84
10.86
10.86
11.42
11.42
7.05
7.05
15.02
15.02
6.96
6.96
25.00
25.00
16.10
16.10
3.94
3.94
18.93
18.93
5.92
5.92
17.47
17.47
7.92
7.92
12.99
12.99
5.94
5.94
12.00
gradient
value
relationship
to threshold mean
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
-0.19
0.47
-0.35
0.35
-0.34
0.36
-0.41
0.22
0.92
0.70
0.97
0.76
0.41
0.18
0.43
0.19
0.56
0.31
0.44
0.23
0.50
0.32
15.17
5.67
17.13
8.17
0.20
0.05
0.25
0.08
0.30
percentile
min
-1.39
-0.88
-1.39
-0.96
-1.39
-0.96
-1.39
-1.21
0.21
0.10
0.21
0.10
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
50*
25* (median)
-0.65
0.05
-0.74
-0.01
-0.74
0.00
-0.80
-0.17
0.80
0.51
0.91
0.58
0.22
0.01
0.26
0.01
0.44
0.09
0.33
0.10
0.46
0.17
9.00
1.75
12.00
2.00
0.05
0.00
0.12
0.00
0.18
-0.25
0.47
-0.39
0.32
-0.39
0.32
-0.44
0.20
0.98
0.68
1.02
0.77
0.42
0.11
0.45
0.12
0.59
0.29
0.49
0.22
0.54
0.33
17.00
4.00
19.00
6.00
0.19
0.00
0.25
0.01
0.32
75*
0.21
0.82
-0.04
0.70
-0.01
0.71
-0.12
0.65
1.09
0.93
1.11
0.98
0.60
0.34
0.62
0.33
0.71
0.47
0.57
0.35
0.59
0.47
21.00
8.00
23.00
13.00
0.30
0.04
0.37
0.13
0.43
max
1.64
1.63
1.62
1.64
1.62
1.64
1.62
1.64
1.27
1.22
1.27
1.23
0.92
0.77
0.90
0.92
0.86
0.92
0.75
0.67
0.75
0.72
34.00
29.00
34.00
29.00
0.72
0.61
0.72
0.61
0.62
IQR
0.86
0.76
0.70
0.70
0.72
0.71
0.68
0.82
0.29
0.41
0.20
0.40
0.38
0.33
0.36
0.32
0.27
0.38
0.23
0.24
0.13
0.30
12.00
6.25
11.00
11.00
0.26
0.04
0.25
0.13
0.25
                                           167

-------
Table C.3.

-------
Table C.3.

-------
Table C.3.
Gradient
Chlorophyll a
(mg/m2)


























Chlorophyll a
(mg/m2)









-------
Table C.3.

-------
Table C.3.
0)
U)
ro
-Q
1
Gradient JJ All
RAWpropBiovolZHR
RAWpropBiovolZHR
RAWpropGreenCRUS
RAWpropGreenCRUS
RAWpropGreenCRUS
RAWpropGreenCRUS
S2
S2
S2
S2
BMIcommunity
BMIcommunity
BMIcommunity
BMIcommunity
TN (mg/L) BMI BMIcommunity
BMIcommunity
BMIcommunity
BMIcommunity
CSCI
CSCI
CSCI
CSCI
CSCI
CSCI
CSCI
CSCI
EPT_Percent
EPT_Percent
EPT_Percent
EPT_Percent
EPT_Percent
-TM / /, \ n»n EPT Percent
TN (mg/L) BMI
EPT_PercentTaxa
EPT_PercentTaxa
EPT_PercentTaxa
EPT_PercentTaxa
EPT_Taxa
EPT_Taxa
EPT_Taxa
EPT_Taxa
lntolerant_Percent
lntolerant_Percent
lntolerant_Percent
lntolerant_Percent
lntolerant_PercentTaxa
lntolerant_PercentTaxa
lntolerant_PercentTaxa



analysis type
SiZer
SiZer
piecewiseregression
piecewiseregression
SiZer
SiZer
piecewiseregression
piecewiseregression
SiZer
SiZer
CART
CART
ncpa.bc
ncpa.bc
ncpa.euc
ncpa.euc
TITAN. decreasers
TITAN. decreasers
BRT_exhaustion
BRT_exhaustion
BRT_resistance
BRT_resistance
piecewiseregression
piecewiseregression
SiZer
SiZer
BRT_exhaustion
BRT_exhaustion
piecewiseregression
piecewiseregression
SiZer
SiZer
piecewiseregression
piecewiseregression
SiZer
SiZer
piecewiseregression
piecewiseregression
SiZer
SiZer
piecewiseregression
piecewiseregression
SiZer
SiZer
BRT_exhaustion
BRT_exhaustion
piecewiseregression



threshold
42.72
42.72
103.37
103.37
19.42
19.42
113.46
113.46
19.42
19.42
0.29
0.29
0.32
0.32
0.32
0.32
0.20
0.20
0.80
0.80
0.30
0.30
0.74
0.74
0.14
0.14
0.60
0.60
0.59
0.59
0.27
0.27
0.68
0.68
0.14
0.14
0.63
0.63
0.14
0.14
0.54
0.54
0.14
0.14
0.55
0.55
0.62

gradient
value
relationship
to threshold
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below
above
below



mean
0.38
0.12
0.25
0.71
0.16
0.50
57.80
27.38
63.46
42.55
-0.36
0.36
-0.34
0.40
-0.34
0.40
-0.42
0.30
0.92
0.64
0.98
0.69
0.93
0.63
0.99
0.77
0.42
0.18
0.42
0.18
0.46
0.21
0.45
0.18
0.51
0.29
15.38
3.00
18.46
7.77
0.19
0.01
0.24
0.07
0.29
0.03
0.28



min
0.00
0.00
0.00
0.00
0.00
0.00
0.00
2.00
0.00
0.00
-1.39
-1.34
-1.39
-1.34
-1.39
-1.34
-1.39
-1.34
0.21
0.10
0.36
0.10
0.21
0.10
0.36
0.10
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.15
0.00
0.00
0.00
4.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00



25th
0.00
0.00
0.00
0.40
0.00
0.00
40.00
13.00
52.00
20.00
-0.74
0.00
-0.71
0.02
-0.71
0.02
-0.78
0.09
0.80
0.49
0.89
0.52
0.81
0.48
0.92
0.60
0.25
0.00
0.25
0.00
0.30
0.02
0.35
0.08
0.46
0.16
9.00
1.00
14.00
2.00
0.05
0.00
0.12
0.00
0.17
0.00
0.16
percentile

50*
(median)
0.22
0.00
0.00
0.95
0.00
0.57
62.00
20.00
67.00
37.00
-0.39
0.35
-0.35
0.43
-0.35
0.43
-0.45
0.27
0.98
0.63
1.02
0.66
0.99
0.63
1.03
0.77
0.42
0.07
0.42
0.07
0.47
0.13
0.48
0.17
0.53
0.28
16.00
2.00
19.00
6.00
0.17
0.00
0.24
0.00
0.30
0.00
0.30



75th
0.80
0.09
0.51
1.00
0.00
0.99
78.00
35.00
78.00
65.00
-0.02
0.77
0.04
0.79
0.04
0.79
0.08
0.67
1.08
0.77
1.10
0.85
1.08
0.76
1.11
0.97
0.59
0.29
0.59
0.29
0.63
0.36
0.56
0.25
0.59
0.41
21.00
4.00
23.00
12.00
0.30
0.00
0.35
0.11
0.41
0.05
0.41



max
1.00
1.00
1.00
1.00
1.00
1.00
100.00
87.00
100.00
97.00
1.12
1.64
1.12
1.64
1.12
1.64
1.12
1.64
1.27
1.20
1.27
1.26
1.27
1.20
1.27
1.26
0.92
0.90
0.92
0.90
0.92
0.90
0.75
0.55
0.75
0.67
34.00
17.00
34.00
27.00
0.72
0.53
0.72
0.61
0.62
0.33
0.62



IQR
0.80
0.09
0.51
0.60
0.00
0.99
38.00
22.00
26.00
45.00
0.72
0.77
0.75
0.77
0.75
0.77
0.86
0.58
0.28
0.28
0.20
0.34
0.27
0.28
0.18
0.37
0.34
0.29
0.34
0.29
0.33
0.34
0.21
0.17
0.13
0.25
12.00
3.00
9.00
10.00
0.24
0.00
0.23
0.11
0.25
0.05
0.24
172

-------
Table C.3.

-------
Table C.3.

-------
Table C.3.

-------
Table C.3.

-------
Table C.3.

-------
Table C.3.

-------
Table C.4. Summary of recommended numeric endpoints for stream NNE
indicators, by beneficial use, from Tetra Tech (2006).
Beneficial Use Risk Category I. Presumptive unimpaired (i.e., the beneficial use is supported)
Beneficial Use Risk Category II. Potentially impaired (i.e., the site may require an impairment
assessment)
Beneficial Use Risk Category III. Presumptive impaired (i.e., the beneficial use is not supported or is
highly threatened)
BURC
Response Variable Boundary
Benthic algal biomass- l/ll
max
(mg chlorophyll o m"2) "'"
Dissolved oxygen - \/\\
mean of 7 daily min.
(mgL-1) "/III
pH maximum — I/M
photosynthesis-driven n/m

COLD
100

150
9.5

5.0
9.0
9.5

WARM
150

200
6.0

4.0
9.0
9.5

REC-1
C

C
A

A
A
A

REC-2
C

C
A

A
A
A

MUN
100

150
A

A
A
A

SPWN
100

150
8.0

5.0
C
A

MIGR
B

B
C

C
C
A
A - No direct linkage to the beneficial use
B - More research needed to quantify linkage
C - Addressed by Aquatic Life Criteria
                                             179

-------
Appendix D. Graphics and Tables Supporting Evaluation of
   Benthic Biomass Response Models the NNE Benthic Biomass
   Spreadsheet Tool
Figure D.I. Comparison of the NNE stations (3053) to the daily precipitation station from NOAA
(981, July 2012).
• NNE Stations
• Daily precipitation stations (NOAA)
                                 180

-------
Figure D.2. Comparison of the predicted precipitation between the PRISM and ARCGIS (predicted for
this study). A good fit was observed for the predicted precipitation between the PRISM model and the
ARCGIS model.
                                                 s
• Precipitation predicted by PRISM Model
Precipitation predicted by ARCGIS model
created for this study
                                              181

-------
Figure D.3. Ranking by variable importance for the Dodds 97 model for AFDM using random forests.


                                    MeanAlgalDen_Dodds97
       Monthly Precip 3 mo
       CODE 21 WS
       Fine Substrate
       URBAN WS
       Turbidity
       Solar Radiation 3 mo
       Road Density WS
       Catchment slope 1km
       Light Extinction
       Ag WS
       Geolgy Quarternary
       TurbidityNTU
       Catchment Slope WS
       Sample Site Elevation
       TKN mg L
       Strahler Order
       Ag 1 Km
       Alkalinity mg L
       Ammonia mg L
       Mean width wet channel
       MaxW1 HALL
       Temperature
       Reach Slope
       TN mg I
       AREAKm2
       Solar Radiation same mo
       URBAN 1K
       Sedimentary Geology-
       Geology Volcanic
       Measured Water Depth
    o

   o
                                            %lncMSE
                                    MaxAlgalDen_Dodds97
  TP mgL

  OPO4 mg L

  Fine Substrate

  TN mg I

  Conductivity uS cm

  Solar Radiation same mo

  Reach Slope

  Sample Site Elevation

  Road Density 1km

  NOX mg L

  Substrate sand size

  CPOM

  CODE 21  WS

  Mid Channel Shade Canopy Cover

  Mean width wet channel
                                        i

                                       10
 i

15
 !

20
 i

25
 !

30
                                            %lncMSE
                                               182

-------
Figure D.4. Ranking by variable importance for the Dodds 02 model for AFDM using random forests.


                                     MeanAlgalDen_Dodds02
      URBAN WS
      Monthly Precip 3 mo
      Fine Substrate
      CODE 21 WS
      TurbidityNTU
      Catchment slope 1km
      Road Density WS
      Light Extinction
      Turbidity
      Max Air Temp 3mo
      Strahler Order
      Measured Water Depth
      Solar Radiation same mo
      Solar Radiation 3 mo
      Catchment Slope WS
      Precip same Mo
      Geolgy Quarts rnary
      Days of Accrual
      Mean Depth wet channel
      Alkalinity mg L
      Ag 1 Km
      Ag WS
      Mid Channel Shade Canopy Cover
      Mean width wet channel
      Road Density 1km
      Geolgynon sedimentary volcanic
      Geology Volcanic
      Sampfe Site  Elevation
      TPmgL
      AREA Km 2
             o-
             o
          -o-
          -(J-
          o

          o
        o
       o
      o
      o
     o
    o
   -o-
 o
 o
 a
o
                                   r
                                             %lncMSE
                                      MaxAlgalDen  Dodds02
     TPmgL

     TNmgl

     NOX mg L

     Fine Substrate

     OP04 mg L

     Reach Slope

     Conductivity uS cm

     URBAN 1K

     CPOM

     Chloride mg L

     Solar Radiation same mo

     Substrate sand size

     Mean width wet channel

     Mid Channel Shade Canopy Cover

     Ammonia mg  L
       o


       o
 o


o
                                   5     10     15     20     25


                                             %lncMSE
                                                 183

-------
Figure D.5 Ranking by variable importance for the QUAL2K model for AFDM using random forests.
                         standardQua!2k_MaxAlgaeDensity
                       revisedQual2k_MaxAlgaeDensity

NOX mg L
Fine Substrate


Conductivity uS cm
unionoerngL
UrU4 mg L
TP mgL
water lemp


Mean width wet channel

HOW
Solar Radiation 3 mo

















i i i
10 20 30 4

Fine Substrate
IN mg I


URBAN 1K
Temperature
TPrng L
Conductivity uS cm
UrU4mg L


Solar Radiation same mo

Mean width wet channel
Flow
]
















I I I I
5 10 15 20
                                    %lnch1SE
                                                                                             %lncMSE
                                                    revisedQua!2k_MaxAlgaeDensity_accrual
                                Fine Substrate
                                TNmgl
                                Max Air Temp same mo
                                Geolgy Cenoz
                                Max Air Temp 3mo
                                Mean Depth wet channel
                                GeolgyQuarternary
                                Road Density WS
                                Measured Water Depth
                                Turbidity NTU
                                Monthly Precip 3 mo
                                AREA Km2
                                Solar Radiation same mo
                                Geolgy non sedimentary volcanic
                                URBAN WS
                                Reach Slope
                                Ammonia mgL
                                Cloud Cover same mo
                                Alkalinity mg L
                                Days of Accrual
                                URBAN 1K
                                OP04mgl_
                                Catchment Slope WS
                                Geology Volcanic
                                Substrate sand size
                                MinAirTemosameMo
                                Sample Site Elevation
                                Cloud Cover 3 mo
                                Light Extinction
                                Solar Radiation 3 mo
•o

0-

0
                                                                   %lncMSE
                                                           184

-------
Figure D.6 Ranking by variable importance for the Dodds 97 for chlorophyll a using random forests.
                                 MeanBenthlcChlA Dodds97
        Reach Slope

        NOX mg L

        TNmgl

        CODE 21 WS

        TPmgL

        Catchment slope 1 km

        Chloride mg L

        Alkalinity mg L

        Solar Radiation same mo

        OP04 mg L

        Substrate sand size

        Conductivity uS cm

        MaxW1 HALL

        Water Temp

        Temperature
 o

o
                                     5
                   10
15
                                          %lncMSE
                                  MaxBenthicChIA Dodds97
   Ag_ 1 Km
   CPOM
   MaxAirTemp 3mo
   Mean width wet channel
   Mid Channel Shade Canopy Cover
   Alkalinity mg L
   URBAN 1K
   Geolgy Quarternary
   URBANWS
   Ammonia mg L
   Road Density 1km
   Substrate  sand size
   CODE 21 IKm
   Solar Radiation 3 mo
   AgWS
   OP04 mg L
   Min Air Temo same Mo
   Flow
   Days of Accrual
   Sample Sjte Elevation
   Conductivity uS cm
   Catchment Slope WS
   TKN mg L
   Chloride mg L
   Sedimentary Geology
   MaxAirTemp same mo
   Geolgy non sedimentary volcanic
   Catchment slope 1 km
   Mean Depth wet channel
   AREA Km2
                    o
                    o
           o
          o
         o
        o
        o
       o
       o
       o
      o
    o
    o
    o
    o
    o
   o
                                      246


                                            %lncMSE
                                              185

-------
Figure D.7. Ranking by variable importance for the Dodds 02 for chlorophyll a using random forests.
                                MeanBenthicChlA  Dodds02
  AQ 1 Km
  MaxAirTernp 3mo
  Mid Channel Shade Canopy Cover
  CPOM
  Alkalinity mg L
  Mean width wet channel
  Substrate  sand size
  Flow
  AgWS
  Geolgy Quarternary
  Chloride mg L
  AREA Km2
  Ammonia mg L
  Monthly Precip 3 rno
  Geolgy Cenoz
  Cloud Cover same mo
  CODE 21 1Km
  Min AirTemo same Mo
  Conductivity uS cm
  URBAN 1K
  Solar Radiation 3 mo
  Turbidity NTU
  Road Density WS
  Fine Substrate
  Geolgy non sedimentary volcanic
  Catchment slope 1 krn
  Catchment Slope WS
  Days of Accrual
  Reach Slope
  Road Density 1km
                              o
                             o
           o
          o
     o
     o
    o

    o
    o
   o
   o
   o
  o
  o
 o

o
                                           %lncMSE
                          MaxBenthicChIA Dodds02
   TPmgL
   TNmgl
   NOX mg L
   OP04 mg L
   Substrate sand size
   Alkalinity mgL
   Conductivity uS cm
                         10
   15
20
25
                                  %lncMSE
                                                 186

-------
Figure D.8. Ranking by variable importance for the QUAL2K models for chlorophyll o using random

forests.
                              standardQuaEk BenthicChlora
                                               RevisedQual2k  BenthicChlora


Road Density 1 km

TKNrngL
rrnjL
Chloride mg L
NOX mg L
Alkalinity mg L
OP04 mg L
Ammonia mj L
Conductivity uS cm
Substrate sand size
Reach Slope
Temperature
Water Temp















0
I 1 I I
-5 0 5 10


URBAN 1 K
r fnij L
Road Density 1km
iwx mg L
Substrate sand size
URBAN WS
UrU4 rnCJL
Chloride mg L
Alkalinity mgL
Conductivities cm
Temperature
water lemp
AREAKinJ
Geolgy Quarternary















0
i I I I I
0 5 10 15 20
                                      %lncMSE
                                                        %lncMSE
                                 benthic_qua!2kaccrual
  CODE 211Km
  Chloride rng L
  Max Air Temp 3mo
  Substrate sand size
  Alkalinity rng L
  Mid Channel Shade Canopy Cover
  TKN mg L
  Ag1 Km
  Mean width wet channel
  TPmgL
  CPOW
  Catchment slope 1km
  Flow
  URBAN 1K
  Monthly Precip 3 mo
  Conductivity uS cm
  Solar Radiation 3 mo
  Max Air Temp same mo
  Road Density 1km
  AREAKm2
  Mean Depth wet channel
  Measured Water Depth
  Fine Substrate
  NOX mg L
  Ammonia mgL
  Road Density WS
  MaxW1 HALL
  Strahler Order
  Geolgy Cenoz
  Catchment Slope WS
              o
              o
             o
        0

       0
     0

     0

     o

     0

     0

     0
   0

   0

   0

  0

  o

  o
  o
 o
 o
o
                                        %lncMSE
                                                      187

-------
Table D.I Details on the models. PCT_MAP is macroalgal percent cover, PCTJV1CP is macrophyte percent
cover, and PCT_MIAT1 is percent presence of thick (lmm+) microalgae. Dashes correspond to predictors
that were not included in the final model for the biomass response variable in question.
Predictor
canopy cover (%)
sand & fines (%)
conductivity
fines (%)
stream
CODE_21_2000_5K
latitude
coarse particulate
discharge
Ag_2000_WS
alkalinity
slope, reach
PH
ecoregion
NOx
NH4
mean monthly max
turbidity
longitude
SRP
watershed area
days of accrual
CODE_21_2000_WS
TN
W1_HALL (riparian
TP
elevation
site disturbance
stream width
sedimentary
URBAN_2000_WS
mean monthly solar
URBAN_2000_1K
mean monthly %
total precipitation
Ag_2000_5K
stream depth

3
2
2
Chlorophyll
a
.3
.4
.4

4
1
3
Q
LL.
<
07
95
09
11.9
9
9
-
2
2
4
7
1
7
-
6
2
5
2
3
4
-
-
2
3
2
-
-
-
2
4
2
2
-
-
-
-
1
.9
.5

.9

.5
.3
.9
.7


.8

.8
.7
.1


.4
.8
.2



.5
.8
.8
.1




.6
2
4
3
3
5
2
1
6
2
4
-
7
1
4
3
3
1
2
2
1
3
-
-
-
3
-
1
2
-
1
1
-
1
03
21
86
11
63
86
21
78
2
82

7
23
29
15
83
37
55
26
38
28



99

17
1

18
34

47
Soft Algal
Total
Biovolume
7.13
4.84
14.81
5.15
8.27
2.17
10.68
-
1.56
-
2.45
2.54
2.51
3.72
3.21
2
-
2.44
3.08
2.08
2.17
2.16
2.68
1.51
1.98
3.31
1.24
-
-
1.67
-
2.12
-
2.52
-
-
-
Q_
<
d'
Q.
13.4
6.63
12.7
2.5
2.73
-
1.11
2.68
2.59
-
5.96
5.91
2.84
1.36
3.17
2.25
1.88
2.93
0.76
1.46
2.44
1.38
-
2.73
1.39
1.49
4.05
0.73
1.63
1.13
-
2.01
2.1
1.26
1.5
1.34
1.97
Q.
U
d'
Q.
2.23
17.4
1.01
7.8
1.39
-
3.39
14.5
5.34
-
2.42
2.86
1.75
3.55
1.16
1.45
2.06
1.94
3.44
0.77
3.8
3.16
-
1.8
0.88
0.72
0.96
3.64
0.82
1.47
-
1.49
1.24
1.09
1.2
2.35
0.9

6
3
1
1
9
-
5
1
6
-
2
1
4
3
2
3
5
4
1
3
2
3
-
2
3
3
2
-
1
1
-
1
2
2
1
0
1
iH
5
G'
Q.
8
6
2
3
2

7
2
9

5
5

6
9
1
8
6
9
6
6
1

7
7
4
6

2


7
2
8
9
8
2
Mean
Relative
6.16
6.13
5.86
5.73
5.58
5.28
4.94
4.87
4.00
3.66
3.65
3.59
3.51
3.41
3.27
3.22
3.19
3.16
2.67
2.62
2.48
2.47
2.44
2.31
2.24
2.23
2.21
2.19
2.03
2.00
1.97
1.91
1.83
1.76
1.49
1.48
1.43
Influence





































                                              188

-------
Detailed  Explanations of Bayesian CART Analysis
Bayesian CART Analysis Approach

Chipman et al  (2002) have developed a Bayesian approach to CART which allows the user to specify a prior
probability distribution p(O, T), where 0  represents the regression model parameters, 7 represents the tree
structure and  p(O, T ) = p(O  / 7 )p(T ). The BCART program uses the prior probability distributions to
determine the chance of splitting each node and selects from a distribution of model parameters (P,o) to
produce a set of candidate trees. Four parameters are used to characterize prior probability distributions:

       Pr(node splits | depth = d) = a(l+d)"3

where a determines the number of final nodes, depth represents the order of splits or "level" in the tree,
and B determines the shape of the tree ("bushiness").

Chipman et al. (2002) suggest the following default values: a = 0.5 and (3 = 2. The user also specifies a prior
distribution for model coefficients and the residual variance.  Chipman et al. suggest trying two  bracketing
values for normalized regression coefficients (c =  1 and 3), where smaller c values result in  estimated
coefficients that are shrunk towards 0 and smaller trees. Chipman et al. also suggest bracketing values for the
fourth model parameter describing variability, of 0.404 s2 and 0.1173 s2, where s2 is the residual variation. In
practice,  each regression tree  model is  run four times to cover the  span  of  suggested a priori tree
coefficients, and the tree with the largest log-likelihood value  (minimum Aikake criterion)  is  selected. To
avoid overfitting and facilitate selection  of a robust solution, Chipman  et al. suggest choosing  the  "most
visited" tree among the multiple iterations rather than the overall "best" fitting tree.

The CGMIidCART program allows the user to specify a training data set to fit the regression tree models and a
test data set to provide  an independent  validation of the models. We selected a random subset of values
representing approximately 10% of the full data set (n=57) as a test set.  In the first round of analyses, we
included a full  suite of potential classifier variables. In addition to the classification variables in Table 3.2, we
added interaction  terms  for turbidity x depth, stream power (watershed area x slope), stream power x
antecedent precipitation, and stream power x antecedent precipitation x % sands and fines  (an index of
potential substrate disturbance). To test the robustness of the original regression tree results (Dodds-type,
TNTP), we repeated the analysis an additional nine times with different random subsets for the training and
test sets and then chose a subset of classifiers (reduced set) based on their frequency of selection in the full
models.

The ten runs of the Dodds-type TNTP model with full set of classifiers yielded trees of various sizes, i.e., from
9 to 21 final nodes (Table D.2). The first run yielded a relatively high explanatory power for the training test
set (r2 = 0.84), with only a slightly lower value for the validation test set  (r2 = 0.80). Model fit based on AIC
was even better for the Dodds-type DINDIP model (training r2 = 0.91, test r2 = 0.88; Table D.2).
                                               189

-------
Table D.2. Results of Bayesian CART analysis of full data set using all potential classification variables
(see Table  .2 and Section 4). Predictor variables included TN, TP, TN2, TP2, days accrual, and days
accrua!2  (TNTP models) or the same variables with DIN and  DIP substituted for TN and TP (DINDIP
models).
Independent
Variable
Type
DINDIP
TNTP
TNTP
TNTP
TNTP
TNTP
TNTP
TNTP
TNTP
TNTP
TNTP
# Independent
regression
variables
6
6
6
6
6
6
6
6
6
6
6
CGM tree fitting
parameters
C
1
1
1
1
1
1
3
1
1
1
1
Variance
.117s2
.404s2
.404s2
.404s2
.404s2
.404s2
.404s2
.117s2
.117s2
.404s2
.404s2
Training Log
Set Likelihood
1
1
2
3
4
5
6
7
8
9
10
955.081708
887.325251
942.584144
1049.597041
1082.228125
912.540841
840.74299
973.827137
947.849706
886.811485
1032.605069
Most
Visited
Tree Size
17
21
21
20
27
21
9
20
19
12
20
Predicted vs.
Observed r2
AIC Training
-1672.16 0.91
-1480.65 0.84
-1591.17
-1819.19
-1786.46
-1531.08
-1555.49
-1667.65
-1629.7
-1605.62
-1785.21
Test
0.88
0.8









We chose the reduced set of classification variables based on the set of classifiers that were selected in at
least half of the full runs (Table D.3). Even though PSA ecoregion only occurred in half of the full runs, it was
retained for testing because it was potentially redundant with the latitude and longitude classifiers. Other
classifiers  in  the  reduced  set  included  NH4 (mg N/L), CODE_21_2000_5K  (a measure  of localized
urbanization), and Julian day. To avoid redundancy, NH4 was only included as a classifier in the TNTP models
but was dropped from the DINDIP models.

Table D.3. Frequency of inclusion of classification variables in Bayesian CART TNTP models (Training
sets 1-10). Only class variables occurring more than twice are listed.
Frequency
46
44
20

8

7
5
4
4
3

Classification Variable
Longitude
Latitude
NH4

CODE 21 2000 5K

JulianDay
PSAc
Year
Conductivity
REFSITESTAT

Definition
Degrees longitude
Degrees latitude
Instream NH4 value (mg NH4-N/L)
Percent NLCD "Code 21" land use within a 5-km

sampling site
Day of year (1-365)
Perennial Stream Assessment ecoregion (1-6)
Year of sample
Instream conductivity




radius from






Site disturbance status (Reference, Intermediate, Stressed) as
defined in Section 2

                                                190

-------
Table D.4. Results of Bayesian CART analysis of full data set that includes PSA ecoregion. Models used
reduced  set  of four potential  classification variables  (PSA  ecoregion  (PSAci),  Julian Day, NH4,  and
UrbanSK). Training set used is 1. Predictor variables for Dodds-type models included TN, TP, TN2, TP2,
days accrual,  and days accrua!2 (TNTP models) or the same variables with DIN and DIP substituted for TN
and TP (DINDIP models). Predictor variables for  the  QUAL2K-type models also  included temperature,
incident light, turbxdepth (see Table D.3 for definitions). Models were also run with or without latitude
and longitude as predictors. Model numbers are provided for a subset for ease of reference in the text.
Number
Independent independent
Model
No.

Model 1



Model 2
Model
Type
Dodds
Dodds
Dodds
Dodds
QUAL2K
QUAL2K
Variable
Type
TNTP
TNTP
DINDIP
DINDIP
DINDIP
DINDIP
Lat/Long
included
No
Yes
No
Yes
No
Yes
regression
variables
6
8
6
8
9
11
Log
Likelihood
583.02
736.17
473.53
659.40
462.76
635.90
Most
Visited
Tree
Size
5
7
2
4
2
4
Predicted vs.
Observed r2
Final Classification
AIC
-1096.05
-1346.33
-919.064
-1246.8
-885.526
-1175.79
Training
0.51
0.79
0.32
0.7
0.34
0.69
Test
0.44
0.81
0.23
0.63
0.29
0.66
Variables
JulDay NH4
JulDay NH4
Urban
JulDay PSAci
Urban
JulDay Urban
Urban
Urban




Table D.5. Results of Bayesian CART analysis of full data set that uses an empirical rather than PSA
ecoregion.  Models used  include a  reduced set of five  potential classification variables (Latitude,
Longitude, Julian  Day, NH4,  and UrbanSK). Training set used is 1. Predictor variables for Dodds-type
models included TN, TP, TN2, TP2, days accrual, and  days accrua!2  (TNTP models)  or the same variables
with DIN and DIP substituted for TN  and TP (DINDIP models). Predictor variables for the QUAL2K-type
models also included temperature, incident light, turbidity x depth. Models were also run with or without
latitude and longitude as predictors. Model numbers are provided  for a subset for ease of reference in
the text.

Model
No.

Models



Model 4

Model
Type
Dodds
Dodds
Dodds
Dodds
QUAL2K
QUAL2K

Independent
Variable Type
TNTP
TNTP
DINDIP
DINDIP
DINDIP
DINDIP

Lat/ Long
included
No
Yes
No
Yes
No
Yes
Number
independent
regression
variables
6
8
6
8
9
11

Log
Likelihood
1162.23
1249.05
1030.95
1120.95
952.50
1033.65
Most
Visited
Tree
Size
21
23
21
19
16
18
Predicted vs.
Observed r2
Final Classification
AIC
-2030.45
-2084.1
-1767.9
-1899.91
-1585
-1635.3
Training
0.91
0.92
0.84
0.93
0.85
0.9
Test
0.79 Lat
0.57 Lat
0.71 Lat
0.81 Lat
0.66 Lat
0.72 Lat
Variables
Long
Long
Long
Long
Long
Long
JulDay
JulDay
JulDay
JulDay
JulDay
JulDay
NH4
NH4
Urban
Urban
Urban
Urban
                                                191

-------
       y=0.8423x + 6E-06
         R2 = 0.8437
                                                          y=0.9715x-0.0103
                                                             R2 = 0.8024
                                                 0.6   -0.5   -0.4   -0.3    -0.2
                       0.5
         y=0.9088x + 9E-07   0.4
            R2 = 0.9103
                       0.3
y= 0.9984X + 0.0051    0.3
   R2 = 0.8839
                0.2
                                                 0.6   -0.5   -0.4   -0.3    -0.2
                                                                                     0.1    0.2    0.3   0.4
Figure D.9. Predicted versus observed normalized log™ chlorophyll a biomass (mg/m2) for a) TNTP
training set, b) TNTP test set, c) DINDIP training set,  and d) DINDIP test sets used in  Bayesian CART
analysis (Dodds-type model, all potential  classifiers). The  line represents the fit of a linear regression
predicting "predicted benthic algal biomass (chl a)"  as a function  of observed benthic  biomass (loglO
chlorophyll a).
                                                   192

-------
Appendix E. Quality Assurance/Quality Control Summary
Research conducted to produce this report followed guidelines specified in an approved Quality Assurance
Plan (QAPP-AED-WDB-ND-2010-r2-01). The original water quality, biotic,  and  habitat data  used  in these
analyses were  not collected by this project but as part of existing California state or regional monitoring
programs with existing approved  QA plans  (see Sections 2.2.2, 3.2.4, and 4.2.3). The quality assurance
parameters for the California datasets used are based on those established for the Surface Water  Ambient
Monitoring Program (SWAMP  2008). General  Quality Objectives for the state monitoring  programs are
described in Element A7 of SWAMP (2008):

SWAMP seeks to meet the following four objectives:
    •   Create an ambient monitoring program that addresses all of California's hydrologic units
       using consistent and objective monitoring, sampling,  and analytical methods; consistent
       data quality assurance (QA) protocols; and centralized data management.

    •   Document ambient water quality conditions in potentially clean and polluted areas.  The
       scale for these assessments ranges from site-specific to statewide.

    •   Identify specific water quality problems preventing the State Board, the Regional Boards,
       and the public from realizing beneficial uses of water in targeted watersheds.

    •   Provide data to evaluate the overall effectiveness of regulatory water quality programs in
       protecting beneficial uses of California's waters.
By definition, if the general Quality Objectives above are met, then the data collected under these monitoring
programs should be of sufficient quality to meet the needs of the current project.

Measurement  Quality  Objectives  for  the   SWAMP  monitoring  programs   are   available   at
http://www.swrcb.ca.gov/water issues/programs/swamp/mqo.shtml. The changes in MQOs between 2008
and 2013 alluded to on that web site do not apply to any of the parameters included in the present study.
Standard Operating Procedures used in the collection and processing of the samples under the established
monitoring plans  include Ode (2007) for stream benthic macroinvertebrates and habitat parameters  and
Fetscher  et  al.  (2009)  for  benthic stream  algae. QAQC protocols  for  bioassessment methods were
supplemented with guidance found in QA Memos 1 and 2 (see
http://www.swrcb.ca.gov/water issues/programs/swamp/mqo.shtml)

Probability-based  sampling frameworks used by the State of California Perennial Stream Assessment (PSA)
and the southern California Stormwater Monitoring Coalition (SMC) monitoring programs are described in
Section 2.2.2  of this report. Although each monitoring program was designed separately  to assess  the
condition of perennial wadeable streams in California, the geographic scope of each differed so new sample
weights had to be assigned when these data sets were combined. The calculation  of adjusted sample weights
used in creating composite cumulative probability distributions is also described in that section.

Objectives for targeted sampling frameworks differed from probability samples. These data from targeted
sites come from the state's Reference Condition Management  Program (RCMP) and a recently completely

                                              193

-------
project  geared  toward developing stream  algal assemblage data  for use  in bioassessment  of  stream
condition. Selection criteria for these monitoring programs are described in Ode and Schiff (2009)

In some cases,  subsets of  data  from the  monitoring programs were  used based  on the  degree of
anthropogenic disturbance associated with sites. Selection criteria for these disturbance  classes are  also
described in Section 2.2.2 of this report.

In general, the  data  collected under the four  monitoring  programs described above  were used for the
intended purpose, i.e., to describe the ambient and/or reference condition of perennial wadeable streams in
California or regions thereof and to determine  whether designated  uses were being met. As discussed in
Section 2.2, the sampling window for California's bioassessment programs has  been chosen  to maximize the
chances of yielding complete samples across a range of wet and dry years.  It  was  not chosen to assess the
temporal variability in benthic algal biomass or to necessarily capture the maximum values of benthic algal
parameters. However, given the need of the current study to assess the relationship between benthic algal
biomass and community composition, it was appropriate to use samples collected concomitantly.

With the exception of a few field duplicates, the data available for the CA state monitoring programs do not
include time series of stream  nutrient concentrations or measurements of the full suite of nutrient forms. To
assess the representativeness of instantaneous samples of nutrients collected during the growing season for
state bioassessment monitoring programs, we analyzed  data for 60 California NAWQA stations sampled at a
fixed frequency  over the  year. We  downloaded data from the  USGS NAWQA Data Warehouse  (U.S.
Geological Survey 2001).

Table E.I. Nutrient fractions for samples from 47 USGS NAWQA stream  stations in California
sampled biweekly over the year.

Fraction dissolved inorganic N
Fraction dissolved N
Fraction particulate N
Fraction soluble reactive P
Fraction dissolved P
Fraction particulate P
Average
0.53
0.89
0.15
0.75
0.69
0.30
Minima
0.05
0.57
0.01
0.25
0.25
0.00
Maxima
0.97
1.20
0.33
1.55
2.00
0.75
Std
0.24
0.13
0.10
0.27
0.28
0.22
Most of the total N and P in the CA NAWQA streams (Table E.I above) are in dissolved form, as they are in
the CA bioassessment streams. Although DIN and soluble reactive P are the most bioavailable forms, algae
can utilize dissolved organic N and P as well.  For the  NAWQA streams, maximum total N and P tend to occur
in January - February (Figure E.la,b), followed  by  declines, then an increase during the growing  season
(Figure E.2a,b). TP tends to peak in July,  while TN slowly increases from May through September. There is a
fair amount of variability among biweekly TN and TP values over the growing season, which contributes to
the noise  in the relationships we have assessed. The predictive power  of relationships could probably be
improved by averaging values for multiple water quality samples over

a)
                                               194

-------
a)
b)
                           Month of maximum TN
              FREQUENCY
                    50 -I
                       1   2   3   4  5   6   7   8   9   10  11  12

                                    month MIDPOINT
                           Month of maximum TP
              FREQUENCY
                    504
Figure E.I. Frequency of month of a) maximum annual total N and b) maximum annual total P in 47
California NAWQA streams.
                                              195

-------
a)
                          Relative TN value by month
   1.00 -
                                                                     I
         1234567
                                      month
                                                          10     11    12
b)
                          Relative TP value by month
 g 0.50 -
   0.25 -
         T
I
                               567
                                      month
                                                     9    10     11    12
Figure E.2. Value for a) total N and b) total P by month relative to maximum monthly value in
47 California IMAWQA streams.
                                                 196

-------
a)
      Relationship between annual maximum and average growing season TN
 TNmax
   40-1
   20
     01234567


                                    TNgsav
b)
       Relationship between annual maximum and average growing season TP
 TPman
    10
                    S  **
                                              -
                                               <
                                                           9     10    11
Figure E.3. Relationship between annual maximum and growing season average values for a) total N

and b) total P in 47 California IMAWQA streams.
                                               197

-------
the months preceding collection of benthic algal samples, or by using nutrient values inferred by diatom taxa
composition (Pan and Stevenson 1996). Lohman and Priscu  (1992) have documented evidence of luxury
consumption of phosphate by Cladophora (in an N-limited portion of the Columbia River) such that ambient
SRP was inversely related to cellular P content. In that same river segment, however, cellular N content of
Cladophora did track ambient dissolved inorganic N levels, the limiting nutrient. In most cases represented by
the CA NAWQA dataset, annual maxima recorded for TN and TP increase linearly in proportion to growing
season averages and thus growing season values should  represent the relative trophic condition of streams
and facilitate cross-comparisons  among systems (Figure  E.3a,  b). The few outliers in this relationship
coincided with enriched systems.

The optimum period for stream algal assessments has not  been established. A plot of stream survey data
suggest that South Coast sites may be exhibiting a peak in June-July, although this trend may be confounded
by disturbance  class of sites sampled (e.g., urban streams sampled in late season). No seasonal  peak is
obvious for the remaining sites in the State (Figure E.4). When biomass values are normalized for TN, there is
no consistent pattern of increasing biomass over the growing season  in either South Coast or other sites,
which would be captured by the "accrual" term in prediction  equations (Figure E. 5).
   co  T—  —i— —i—  —i—  —i—  —i	u —i— —i—  —i— —i—  —i— —i—    region
   2 May  Jun   Jul   Aug  Sep  Oct     May  Jun  Jul  Aug  Sep  Oct    _». non-South Coast
Q.'
O
.C
o
'ov
o
     1-
                     2010
                                                     2011
                                         3-
                                         2-
                                         1-
                                         0-
                                                                                South Coast
          Jun    Jul    Aug   Sep   Oct      Jun    Jul   Aug   Sep   Oct
                                       Date

Figure E.4. Chlorophyll a levels (log-transformed) across sampling dates, by year, for South Coast (blue)
and all other sites (red) within the state. Curves show time-averaged trends.
                                               198

-------
    120-
     80-
   40-
§  0-
                     2008
.	H
                                                      2009
                                       300-
                                     200-
                                       100-
                                         o-
                     2010
                                                      2011
>    May   Jun  Jul   Aug   Sep  Oct      May  Jun   Jul   Aug  Sep  Oct
p
TO
6
§>300-
    200-
    100-
      o-

                                       300-
                                       200-
                                       100-
                                       0J
                                region
                                •*- non-South Coast
                                — South Coast
           Jun   Jul    Aug   Sep   Oct       Jun   Jul   Aug  Sep  Oct
                                       Date

Figure E.5. Chlorophyll o levels (log-transformed) across sampling dates normalized by log total N, by
year, for South Coast (blue) and all other sites (red) within the state. Curves show time-averaged trends.
Literature Cited
Fetscher A.E., Busse L.B., and P.R. Ode. 2009. Standard Operating Procedures for Collecting Stream Algae
    Samples and Associated Physical Habitat and Chemical Data for Ambient Bioassessments in
    California. California State Water Resources Control Board Surface Water Ambient Monitoring
    Program (SWAMP) Bioassessment SOP 002. (updated May 2010)
Lohman, K. and J.C. Priscu. 1992. Physiological indicators of nutrient deficiency in Cladophora
    (Chlorophyta) in the Clark Fork of the Columbia River, Montana. Journal of Phycology, 28:443-448.
Ode P. 2007. SWAMP Bioassessment Procedures: Standard operating procedures for collecting benthic
    macroinvertebrate samples and associated physical and  chemical data for ambient bioassessment in
    California. Available from http://swamp.mpsl.mlml.calstate.edu/wp-
    content/uploads/2009/04/swamp sop  bioassessment  collection 020107.pdf
Ode P and K. Schiff. 2009. Recommendations for the Development and Maintenance of a Reference
    Condition Management Program (RCMP) to Support Biological Assessment of California's Wadeable
    Streams. SCCWRP Technical Report #581. www.sccwrp.org
                                              199

-------
Pan, Y. and R.J. Stevenson. 1996. Gradient analysis of diatom assemblages in Western Kentucky wetlands.
    Journal ofPhycology 32:222-232.
SWAMP. 2008. Quality Assurance Program Plan of the California Surface Water Ambient Monitoring
    Program.
    http://www.waterboards.ca.gov/water issues/programs/swamp/docs/qapp/qaprp082209.pdf

U.S. Geological Survey 2001. National Water Information System (NWISWeb) [Surface Water/Bed
    Sediment]: U.S. Geological Survey database, accessed July 28, 2014, at
    http://water.usgs.gov/nawqa/data.
                                               200

-------
&EPA
      United States
      Environmental Protection
      Agency


      Office of Research and Development
      National Health and Environmental^
      Effects Research Laboratory
      Atlantic Ecology Division
      Narragansett, Rl 02882

      Official Business
      Penalty for Private use
      $300

-------