U.S. Environmental Protection Agency
Office of Water/Office of Research and Development
Washington, D.C. 20460
EPA841-R-09-001a
National Lakes Assessment:
Technical Appendix
Data Analysis Approach
January 2010
-------
National Lakes Assessment: Technical Appendix
Data Analysis Approach
U.S. Environmental Protection Agency
Office of Water
Office of Research and Development
January 2010
EPA841-R009-001a
-------
National Lakes Assessment: Technical Appendix
Data Analysis Approach
Table of Contents
Overview 4
Objectives of the NLA 4
Reference Condition 5
Sources of Reference Sites 5
Screening NLA Site Data for Biological Reference Condition 5
Screening NLA Site Data for Nutrient Reference Condition 6
Biological Data 11
Data Preparation: Standardizing Counts 11
Operational Taxonomic Units 12
Sediment Diatom Metric Development 12
Metric Evaluation and Selection 12
Metric Selection, Scaling, Transformation and Calculation of Lake Diatom 13
Condition Indices 13
Plankton 0/E: Predictive (RIVPACS) Models 18
Relative Risk, Attributable Risk, and Relative Extent 20
Relative Risk (RR) 21
Attributable Risk (AR) 22
Water Chemistry Analysis 23
Trends Studies 24
Trend Analysis of National Eutrophication Survey (NES) 24
Diatom Sediment Core Analysis 28
Sediment Diatom Transfer Function 28
Calibration and Development of Transfer Functions 30
Top-Bottom Changes 31
Development of Indices for Lakeshore and Littoral Habitat Condition 33
Introduction 33
Lakeshore Disturbance Index 36
Condition Criteria for RDisJX: 37
Lakeshore Habitat Index 37
Shallow Water Habitat Index 39
Physical Habitat Complexity Index 40
Physical Habitat Index Precision and its Interpretation 41
Reference Site Screening 43
Habitat Indicator Expectations 44
Setting Condition Criteria 45
PHAB Metric Performance 46
NLA Index Precision and Interpretation 55
Quality Assurance Summary 57
References 60
-------
Overview
This document provides additional information to supplement the results and
discussion presented in the 2007 National Lakes Assessment (NLA). It is intended to
serve as a more technical reference than the report itself on the conceptual basis and
the methods and procedures used for the NLA. Although it is intended to provide a
comprehensive summary of these procedures, it is not intended to present additional
data analysis results or an in-depth report of the design, sampling, or analysis protocol.
For additional details, citations are provided.
Objectives of the NLA
The objective of the NLA is to characterize the ecological condition of the nation's
lakes throughout the conterminous United States. The NLA is an ecological
assessment of lakes based on chemical, physical, and biological data. It employs a
statistically-valid probability design stratified to allow estimates of the condition of
streams on a national and regional scale. The two key questions the NLA addresses
are
• To what degree are the Nation's lakes in good, fair, and poor condition?
• What is the relative importance of the different stressors evaluated in the
NLA?
The NLA is a collaboration among the U.S. Environmental Protection Agency
(EPA), states, tribal nations, U.S. Geological Survey (USGS), and other partners. It is
intended as a document for the public and Congress. It is not a technical document, but
rather a report geared towards a broad audience. This Technical Addendum is a
supplemental document used to support the results in the NLA report. It describes the
process used to collect, evaluate, and analyze data for the NLA. It outlines steps taken
to assess the biological condition of the nation's freshwater resources and identify the
relative impact of stressors on this condition. Results from the analysis are included in
this 2007 NLA Report; the data collected and methods described will continue to be
studied and used for future analyses.
The NLA data analysis procedures described in this technical report were
developed from the input and experience of the participating cooperators and technical
experts. A small workgroup was held in the spring of 2009 to consider approaches for
data analysis. Findings from this workshop were presented at the larger group of
cooperators and lake managers at the National Lakes Meeting in April 2009. Here,
state agencies, universities, non-profits, and EPA participated in a one day workshop
where they discussed topics such as analysis options, data presentation, and reference
sites. Discussions from this meeting were used to help define the steps taken for the
data analysis presented in the final report.
NLA analysts used two processes for establishing the good/fair/poor findings in
the NLA report. For trophic status and recreational indicators, alysts used fixed,
nationally consistent thresholds. This approach is not covered in detail in this Techincal
Addendum. The second approach was to establish regionally consistent reference-
based thresholds. Detailed information on this approach is presented below.
-------
Reference Condition
To assess current ecological condition, it is necessary to compare
measurements today to an estimate of "good" quality. Setting reasonable expectations
for each indicator was one of the greatest challenges for the NLA analysts. Because of
the difficulty in estimating historical conditions for many NLA indicators, the 2009 NLA
used "least-disturbed condition" as the reference condition. Least-disturbed condition
can be defined as the best available chemical, physical, and biological habitat
conditions given the current state of the landscape - or "the best of what's left"
(Stoddard et al. 2006). Data from reference sites were used to develop seven regional
specific reference conditions against which test results could be compared.
Sources of Reference Sites
Sites sampled during the NLA index period using consistent sampling protocols
and analytical methods were screened to meet regional specific physical and chemical
criteria. These included both sites selected from the probability sample sites and an
additional 124 hand-picked sites thought to be reference by best professional judgment.
Like the probability sample sites, the hand-picked sites were sampled using the NLA
methods. These sites were obtained from a number of sources. Some states
submitted their best reference sites to be sampled as part of the NLA while other sites
from the west and northeast were selected in a prescreening analysis utilizing landuse
to find least-disturbed lake watersheds. Regardless of whether sites were probability-
based or hand-selected, only those that met the final screening criteria were used in
developing the reference condition.
Screening NLA Site Data for Biological Reference Condition
Prior to identifying reference lakes, all lakes from the NLA were grouped into
distinct regional clusters based on nine environmental variables. This clustering was
undertaken in order to identify regional reference lakes. These variables took into
account geographic and geologic differences such as elevation, precipitation, air
temperature, longitude, latitude, and calcium concentrations. In addition to these
geographic/geologic variables, other variables such as lake area, depth, and shoreline
development were also used to segregate lakes.
Seven regional clusters were identified during this process, and these seven
regions were grouped into one of three larger regions, eastern highlands (EHIGH; which
constituted the Appalachians and the Northeast), plains and lowlands (PLNLOW; which
constituted the coastal plains, northern and southern plains, and Midwest), and western
mountains (WMNTS; which constituted the western mountains and xeric region of the
west). The PLNLOW region, which was the largest of the three combined regions, was
stratified along 40 degree latitude to insure that reference sites south of the upper
Midwest would be included in the analysis. It is important to keep in mind that the
seven regional clusters were identified to group like lakes for purposes of identifying
regional reference lakes, but are different from the NLA reporting regions. Lakes from
more than one lake cluster can and does exist within the reporting regions.
-------
To identify biological reference sites for purposes of the NLA, analysts used the
chemical and physical data collected at each site to determine whether any given site is
in least-disturbed condition for its region. In the NLA, screening values were
established for ten chemical and physical parameters to screen for reference sites.
These parameters included total nitrogen, total phosphorus, chloride, sulfate, turbidity,
euphotic zone dissolved oxygen, acid-neutralizing capacity, shoreline disturbance by
agriculture, shoreline disturbance by non-agriculture, and shoreline disturbance intensity
and extent. If a site exceeded the screening value for any one stressor, it was dropped
from reference consideration.
Given that expectations of least-disturbed condition vary across regions, the
criteria values for exclusion varied by region. The seven aggregate reference clusters
developed for the NLA used regionalize biological reference condition thresholds (Table
A-1). The first threshold value in each of the 10 screening variables, from Table A-1, is
the reference threshold. All sites in the NLA (both probability and hand-picked) that
passed all criteria were considered to be biological reference sites for the NLA (Table A-
2, Figure A-1). However, if any site exceeded one or more threshold, then it was not
considered a reference lake.
In addition to selecting biological reference sites, analysts also determine poor
quality (highly disturbed) sites that would be used in the biological assessment of the
nation's lakes. Similar to the reference selection process, thresholds were used to
determine which lakes were to be considered poor, in each of the seven cluster regions.
The second threshold value in each of the 10 screening variables, from Table A-1, is the
poor threshold. If any site exceeded the threshold for any one of these screening
criteria, then the site was considered to be in poor condition. However, in regional
clusters C, D, and E, a site had to exceed two or more of these thresholds to be
considered in poor condition. Analysts incorporated this rule due to the high number of
highly disturbed condition sites in these regions, when we applied a single failure as the
screening variable threshold.
Note that the NLA did not use data on landuse in the watersheds for the final
reference site screening—sites in agricultural areas (for example) may well be
considered least disturbed, provided that their chemical and physical conditions are
among the best for the region. Additionally, the NLA did not use data from the biological
assemblages themselves because these are the primary components of the lake
ecosystems being evaluated and to use them would constitute circular reasoning.
Screening NLA Site Data for Nutrient Reference Condition
Setting reference condition for nutrients requires a different process then the one
used for biological reference condition evaluation. Because nutrients (TN, TP) were
used to select biological reference sites, the biological reference sites could not be used
as nutrient reference lakes due to circularity. During the development of nutrient
reference sites, 11 nutrient ecoregions were utilized to categorized different portions of
the conterminous United States (USEPA 2000). These included Coastal Plain,
Temperate Plains, Southeastern Plains and Piedmont, Grass Plains, Cultivated Great
Plains, Southern Glaciated, Northern Glaciated, Southern Appalachian Mountains, Xeric
West, and Western Mountains. The Grass Plains was separated into to categories,
natural and man-made, due to the Sand Hills high natural nutrient levels (Table A-3).
-------
As with selection of biological reference lakes, chemical and lake riparian and
littoral condition thresholds were used to select nutrient reference lakes. An initial
screening of all ecoregions for inorganic acidity excluded all lakes with an ANC < 50
ueq/L and DOC < 5 mg/L. Once these lakes were excluded, selection of reference
conditions by nutrient ecoregion was conducted, using chloride, sulfate, shoreline
disturbance by agriculture, shoreline disturbance by non-agriculture, and shoreline
disturbance intensity and extent, and in field assessment of agricultural (Assess ag),
residential (Assess resid.), and industrial (Assess ind.) landuse from field data form
(Table A-3). Similar to biological reference selection, if a lake exceeded any one of
these eight selection criteria then the lake was not considered a reference lake.
However, chloride was not used to select reference lakes in two Omernik level III
ecoregions of the Western Mountains (ecoregion 1) and Northern Glaciated (ecoregion
82) nutrient ecoregions due to ocean influence.
Once the nutrient reference lakes were selected, nutrient levels for separating
Good, Fair, and Poor were determine from the distribution of reference lake nutrient
concentrations from the 11 nutrient ecoregions. Nutrient levels were determined for
both total phosphorus (TP) and total nitrogen (TN). The cutoff between Good and Fair
lakes was set at the 75th percentile (Q3) of reference lakes, and the cutoff between Fair
and Poor lakes was set at the 95th percentile (P95) of reference lakes (Table A-4). If a
nutrient ecoregion had < 20 lakes, then the cutoff between the Fair and Poor lakes was
the maximum nutrient concentration (P95 = maximum) for reference lakes in that
nutrient ecoregion.
In addition to developing thresholds for nutrients, we determined thresholds from
population percentiles in the reference lakes in each of the nutrient ecoregion for
chlorophyll-a and turbidity (Table A-5). Like the nutrient thresholds, these percentile-
based thresholds were used to determine Good, Fair, and Poor lake conditions for the
NLA. With the cutoff between Good and Fair lakes set at the 75th percentile (Q3), and
the cutoff between Fair and Poor lakes set at 95th percentile (P95).
-------
Table A-1. Regional biological reference thresholds for the reference/highly disturbed lakes
A
B
C
1,2
C
1,3
D
1
E
1
F
G
Phosphorus
M9/L
12/100
10/100
15/125
50/125
75 / 250
100/500
10/100
50 / 250
Nitrogen
M9/L
300 /
1500
300 /
1500
500 /
1500
750 /
1500
750 /
1500
15007
5000
300 /
1500
750 /
1500
Chloride
ueq/L
200 /
10,000
250 /
10,000
250 /
10,000
250 /
10,000
NA / 2000
600 /
10,000
250 /
10,000
500 /
10,000
Sulfate
ueq/L
400 /
1000
250 /
1000
250
/1000
NA/
1000
250 /
1000
15007
10,000
250 /
1000
500 /
4000
Turbidity
NTU
5/50
2/50
5/50
10/50
10/50
10/50
2/50
10/50
ANC4
ueq/L
<50
<50
<50
<50
<50
<50
<50
<50
Dissolved
oxygen
mg/L
>4/<3
>4/<3
>4/<3
>4/<3
>4/<2
>4/<3
>4/<3
>4/<3
Agriculture
disturbance
RDISINAG
0/0.5
0/0.5
0/0.3
0/0.3
0/0.5
0.1 70.5
0/0.5
0.1 70.5
Nonagricultural
disturbance
RDISINNONAG
0.6/0.80
0.5/0.75
0.6/0.8
0.6/0.8
0.6/0.75
0.6/0.75
0.5/0.75
0.5/0.75
Disturbance
intensity
RDISINEX1A
0.5/0.85
0.4/0.85
0.5/0.85
0.5/0.85
0.6/0.85
0.6/0.85
0.4/0.85
0.5/0.85
00
1 Because of the number of highly disturbed sites in these four clusters, a site had to exceed two thresholds to be
categorized as highly disturbed, unlike the other cluster where a site only had to exceed one threshold to be considered
highly disturbed.
2 Lakes at latitude greater than 40 degrees - the lake classification number is the sum of reference or highly disturbed
sites within the cluster.
3 Lakes at latitude less than or equal to 40 degrees.
4 ANC thresholds were only used to determine if a lake would be considered highly disturbed (i.e. adversely affected by
atmospheric deposition). If the ANC value was >50 ueq/L, or if dissolve organic carbon is > 5mg/L (even with an ANC
value < 50 ueq/L) then the lake was assumed to be an acceptable candidate for as a reference lake with regard to this
parameter.
-------
Table A-2. Biological Reference Sites
Reference Clusters
Eastern highlands
A
B
Plains and lowlands
C
D
E
Mountain West
F
G
Total
Data Source
Hand-pick
6
16
6
5
2
8
0
43
Random
11
14
24
14
22
32
10
127
Total
17
30
30
19
24
40
10
170
Figure A-1. Biological reference sites by seven regional clusters
-------
Table A-3. Nutrient reference site screening criteria by nutrient ecoregion
Nutrient
Ecoregion
Coastal Plain
II. Western Mts.
Ill.XericWest
IV. Grass Plains-Man-
made
IV. Grass Plains-Natural
IX. SE Plains/Piedmont
V. Cultivated Great Plains
VI. Temperate Plains
VII. Southern Glaciated
VIM. Northern Glaciated
XI. S. Appalachian Mts.
Chloride
(ueq/L)
>1000
>20
>500
>1000
>400
>200
>1000
>1000
>400
>20
>500
Sulfate
(ueq/L)
>400
>50
>10000
>10000
>400
>400
>10000
>10000
>400
>200
>500
Habitat
ag
disturb
>0
>0
>0.1
>0.2
>0
>0
>0.2
>0
>0
>0
>0.1
Habitat
non-ag
disturb
>0.6
>0.2
>0.6
>0.6
>0.1
>0.4
>0.6
>0.6
>0.6
>0
>0.5
Habitat
Ex1a
disturb
>0.6
>0.2
>0.6
>0.6
>0.1
>0.4
>0.6
>0.6
>0.6
>0
>0.5
Assess
ag
>4
>4
>6
>9
>5
>4
>9
>9
>9
>4
>9
Assess
resid.
>9
>4
>6
>9
>5
>9
>9
>9
>9
>9
>9
Assess
ind.
>4
>4
>6
>9
>5
>4
>9
>9
>9
>4
>9
Table A-4. Good/Fair/Poor Condition Class thresholds for total phosphorus
(TP) and total nitrogen (TN)
Nutrient
Ecoregion
Coastal Plain
II. Western Mts.
Ill.XericWest
IV. Grass Plains-Man-
made
IV. Grass Plains-Natural
IX. SE Plains/Piedmont
V. Cultivated Great Plains
VI. Temperate Plains
VII. Southern Glaciated
VIM. Northern Glaciated
XI. S. Appalachian Mts.
#Ref
Lakes
14
23
14
9
6
30
16
10
13
24
21
TP (ug/L)
Good-Fair
26
15
48
37
839
62
117
108
24
16.5
10
TP (ug/L)
Fair-Poor
75
19
130
56
1719
176
159
193
102
36
29
TN (ug/L)
Good-Fair
629
278
514
513
8647
680
1106
1240
828
674
311
TN (ug/L)
Fair-Poor
2311
380
2286
824
9359
1531
1355
2447
1410
1174
665
10
-------
Table A-5. Condition Class thresholds for chlorophyll-a and turbidity
Nutrient
Ecoregion
Coastal Plain Ecos
II. Western Mts.
Ill.XericWest
IV. Grass Plains-Man-
made
IV. Grass Plains-Natural
IX. SE Plains/Piedmont
V. Cultivated Great Plains
VI. Temperate Plains
VII. Southern Glaciated
VIM. Northern Glaciated
XI. S. Appalachian Mts.
Total # of
Lakes in Data
89
165
88
40
24
186
122
106
125
140
72
Chl-a (ug/L)
Good-Fair
29.1
1.81
7.79
13.9
118
31.7
49.9
37.8
8.56
7.56
5.34
Chla-a (ug/L)
Fair-Poor
75.6
2.74
29.5
25.1
144
84.0
76.3
49.6
46.4
12.5
23.8
Turb. (NTU)
Good-Fair
6.30
1.44
3.69
4.49
47.5
11.4
26.5
10.7
5.19
2.75
1.91
Turb (NTU)
Fair-Poor
19.9
5.47
24.9
14.4
75.9
37.3
50.3
19.7
102
5.41
2.38
Biological Data
Data Preparation: Standardizing Counts
NLA analysts standardized the number of individuals in a sample to a constant
number to provide an adequate number of individuals (i.e. diatom valves, natural algal
units, microcrustaceans and rotifers) that was the same for nearly all samples and that
could be used for both multimetric index development and/or 0/E predictive modeling
index. For sediment diatoms, a subsample was place on a microscope slide to be
enumerated. All samples were scribed with transect lines, which were used for counting
a know field of the subsample. Taxonomists were to enumerate no more than 600
valves per sample. For phytoplankton, a subsample was place on a microscope slide
and the taxonomists were to enumerate up to 300 natural algal units. Finally, for
zooplankton, two subsamples were enumerate for microcrustaceans and rotifers. For
each taxonomic group, taxonomist were supposed to count a minimum of 200
individuals and not more than a maximum of 400 individuals.
11
-------
Samples that did not contain the minimum number of units/individuals were
reviewed and retained for further analysis when appropriate (i.e. if the sampling effort
was determined to be sufficient) because low counts can indicate a response to one or
more stressors. For example, samples from sites classified as least disturbed were
retained if zooplankton counts were 100 or more individuals.
Operational Taxonomic Units
To provide a nationally consistent database for the diatoms, phytoplankton and
zooplankton, taxonomic lists were reviewed for discrepancies. In some cases it was
necessary to combine taxa to a coarser level of common taxonomy. This new
combination of taxa is called the "Operational Taxonomic Unit" or OTU and improves
the level of confidence in an overall assessment.
Sediment Diatom Metric Development
The taxonomic composition and relative abundance of different taxa that
compose the sediment diatom assemblages in lake sediments were used to develop a
diatom Index of Biological Integrity (IBI) or a Lake Diatom Condition Index (LDCI). IBIs
for fish and benthic macroinvertebrates have been used extensively in North America,
Europe, and Australia to assess how human activities affect ecological condition
(Barbouretal., 1995, 1999; Karr and Chu 1999). IBIs usually contain multiple
measures of a given assemblage, such as structural, functional, and/or tolerance
metrics, that respond positively or negatively to anthropogenic stressors (Barbour et al.
1999). The purpose of these indicators is to present the complex data represented
within an assemblage in a way that is understandable and informative to resource
managers and the public. This approach has been recommended for use in previous
EPA surveys such as the Wadeable Streams Assessment.
While diatoms have been used extensively in North America, Europe, and
Australia to monitor water quality, development of diatom IBIs has been much more
limited compared to other biological assemblages (Bahls 1993, Hill et al. 2000, Wang et
al. 2005). Additionally, most of these IBIs have been developed for lotic ecosystems
with a few exceptions for wetland ecosystems (Wang et al. 2006). This study contains
the first known published IBI for lentic ecosystems.
The following sections provide a general overview of the approach used to
develop ecological indicators based on sediment diatoms, followed by details regarding
data preparation and the process used for each approach to arrive at a final indicator.
Metric Evaluation and Selection
Candidate metrics were derived from the sediment diatom count data and traits of
each taxon. Morphological and growth form traits were obtained from literature or best
professional judgment. Indicator species analysis was used to determine diatom taxa
that were characteristically found in reference or impaired lakes, and to determine
diatom taxa characteristically found in either high TN and/or TP, or low TN and/or TP
(Dufrene and Legendre 1997). In most cases, three variants of each candidate metric
were calculated: one based on taxa richness, one based on the proportion of
12
-------
individuals, and one based on the proportion of taxa. All candidate metrics were
assigned to one of the following five categories representing different aspects of biotic
integrity (Barbouret al., 1999; Karr, 1993; Karret al., 1986; Stoddard et al., 2005).
o Similarity to Reference Condition: Proportions of individuals and taxa
characteristically found in reference or impacted sites
o Diversity: e.g. taxa number observed in samples and evenness of the
distribution of individuals across taxa
o Composition: e.g. the relative abundance of different genera
o Morphological and growth forms: e.g. Benthic, planktonic, motile, epiphytic,
colonial, chainforming
o Tolerance: e.g. low and high nutrient
Three performance evaluations were conducted to identify the best metric from each
metric category. Candidate metrics that failed a test were eliminated from additional
consideration and testing.
• Signal to noise (S:N) test: "Signal to noise" is the ratio of variance among sites and
the variance within a site (based on repeated visits to the same site). A low S:N
value indicates a metric that cannot distinguish among sites very well. S:N ratios
were calculated for each assessment region. Generally, candidate metrics having
S:N values < 1 were eliminated.
• Mann-Whitney (/test: Metrics were selected using this method when these tests
showed significant differences (a=0.05) between reference and highly disturbed
sites (see the description of how reference and poor sites were identified under
Setting Expectations). Additionally, analysts evaluated the separation power of each
significant metric using deviation in median ranks of metrics in reference and
impaired sites and the Z-statistic. Separation power has defined as the amount of
overlap (i.e. 25th and 75th percentile) in box plots of values of metrics for reference
and impaired sites (Barbour et al. 1996, 1999). The Z-statistic accounts for the
separate variation in ranks within reference and impaired groups as well as the
difference in magnitude of ranks among groups.
• Independence among metrics in different metric categories was evaluated using
correlations among metrics. Metrics within categories were often highly correlated.
Independence among metrics was maintained by calculating averages of metrics
within categories before calculating an overall average LDCI.
Metrics with the highest S:N and Z-statistics (either positive or negative) and lowest
correlations with other categories were selected for inclusion in the LDCI.
Metric Selection, Scaling, Transformation and Calculation of Lake Diatom
Condition Indices
Multiple versions of the Lake Diatom Condition Index were calculated to evaluate
their relative performances for distinguished reference and highly disturbed sites. The
13
-------
same metrics were used in calculating all versions of the LCDIs. All LCDIs were based
on metrics that were scaled to the same range.
Structural and Tolerance Metrics
Most structural and tolerance metrics were scaled to a 0-1 range using the
"Blocksom - 5th-95th percentile" method (Blocksom 2003):
• determine 5th and 95th percentiles of metrics;
• subtract the 5th percentile of the metric from metric value at a site; and
• divide that quantity by difference between the 5th and 95th percentiles
(Table A-6).
If metrics were positively correlated with the chemical principle components analysis
(PCA) factor score, they were subtracted from 1.0 to reverse the scale such that a 0.0
and 1.0 metric scores indicated low and high biological condition, respectively. If
metrics were positively correlated with the chemical PCA factor, they were not
subtracted from 1.0 to reverse the scale.
Genus Level Composition Metrics and Percent Epiphytic Individuals
Genus level species composition metrics and percent epiphytic individuals were
normalized using a 0, 0.5, 1.0 values that were assigned to metric values based on
quartile separations in the ranges of the metrics (Table A-7). This scale was used
because many sites had 0.0 relative abundances of genera and percent epiphytic
individuals, which would skew the distribution of a metric normalized using the 5-95th
percentile method described above.
• If metrics were negatively correlated to the chemical PCA factor score, high
values of metrics indicated high biological condition. In this case 0 was assigned
to metric values if they were less than the 25th percentile of the metric range, 0.5
was assigned to metric values if greater than or equal to the 25th percentile and
less than the 75th percentile, and 1.0 was assigned to metric values if greater
than or equal to the 75th percentile.
• If metrics were positively correlated to the chemical PCA factor score, high
values of metrics indicated low biological condition. In this case 10 was assigned
to metric values if they were less than the 25th percentile of the metric range, 0.5
was assigned to metric values if greater than or equal to the 25th percentile and
less than the 75th percentile, and 0.0 was assigned to metric values if greater
than or equal to the 75th percentile.
14
-------
Table A-6. The 5th and 95th percentile of final evaluated structural and tolerance
metrics.
Metric
Prop. Impacted
spp
Prop. Reference
spp
Shannon
diversity /-/'
Richness
Prop. Colonial
Individuals
Prop. Low TP
Taxa
Prop. High TP
Taxa
Prop. Low TN
Taxa
Prop. High TN
Taxa
5th percentile
0.000
0.026
1.414
19
0.034
0.034
0.042
0.045
0.029
95th percentile
0.349
0.050
3.486
71.45
0.825
0.606
0.651
0.611
0.618
If more than 25% of the sites had zero relative abundance at a site, then greater than
25% of the values would be assigned either a 0.0 or 1.0, depending upon the
relationship between the metric and the PCA score. For example, the % Cocconeis
individuals at 41 % of the sites were 0.0 and this metric was positively correlated to the
chemical PCA factor score of the site; accordingly 41 % of the sites with relative
abundances equal to 0.0 were assigned a 1.0, 34% of site were assigned with relative
abundances between the 41st and 75th percentiles were assigned a 0.5, and the
remaining 25% of values was assigned a 0.0.
15
-------
Table A-7. The 25th and 75th percentile of example genus/growth form metrics.
Metric
Prop.
Achnanthidium
Individuals
Prop. Cocconeis
Individuals
Prop. Cvclotella
and
Stephanodiscus
Individuals
Prop. Epiphytic
25tn percentile
0.002
0.000
0.003
0.002
75tri percentile
0.045
0.012
0.234
0.024
The LDCI was calculated in three steps to produce a multimetric index ranging
between 0 and 100. First the averages of selected metrics within the five metric
categories were determined. Then these averages of each metric category were
weighted evenly by multiplying by 20 and these products were summed to calculate the
final value of the observed LDCI. Finally, LDCI was calculated as the deviation in
observed LDCI values and the expected LDCI value for a specific lake, where the latter
accounted for variation in LDCI values due to natural lake and lake watershed features.
Performances of different models predicting expected LDCI (using the 10
selected and scaled metrics in Tables A-6, A-7) were evaluated by calculating variation
in expected LDCI among reference sites. Models with low variation in expected
condition at reference sites will more precisely distinguish between reference and
impaired condition. Models of expected LDCI differed as a result of two calculation
methods, whether individual metrics or the multimetric LDCI were predicted by models,
and in the variables used to account for natural variation in expected LDCI.
Expected LDCIs models were calculated using the following (Table A-8):
1. no natural variables;
2. natural versus man-made lakes;
3. seven lake type clusters;
4. nine ecoregions
5. Classification and regression tree (CART) model predictions of natural variation
in metrics at reference sites using all "CIS" variables as in Cao et al. (2007);
6. adjusted using CART model predictions of natural variation in LDCI at reference
sites using all "CIS" variables;
16
-------
7. regression predictions of natural variation in LDCI at reference sites using all
"CIS" variables and forward stepwise regression;
8. regression predictions of natural variation in LDCI at reference sites using "CIS"
variables selected for first principles as important causal variables and forward
stepwise regression;
9. regression predictions of natural variation in LDCI at reference sites using "CIS"
variables selected for first principles as important causal variables and using all
subset regression; and
10. regression predictions of natural variation in LDC at reference sites using "CIS"
variables selected for first principles as important causal variables, minus
maximum lake depth because it was negatively related to LDCI, and using all
subset regression.
Table A-8. The amount of variation in expected LDCI among reference sites using
the models above.
Model
1
2
3
4
6
w
7
8
VJ
9
10
1 W
LDCI model
None
lake type clusters
lake vs reservoir
ecoregions level 9
all natural factors for metrics by
j
CART model
all natural factors for LDCI by
CART model
all natural factors for LDCI by
GLM
1st P natural factors for LDCI by
GLM
all subset NF for LDCI by GLM
all subset NF - depth for LDCI
by GLM
Ref Var
226.6
157.9
210.6
140.2
Q^ 9
\J\J . £-
1166
1 1 \J . \J
102.8
97 2
\J 1 .£-
97.2
107 2
I \J I . ^
Option 10 was chosen, with reference variance of 107.2. This model had lower
variance than any of the a priori categorical classifications of lakes. Model 5(14 metrics
independently adjusted by CART model) had lower variance among reference site LDCI
values, but requires more evaluation before finalizing. Of the last four models, we
decided that all subset regression models (models 9 and 10) were better than models
using forward stepwise regression (models 7 and 8). Model 10 was chosen because
maximum lake depth, a predictor variable in the LDCI was negatively related to LDCI in
model 9, which does not make sense based on first principles, i.e. we predicted that
deep lakes should naturally have higher LDCIs than shallow lakes. Exploration of
regional patterns indicted maximum lake depth was positively and negatively correlated
to LDCI at reference sites, depending upon region. Model 10 was:
17
-------
ExpectedLDC = (73.363 - 79.948 x KFCT_AVE + 0.008167 xLAT_DD2 -1.367
I.ll6x^/SUMP _PT))
In the model, KFCT_AVE is watershed mean soil erodibility factor of soils (no
units) from State Soil Geographic (STATSGO) Database, LAT_DD is latitude in decimal
degrees, BASIN_LAKE_RATIO is ratio of basin area to lake area, ELEV_PT is site
elevation (meters) from the National Elevation Dataset, LON_DD is longitude in decimal
degrees, and SUMP_PT is annual sum of the predicted mean monthly precipitation
(mm) derived from the PRISM data.
Plankton O/E: Predictive (RIVPACS) Models
Observed over Expected (O/E) indices provide a quantitative measure of
biological condition by measuring the agreement between the taxonomic composition
expected under reference conditions and that observed at individual sites. For the NLA,
we developed a combined phytoplankton-zooplankton O/E index based on the 259
plankton taxa observed across reference-quality lakes (351 total plankton taxa were
identified across all of the 1157 NLA lakes).
Because taxonomic composition can vary markedly with natural environmental
factors, application of the O/E index depends substantially on development of models
that predict how taxonomic composition varies with natural environmental setting. These
models are calibrated with data collected at reference sites.
One hundred and seventy lakes were initially identified as candidate reference
lakes for use in calibrating 3 regional (western mountains and xeric (WMTNS), plains
and lowlands (PLNLOW), and eastern highlands (EHIGH)) models. As described, these
reference lakes were selected from the 1157 lakes sampled for the NLA based on
application of regional screening criteria (see pages A-2). Of these 170 lakes, 14 large
PLNLOW lakes lacked littoral predictor variable data and were dropped from model
development resulting in156 calibration lakes for purposes of the O/E model (Fig. A-2).
The 50 (WMTNS), 59 (PLNLOW), and 47 (EHIGH) lakes were then classified into
4 (WMTNS = groups 1-4), 5 (PLNLOW = groups 5-9), and 3 (EHIGH = groups 10-12)
groups based on their combined phytoplankton and zooplankton taxa composition (Fig
A-2). Following reference site classification, discriminant functions models were
developed to predict the probability of group membership from naturally occurring
landscape attributes. For the WMTNS model, logic average available water holding
capacity (proportion), log 10 soil permeability (inches/hr), average depth to the water
table (feet), longitude (decimal degrees), and logic average calcium oxide content (%)
of the watershed's bedrock were identified as predictors. For the PLNLOW model, mean
long-term maximum monthly air temperature (°C), mean long-term maximum monthly
precipitation (mm), log™ lake surface area (km2), log™ watershed area (km2), log™ lake
perimeter (km), dummy variable specifying natural lake (1) or reservoir (0), square root
18
-------
of the percent littoral cover as aquatic macrophyte, and square root of the percent
littoral cover as organic matter were identified as predictors of plankton composition. For
the EHIGH model log™ average available water holding capacity (proportion), average
bulk soil density (g/cm3), and average depth to the water table (feet) were identified as
predictors.
Figure A-2. Location of the 156 reference lakes used to develop the
phytoplankton-zooplankton O/E indices. Individual lakes are symbol and
color coded by the groups they were assigned to based on similarity in
taxa composition.
These discriminant functions models were used to predict the probabilities of
each of the 1157 NLA lakes belonging to each of the biologicacally-defined groups.
These probabilities, together with the observed frequencies of occurrence of each taxon
among the lakes in each group, were used to predict the probabilities of observing each
of the 259 reference lake taxa at each of the 1157 NLA lakes. For each lake,
probabilities of detection > 0.5 were then summed to estimate the number of reference-
lake taxa expected (E) at each lake. The O/E index is the proportion of those reference-
lake taxa predicted at a specific lake that were observed (0) in a sample.
The distribution of O/E values observed at reference-quality lakes is used to
evaluate the condition of assessed lakes. Because the reference-site O/E distributions
19
-------
for the 3 regions had 1 or 2 statistical outliers, we dropped these values when
estimating the 5th and 25th percentile of reference site values. Lakes with 0/E values >
25th percentile of reference site values were considered to be in good condition. Lakes
with 0/E values between the 5th and 25th percent!les of reference site values were
considered to be in fair condition. Lakes with 0/E values < the 5th percentile of reference
site values were considered to be in poor condition. The 5th and 25th percentile
reference site 0/E values, were 0.60 and 0.80.
Relative Risk, Attributable Risk, and Relative Extent
In the NLA, each targeted and sampled lake was classified as being in either
"Good," "Fair," or "Poor" condition, separately for each stressor variable and each
biological response variable. From this data, we estimated the relative extent
(prevalence) of lakes in Poor condition for a specified stressor or response variable. We
also estimated the relative risk (RR) of each stressor for a biological response. RR
measures the severity of a stressor's effect on that response in an individual lake, when
that stressor is in Poor condition (Van Sickle, et al. 2006). Finally, we estimated the
population attributable risk (AR) of each stressor for a biological response. AR
combines RR and relative extent into a single measure of the overall impact of a
stressor on a biological response, over the entire population of lakes (Van Sickle and
Paulsen 2008).
To estimate RR and AR, we first combined the "Good" and "Fair" classes of
condition into a single class designated "Not Poor". Thus, each sampled lake was
designated as being in either Poor (P) or Not-Poor(A/P) condition, separately for each
stressor and response variable.
To estimate the relative extents, RR, and ARfor one stressor (S) and one
response (Y) variable, we compiled a 2x2 table (Table A-9), based on data from all
lakes that were included in the probability sample. A separate table must be compiled
for each pair of stressor and response variables:
20
-------
Table A-9.
Response (Y)
Not-Poor
(NP)
Poor (P)
Stressor (S)
Not-Poor (NP)
a
c
Poor (P)
b
d
Table entries (a,b,c,d) are the sums of the sampling weights of all sampled lakes
that were found to have each combination of P or NP condition for S and Y. Thus,
REs.est = (b+d)/(a+b+c+d) is the estimated relative extent of Poor stressor condition,
calculated by estimating the number of lakes in the population that have Poor stressor
condition (totaled over both classes of response condition), divided by the total
estimated number of lakes. Similarly, REY,est = (c+d)/(a+b+c+d) estimates the relative
extent of lakes in Poor condition for the biological response variable.
RES can also be interpreted as the probability that a lake chosen at random from
the lake population will have Poor stressor condition. In shorthand, this probability can
be written as RES = Pr(S = P). We use similar concepts and language from probability
theory to define and interpret RR and AR.
[Note: RE estimates made from Table A.1 may differ slightly, due to sampling
variation, from our extent estimates that were made using separate data for each S and
Y variable. This is because the weight sums in Table A.1 include only those lakes that
have nonmissing class assignments for both S and Y]
Relative Risk (RR)
Relative risk measures the likelihood (that is, the "risk", or probability) of finding
Poor biological response condition in a lake when the condition of a specific stressor is
also Poor. This likelihood is expressed relative to the likelihood of Poor response
condition in lakes that have Not-Poor stressor condition. That is,
RR =
Pr(Y=P\S=P)
Pr(Y=P\S=NP)
(A1)
Using Table A.1, RR is estimated by:
RRest = [d/(b+d)]/[c/(a+c)J
(A2)
21
-------
RR = 1.0 indicates "No association" between stressor and response, that is, Poor
biological condition in a lake is equally likely to occur whether or not the stressor
condition is Poor. RR < 1.0 indicates that Poor response condition is actually less likely
to occur when the stressor is Poor.
Further details of RR and its interpretation, including estimation of a confidence
interval for RRest, can be found in Van Sickle et al. (2006).
Attributable Risk (AR)
Attributable risk (AR) measures how much of the extent of Poor condition for a
biological response variable can be attributed to the Poor condition of a specific
stressor. AR is based on a scenario in which the stressor would be entirely eliminated
from the lake population, by means of restoration activities. (By "eliminated", we mean
that all lakes in Poor condition for the stressor are restored to the Not-Poor condition.)
Under this scenario, AR is defined as the proportional decrease in the extent of Poor
biological response condition that would occur if the stressor were eliminated from the
lake population. Mathematically, AR is defined as (Van Sickle and Paulsen 2008)
Pr(7 = P)
We estimated AR using Equation A3. We first calculated REY,est (see above), which is
an estimate of Pr( Y = P). Then, using the weight sums in Table A-9,
ARest = [REY>est - c/(a+c)] I REY>est (A4)
We calculated a confidence interval for ARest following Van Sickle and Paulsen (2008).
AR can take a value between 0 and 1 . An AR value of 0 indicates either "No
association" between stressor and response, or else a stressor having zero extent.
A strict interpretation of AR in terms of stressor elimination, as described above,
requires one to assume that the stressor-response relation is strongly causal and that
stressor effects are reversible. Van Sickle and Paulsen (2008) discuss the reality of
these assumptions, along with other issues such as interpreting the AR's of multiple,
correlated stressors, and using AR to express the joint effects of multiple stressors.
However, AR can also be interpreted more informally, as a measure that
combines RR and relative extent into a single index of the overall, population-level
impact of a stressor on a response. After some algebra, AR can be written as (Van
Sickle and Paulsen 2008)
AR= . (A5)
l + REs(RR-V)
22
-------
Equation A5 shows that the numerator of AR is the product of the relative extent
of Poor stressor condition and the "excess" relative risk (RR-1) of that stressor. The
denominator standardizes this product to yield AR values between 0 and 1. Thus, a high
AR for a stressor indicates that the stressor is widely prevalent (has a high relative
extent of Poor condition), and the stressor also has a large effect (high RR) in those
lakes where it does have Poor condition.
Water Chemistry Analysis
Six chemical stressors are summarized in the NLA report: total nitrogen, total
phosphorus, dissolved oxygen, turbidity, acidity and salinity. For total nitrogen and total
phosphorus, threshold values were determined during the reference nutrient process.
For setting nutrient class boundaries, reference sites from the screened NLA dataset
were used (see page 3). Because nutrients were the focus, the two nutrient screening
levels used in defining biological reference sites were dropped and the other screening
factors were used by themselves to identify a set of "nutrient reference sites." Before
calculating percentiles from this set of sites, outliers (values outside 1.5 times the
interquartile range) were removed (Herlihy and Sifneos 2008). For acidity, threshold
values were determined based on values derived during the NAPAP program. Sites
with acid neutralizing capacity (ANC) less than zero were considered acidic. Those with
dissolved organic carbon (DOC) greater than 10 mg/L were classified as organically
acidic (natural). Acidic sites with DOC less than 10 and sulfate less than 300 |jeq/L
were classified as acidic deposition impacted, those with sulfate above 300 were acid
mine drainage impacted. Sites with ANC between 0 and 25 |jeq/L were considered
acidic deposition influenced, but not currently acidic.
Salinity classes and dissolved oxygen were divided into low, medium, or high
classes. Salinity classes were defined by specific conductance using ecoregional
specific values (Table A-10). Dissolved oxygen classes were defined by oxygen
concentration for each ecoregion (Table A-10).
23
-------
Table A-10. Select Water Chemistry Criteria for NLA Assessment
Ecoregion
CPL
NAP
SAP
UMW
TPL
NPL
SPL
WMT
XER
Salinity as
Conductivity
(|jS/cm)
Low-
Medium
500
500
500
500
1000
1000
1000
500
500
Salinity as
Conductivity
(|jS/cm)
Medium-
High
1000
1000
1000
1000
2000
2000
2000
1000
1000
Dissolved
Oxygen
(mg/L)
Low-
Medium
>3
>3
>3
>3
>3
>3
>3
>3
>3
Dissolved
Oxygen
(mg/L)
Medium-
High
5<
5<
5<
5<
5<
5<
5<
5<
5<
Trends Studies
Trend Analysis of National Eutrophication Survey (NES)
Monitoring and surveillance programs have, in the past, often dealt with site-
specific questions of ecosystem condition, thus concentrating on single lakes or small
groups of lakes. For example, sites are often monitored for nutrient levels, frequency of
algal blooms, fisheries, fecal coliform counts at swimming beaches, etc. However,
present day pressures on aquatic systems across large geographic areas have driven
the need to assess lakes over far wider regions. This kind of need produced the
National Eutrophication Survey (NES) in 1972-1976. While this survey was National in
scope, the design was not of a rigorus scientific nature. The purpose of the NES was to
assess the trophic condition (defined as the nutrient enrichment) of lakes influenced by
domestic waste water treatment plants (WWTP) with a sprinkling of special purpose
lakes that were not necessarily influenced by WWTP's. The specific purpose of the
survey was to measure nutrient inputs from all sources in the watershed relative to
those of the WWTP source to determine if WWTP upgrades might be successful in
modifying the lake or reservoir trophic state.
The trophic state definition and assessment approach to water quality employed
by the NES did not necessarily result in the identification of degraded water quality, but
was used simply to characterize the water quality based on the nutrient levels. The
perception of good or poor water quality based on the nutrient level depends on the
intended beneficial use of the water body. For example a reservoir managed for a warm
water fishery can tolerate a much greater degree of nutrient enrichment than can a lake
intended for a cold water trout fishery. Therefore, nutrient enrichment, or degree of
eutrophication was employed as a tool to gauge the water quality, while the
24
-------
interpretation of good or poor condition depended on the intended use of the lake or
reservoir. While the NES survey assessed the condition of selected lakes across the
United States, the focus of the survey was to assess the condition of individual lakes in
detail rather than to extrapolate results to the condition of all lakes. Trophic state
condition based on nutrient levels with some characterization of correlative Secchi disk
transparency, phytoplankton, and chlorophyll-a became the focus for individual NES
lake reports. Because of the correlative nature between nutrients and chlorophyll-a
concentration, NLA analysts decided to utilize chlorophyll-a concentrations as the
trophic condition indicator for the current comparison between studies. Additionally,
because of the significance of nutrients in inland waters, NLA analysts also made a
comparison of nutrient concentrations between studies.
Since 1972 there has been increased interest in characterization of lake and
reservoir quality on a regional basis relative to biogeochemical and land use
characteristics. This approach seeks to quantify those characteristics relative to lake
water quality, such that regionalized management techniques might be utilized to
minimize adverse effects on water quality. Relatively new developments in statistical
sampling designs, ecoregional landscape classification and TM and AVHRR
technologies coupled with CIS capabilities and other techniques make it possible to do
regional resource analyses in ways that did not exist at the time of the NES. We now
have tools to conduct a regional census, survey, modeling effort or other techniques
designed to describe, infer, or extrapolate lake, reservoir, and wetland conditions across
temporal, spatial and biological scales (boundaries).
Because the design of the NLA selected lakes on a probability basis by number
and size we can infer analytical results of the sampling to the population of lakes from
which the sample was drawn. As part of the NLA design process, the team also had the
opportunity to include a subset of lakes from the NES survey of 1972-1976 and make a
comparison of trophic state, i.e. how have these lakes changed in trophic state over 35
years. This afforded NLA analysts the opportunity to use the rigorous statistical design
of the regional NLA to provide a benchmark of condition change for lakes originally
selected and evaluated on a site-specific basis. As a result, the subset of NES lakes
sampled for NLA, was used to estimate the current condition of all 800 lakes in the
original NES survey.
The goal of this trend analysis was to compare the 1972-1976 trophic state of a
subset of the NES lakes with the trophic state of those same lakes today (2008). As
described above, the NLA sampling and analysis provided the opportunity to
accomplish this goal. While sampling and analysis techniques differed somewhat for the
two surveys, NLA analysts determined that the differences were insignificant relative to
comparing trophic states using chlorophyll-a and nutrient concentrations. The NLA
sampling consisted of a single, mid-summer integrated water sample at the deepest
spot in the lake and from just below the surface to a depth of about 1.5m (a sampling
tube). The NES sampling consisted of sampling several sites on a lake as well as the
inlets and outlets. However, this sampling also included a site at the perceived deepest
spot in the lake. Sampling was done with a Van Dorn Bottle at just below the surface
25
-------
and at 1 -2 m depth intervals. Therefore, the current comparison used the integrated
sample NLA chlorophyll-a and nutrient concentrations and compared them to NES
samples taken at the site nearest the NLA site and from depth(s) that most nearly
mimicked the depth of the NLA integrated depth sample. The accuracy and precision of
analytical results are considered comparable to each other based on the methods and
the QA of both the NES (USEPA 1974; USEPA 1975a; USEPA 1975b) and the NLA.
We used a 4 X 4 factorial design to assess the trophic state of the original NES
survey and the NLA. We determined how many lakes were in each of the following
trophic classes hypereutrohic (HE), to eutrophic (E), to mesotrophic (M), to oligotrophic
(0), as well as the reverse series from 0, to M, to E, to HE, during 1972 and 2007. The
results are summarized in Figure A-3.
National Eutrophication Survey Lakes
1972 & 2004
Number of
Lakes
Oligotrophic
Mesotrophic
Eutrophic
Hypereutrophic
49%
43
117
233
189
130
215
394
279
40 BO
Percentage of Lakes
so
100
1972-NES
2007 NES in NLA
Figure A-3. Percentage and number of NES lakes estimated in each of four
trophic classes in 1972 and in 2007 based on chlorophyll-a concentrations.
The 2007 NLA indicates that 51.1% of the NES lakes remain in the same trophic
state category as they were in 1972 (Figure A-4). Another 22.6 degraded to lower
trophic state categories. Only 26.3 percent of the lakes actually improved in their trophic
state category. While at first glance this seems a rather bleak picture, the results must
be put into proper perspective. First, most of the original NES lakes were eutrophic or
hypereutrophic to begin with because they were selected for their proximity to domestic
waste treatment plants. Second, there likely has been an increasing population density
26
-------
associated with these lakes, again most likely since they were in the vicinity of waste
treatment plants where populations usually grow.
While the 2007 NLA indicates a modest improvement in trophic state, as
assessed as chlorophyll-a, there does appear to be a much more substantial
improvement in total phosphorus concentrations between 1972 and 2007. The 2007
NLA indicates that 50.4% of lakes decreased in there total phosphorus concentrations,
while only 25.9% increased, with another 23.6% showing no change. The most likely
reason for this out come is the improvements in waste treatment plants between 1972
and 2007. However, this inconsistency between chlorophyll-a and total phosphorus
concentrations is difficult to resolve.
While we have land use cover for the 2007 NLA we do not have similar land
cover data for the original NES, that we might be able to make land use change
associations with the trophic state changes (chlorophyll-a).
NLA analysts did not identify the causes for the improvements or the declines,
and other factors may be influencing chlorophyll-a that are not influencing total
phosphorus concentrations. Additional analyses are recommended to delve into these
results in greater detail.
Change in Trophic State (Chlorophyll)
i i Degraded i ' Unchanged ' ' Improved
Figure A-4. Change in trophic state of lakes between the 1972 NES and 2007 NLA
studies.
27
-------
Change in Phosphorus
NES Lakes 1972-2007
Figure A-5. Change in total phosphorus of lakes between the 1972 NES and 2007
NLA studies.
Diatom Sediment Core Analysis
Sediment Diatom Transfer Function
Variations in fossil diatom species composition were used to assess the amount
of change that has occurred in lake systems cored within the NLA since the European
settlement. Diatoms are one of the most powerful water quality indicators used in
paleolimnological studies. They colonize virtually every freshwater microhabitat and
many diatom species have well-defined optima and tolerances for environmental
variables such as lake pH, nutrient concentration, water salinity or color (Stoermer and
Smol 1999). Thus, they constitute a powerful approach to allow lake managers
characterize natural background or reference conditions and to track past changes in
lake systems (Smol 1992; Charles et al. 1994).
In order to reconstruct changes in study lakes, the following main steps were
taken: 1) Calibration and development of transfer functions; 2) Reconstruction and
assessment of the magnitude of change in lake characteristics (see summary of method
in Figure A-6).
28
-------
a) Calibration step: establish relationships between chemical
variables measured in lake water column (X) and (diatom species (Y)
identified in lake surface sediments
Study Lakes
Water Chemistry
Measurements
Surface Sediment
^Diatom Samples
VsX ViX VS*' Xix
vs^ \^s \^S X^X
Diatom Species in
Surface Sediments
Lake Water Chemistry
Variables
Species 1, Species 2, .. Species m
Lakel
Var. 1,Var. 2, .. Var. p
Lake 2
Diatom species are a function U of
lake water chemical variables (e.g.,
TP, TN)
Y=UX
b) Reconstruction step: use fossil diatom species
preserved in lake sediments to infer past lake water
chemistry (TP, TN)
TP, TN are the function U'1 of fossil
diatom species
Top / Bottom Stratigraphic
Cores Cores
X = U'1Y
• Bottom samples (provide pre-European
settlement conditions)
• Stratigraphic samples (provide pre-European
to present-day conditions)
Figure A-6. Quantification of relationships between diatom species in lake surface
sediment samples (calibration set) and measured water chemistry (Y = UX) and
development of transfer functions (X = U~1Y) to infer water characteristics (e.g.,
TP, TN) based on the composition and number of diatom species in top and
bottom (or Stratigraphic) sediment samples.
29
-------
Calibration and Development of Transfer Functions
1 Ordinations in reduced space. Prior to the development of diatom-based transfer
functions, diatom distributions and relationships with ecological data were explored.
This step is necessary to examine whether linear or unimodal methods are appropriate
for the available training set in relation to the environmental variables of interest.
Detrended Correspondence Analysis (DCA) of diatom species data was performed to
determine if the length of environmental gradient is >2.5 SD in order to allow the use of
unimodal statistical modeling. Since the environmental gradient was >2.5 SD, unimodal
methods were considered appropriate to explore variation in surface sediment diatom
assemblages among the lakes and to determine which chemical and other
environmental factors explain statistically significant proportions of their variance.
A CCA (and associated Monte Carlo permutation test) using all measured
environmental variables was carried out to find a minimal set of variables that explain, in
a statistical sense, some variance within the diatom species data. Species relative
abundances were log-transformed and down weighted for rare taxa. The statistical
significance of each variable was assessed using Monte Carlo unrestricted permutation
tests involving 999 permutations (ter Braak, 1989). CCA allowed us to determine how
strongly diatom species composition is related to ambient lake water measured
parameters. Within the measured variables included in the NLA data set, PTL, COND,
NTL, and pH were found to explain highly significant proportions of the species variance
(p=0.001). CCA also allowed for the identification of outlier samples (i.e., samples with
unusual diatom assemblages, unusual combination of environmental variables or a
diatom assemblage with poor relationship to environmental variables) that have extreme
(more than 5 times) influence and very high squared residual chisquared distance.
Other analyses performed within this step were Principal Component Analysis to assess
variations in lake physico-chemical caracteristics, and a Pearson correlation matrix with
Bonferroni adjusted probabilities to idenrify groups of significantly (p < 0.05) correlated
environmental variables. Multivariate analyses were performed using the computer
program CANOCO (ter Braak and Smilauer 2002) and correlations using the program
JMP 5.01 (SAS Institute Inc.).
2. Development of transfer functions. In this step, diatom species from surface sediments
('modern samples', constituting the calibration or 'training' data set) that are significantly
influenced by select variable (i.e., they have well defined optima and narrow tolerances
with regard to TP, TN, pH or CON) can be used to develop inference models for these
variables. Models to infer TP, TN, conductivity and pH were developed using weighted-
averaging partial-least-squares (WA-PLS) models. This reconstruction procedure has
the advantage of taking into account the residual correlations that remain after fitting the
environmental variable of interest (ter Braak, 1993; Birks 1998). In WA-PLS, the first
component is selected to maximize the covariance between the vector of species
weighted averages and the environmental variable of interest. Subsequent components
are chosen in the same way but with the restriction that they be orthogonal and hence
uncorrelated to earlier components (ter Braak, 1993). The number of components to be
retained is determined by cross-validation (leaveone-out-jackknifing) on the basis of
30
-------
prediction error sum-of-squares (PRESS), to minimize RMSEP and bias. WA-PLS
calibration functions were developed on log-transformed species data. The species data
set was based on species names corrected for misspelling and synonyms. The data set
includes 220 diatom species that occur with at least 2% relative abundance in at least 6
samples. Transfer Functions computed including species < 2% had lower predictive
power except for conductivity. The chemistry data was log transformed except for pH; if
chemistry data were available for more than 1 visit, values were averaged.
Calculations were done using the computer program C2 (Juggins 2003). Error
estimates based on cross-validation were provided for each model. The models were
subsequently used to infer lake water TP, TN, CON and pH in top and bottom samples
and assess the amount of change that occurred in sediment cores.
Table A-11. Transfer functions developed for the national calibration data set
Transfer
function
TP
TN
PH
Cond
R2
0.74
0.67
0.59
0.78
Jack R2
0.67
0.58
0.52
0.75
RMSEP Jack
0.36
0.29
0.63
0.28
Equation
Inferred Log TP = 0.4502122 +
0.7038829*0bsrved Log TP
Inferred Log TN = 1 .0462799 +
0.62751 95*0bserved Log TN
Inferred pH = 3.6986058 +
0.542196*ObservedpH
Inferred Log Cond = 0.5794113 +
0.7501 362*0bserved Log Cond
Models were developed using the whole data set and by splitting data for
individual clusters defined within this project. The predictive power was also explored for
the original data set (fixed names) and for data sets with lumped diatom species.
Top-Bottom Changes
A fast way to quantify the changes that have affected lake systems, beginning
with the European settlement, is to examine how much the composition and relative
abundance of sedimentary diatom assemblages changed between the top (representing
present-day conditions) and at the bottom of sediment cores, representing the
reference, undisturbed conditions (i.e., top-bottom approach). The top-bottom approach
provides two 'snap-shots' of environmental conditions, before and after human impacts,
and has proven successful in addressing diverse environmental questions such as the
impact of acid rain, eutrophication or global warming (Cumming et al. 1992; Dixit et al.
1999; Smol et al. 2007).
The transfer functions were used to reconstruct pH, TP, TN, and conductivity in
top and bottom samples and assess the amount of change since pre-industrial (pre-
European) times in cores that were considered long enough to reach pre-disturbance
conditions. Since Pb-210 dating of bottom cores was not performed, we used an
31
-------
alternative approach to evaluate whether or not bottom cores may represent reference
conditions (see below).
Determination of whether sediment cores represent pre-disturbance conditions
We assigned each sediment core to one of three categories based on our
confidence that the bottom interval represented time prior to European-settlement
disturbance typical for the region. "Yes" indicates confidence that the bottom
represents a pre-disturbance time period. Usually these are from longer cores and / or
from lakes with lower sedimentation rate. "No" means it is unlikely that the interval is
sufficiently deep to represent pre-disturbance time. These are usually from shorter
cores and/ or lakes with presumed high sedimentation rates. "Uncertain" means that it
is difficult to make a determination. This category was used for lakes officially
designated "man-made" (reservoirs), oxbow lakes, and others that were borderline in
terms of core length, presumed sedimentation rate, and disturbance history.
Category assignments were based on several factors, including sediment core
dates from previous studies and evaluation of lake and watershed characteristics that
can have a strong influence on sedimentation rates. Key variables considered were
nutrient ecoregion, lake cluster (A-G), total percent watershed disturbance, total P,
depth, surface area and watershed area. As general principles, lake watersheds with
highly erodible soils and high watershed disturbance (especially Ag) tend to have
greater input of inorganic particles due to erosion. Watersheds with high percent
agriculture and urban tend to have higher algal growth stimulated by increased nutrient
inputs. Sediments in shallower lakes might be mixed to a greater depth than deeper
lakes. In all the cases above, a longer core would be required to reliably represent pre-
disturbance times.
Many final decisions were based on viewing the lake and its watershed using
Google Earth. This was a very valuable source of information and showed many
important characteristics that otherwise would not have been taken into account (e.g.,
land-use disturbance patterns, location of shoreline riparian vegetation, local hydrology).
Over half the lakes were viewed with GE.
There were 501 lakes with both top and bottom sediment core intervals, and that
had sufficient number of diatoms counted (about 30+ lakes had cores for which one or
more intervals had a low number of diatoms counted). Of these, 294 were categorized
"yes", 106 as "no", and 101 as "uncertain." There were also data for 30 duplicate cores
that were not included in the analysis. In most cases, cores from visit 1 were used; visit
2 cores were sometimes used if they were longer.
Even though a sediment core bottom-sample may not be deep enough to
represent a pre-settlement time horizon, it can still represent reference conditions if the
lake has not been disturbed. We made no attempt to make these types of
determinations.
32
-------
General criteria for categories
Yes
Lakes were accepted for analysis if: they occurred in nutrient ecoregions 2 and 8
where sedimentation rates are known to be relatively low, based on previous studies;
lakes with undisturbed, or relatively undisturbed watersheds, and at least moderately
long cores for the region; lakes in the Northeast US greater than 25 cm in length were
generally considered sufficiently long based on results from the EMAP Surface Water
study (NE lakes; ragweed pollen was analyzed in the bottom sediment samples to help
confirm pre-disturbance time period).
No
Lakes were rejected if: cores less than 20 cm in length, except a few reference
lakes that seemed clearly undisturbed and to have low sedimentation rate; all lakes in
nutrient ecoregion 6 with percent watershed disturbance (usually Ag) greater than 50%,
bottom sediments in this region with high % Ag would need to be at least 60 cm depth
to be in pre-settlement time (Dan Engstrom, Pers. comm.); and all cores in this
ecoregion, regardless of percent watershed disturbance, that were less than 30 cm long
were not considered for analysis.
Uncertain
Lakes that were considered uncertain were as follows: man-made lakes
(reservoirs); date of formation (e.g., dam building) was not known; sedimentation rates
were also unknown, and could potentially vary over a wide range; and all oxbow lakes
(as determined using Google Earth).
Development of Indices for Lakeshore and Littoral
Habitat Condition
Introduction
The physical habitat of a lake includes the environment at the bottom of the lake
(substrate), the vegetation and substrate along its shoreline (riparian zone), and the
biotic and abiotic structure of the near shore water (littoral zone). Physical habitat
condition is critically important to benthic communities, fish and other aquatic
organisms.
The NLA and other lake survey and monitoring efforts increasingly rely upon
biological assemblage data to define lake condition. Information concerning the multiple
dimensions of physical and chemical habitat is necessary to interpret this biological
information and comprehensively assess ecological condition. The controlling influence
of littoral structure and complexity on lake biota has been long recognized, and recent
research highlights the roles of littoral woody debris in providing refuges from predation
and affecting nutrient cycling and littoral production. National Lake Assessment field
crews characterized lake depth, water surface characteristics, bank morphology and
33
-------
evidence of lake level fluctuations, littoral and shoreline substrate, fish concealment
features, aquatic macrophytes, riparian vegetation cover and structure, and human land
use activities. These littoral and riparian physical habitat measurements and visual
observations were made in a randomized array of 10 littoral plots (10m x 15m) with
adjoining riparian plots (15m x 15m) systematically spaced along the shoreline of each
sample lake. Metrics describing a rich variety of lake characteristics were calculated
from this raw data, and many of these were determined with moderate precision in the
national dataset.
For the NLA, we summarized the shoreline and littoral physical habitat
information with four integrative measures of lake condition: RipDist, incorporating
measures of the extent and intensity of human land use activities; RipVeg, incorporating
the structure and cover in three layers of riparian vegetation, including inundated
vegetation; LitCvr, a combined biotic cover complexity measure including large woody
snags, brush, overhanging vegetation, aquatic macrophytes, boulders, and rock ledges;
and LitRipCvr, which combines RipVeg and LitCvr in an index of the cover and
complexity of the land-water interface of lakes.
Riparian and littoral habitat structure serves as both an indicator of ecological
condition and a context for interpreting biological information. These habitat
components are important to lake biological assemblages, providing refuge from
predation, living and egg-laying substrates, and food. Shoreline structure also affects
nutrient cycling, littoral production, and sedimentation rates. Human activities along
lakeshores often adversely affect these ecosystem functions by reducing habitat
complexity. Compared with riparian and littoral conditions in lesser disturbed reference
lakes throughout the U.S.A, lakes with moderate or high human disturbances in the
same region have reduced cover and extent of multi-layered riparian vegetation or
natural wetlands. Those with moderate or high disturbance generally also have
reduced snag, brush and emergent aquatic macrophyte cover. Our general expectation
is that wetland and multi-layered riparian vegetation and abundant, complex fish
concealment features foster native fish, macroinvertebrate, and avian assemblage
diversity, whereas extensive and intensive shoreline human activities that reduce
natural riparian vegetation and reduce littoral cover complexity are probably detrimental
to native biota.
Our physical habitat assessment approaches and expectations are based on
previous research. In Midwestern lakes, Christensen et al. (1996) reported negative
associations between lakeshore cabin development and the density of riparian trees
and littoral coarse woody debris, and Jennings et al. (1999) reported cumulative
negative effects on fish assemblages as riparian alteration increased. More recent
Wisconsin lake studies have found reductions in the quantity of woody debris, and the
cover of emergent and floating aquatic macrophytes with increases in cumulative
lakeshore human development (Jennings et al. 2003; Hatzenbeler et al. 2004).
Radomski and Geoman (2001) also reported loss of emergent and floating-leaf
vegetation as a result of human lakeshore development in upper midwest lakes. In the
Northeast, shoreline disturbance has been associated with the decline of species
34
-------
richness of native minnows and with an increase in nonnative predator fish species
(Whittier et al. 1997). Halliwell (2007,2008) described a number of recent and ongoing
studies showing detrimental effects on fish and their habitat resulting from human
development of shorelines. Among these, an ongoing study by Merrell et al. (2008,
2009) showed that human development of shorelines resulted in a decrease of woody
debris (snag habitat), an increase in sandy shorelines, and an increase in submerged
aquatic macrophyte cover in Vermont lakes. Wagner et al. (2006) reported negative
effects of residential lakeshore development on littoral fishes and habitat, citing their use
of near-shore habitat for nesting, foraging and refuge from predators and adverse
conditions. Ness (M.S. 2006 Univ. of Maine) reported that both riparian (shore) and
littoral habitat complexity was simplified (at the site scale), with lower densities of trees
and shrubs, aquatic macrophytes, and in-lake coarse woody debris.
In an early probability survey of Northeastern lakes, Whittier et al. (2002)
reported population estimates of the number of lakes with no direct evidence of human
activities in 27% (± 9%). The methods used in this survey (Kaufmann and Whittier,
1997) were modified by Rowan et al. (2006) for use in surveys serving the needs of the
European Union's Water Framework Directive, and by the USEPA (2007) for use in the
present NLA. Based on these methods, Whittier reported that 67% of NE lakes had
relatively undisturbed shorelines, and at these lakes the median canopy-layer tree cover
was 67%, with a median combined canopy-layer, mid-layer, and ground-layer woody
cover of 170% (of a possible 300%), indicating substantial structural complexity and the
potential for sustaining that complexity over time. At the other end of the spectrum, 23%
(± 10%) of lakes had at least one type of human structure or activity at half or more of
the shoreline stations. This human activity was associated with reduced canopy-layer
cover (median = 35%) and three-layer woody cover (median = 82%). Half of these lakes
had buildings at more than a third of the shoreline stations. Habitat complexity, in the
form of woody snags, overhanging trees, and aquatic plants, was markedly reduced at
lakes with higher levels of human activity along the shoreline. The national findings of
the NLA reinforced and expanded the geographic extent of these earlier findings for
Northeastern U.S. reported by Whittier et al. (2002), who concluded that although
stressors such as non-native fish introductions, mercury contamination, and shoreline
alteration were not generally considered subjects for environmental management, they
were as widespread as eutrophication, and more extensive than acidification, in the
lakes of their survey.
Lakeshore disturbances caused by human activity are direct stressors to littoral
and riparian habitat, and range in impact from minor effects (such as removal of small
areas of riparian tree cover to develop a picnic area) to major alterations (such as
construction of a large year-round lakeshore home complete with retaining walls,
unstable and erosive landscaping, and concrete shoreline walls and docks).
Publications of the effects of lakeshore development on the integrity of littoral habitat
present a compelling story. Effects like sedimentation, loss of native plant growth,
alteration of native plant communities, loss of habitat structure, and modifications to
substrate types are all commonly associated with shoreline human development
(Christensen et al., 1996; Whittier et al., 1992, Engel and Pederson, 1998, Merrell et al.,
35
-------
Human activities along lakeshores often adversely affect shoreline and littoral
structure and ecosystem function by reducing habitat complexity. For example, in the
presence of human activity, habitat complexity, in the form of woody snags,
overhanging trees, and aquatic plants, becomes markedly reduced, resulting in impacts
to macrophytes, fish, and other aquatic biota (e.g., Wagner et al., 2006, Taillon and Fox,
2004, Engel and Pederson, 1998; Whittier et al., 2002). In the following sections, we
describe an index of shoreline disturbance and three measures of shoreline habitat:
shoreline (or riparian) vegetation cover and complexity, littoral habitat structure and
cover for biota, and a combined index of lakeshore and littoral habitat cover and
structural complexity.
Lakeshore Disturbance Index
The Lakeshore Distrubance Index RDisJX was based on field observations
tallying the presence and proximity of 12 types of human activities or disturbances at 10
systematically located shoreline positions. RDisJX incorporates both the extent of
human activities and the intensity of those activities. The extent was expressed simply
as the proportion of the shoreline stations that have at least one type of human activity
recorded within their 15 x 15 m shore plot and adjacent 10 x 15 m littoral plot
(hifpAnydrca). The intensity of human disturbances was expressed by the mean
proximity-weighted tally of the number of types of human land-use activities per
observation station, both agricultural (hiiAg) and non-agricultural (hiiNonAg), where
disturbances observed outside of the plots were given half the weight of those within the
shoreline-littoral plots.
The field procedures tallied nine designated non-agricultural human
disturbances: buildings, commercial developments, parks/man-made beaches,
docks/boats, walls/dikes/revetments, trash/landfill, road/railroad, power lines, and lawns.
Similarly field crews tallied three designated types of agricultural disturbances: row
crops, pasture/range/hayfield, and orchard. The field procedures classified Agricultural
disturbances into one-third as many categories as for non-Agricultural types.
Consequently hiiAg ranges from 0 to 1.1, whereas hiiNonAg has a range 5.6 times as
great (0 to 6.4). To avoid over-representing non-agricultural disturbances and under-
representing agricultural disturbances in RDisJX, the disturbance intensity tallys for
agricultural land use were weighted by 5x. This weighting effectively scales agricultural
land-uses equal in disturbance potential to those for non-agricultural land use. The
index is scaled from 0 to 1, where 0 indicates absence of any human disturbances and
1 indicates extremely high disturbance.
RDisJX = 1 -11 / M + hiiNonAg + (5 x hiiAg )]} + hifpAnvdrca
2
The same formulation of RD/s_/Xwas applied to all ecoregions and all multivariate lake
type classes ("CLUS") in the NLA.
36
-------
Condition Criteria for RDisJX:
We applied uniform condition criteria within the NLA:
Low Disturbance RDisJX 20 but < 0.75
High Disturbance RDisJX >0.75
Whittier et al (2002) defined thresholds of 25 and 50% of shoreline with human activities
as "low" and "high" in an assessment of Northeastern U.S. lakes. The NLA metric
hifpAnydrca, a measure of the proportion of lakeshore with one or more disturbance
types, is directly comparable to Whittier et al's measure and uses the same field
methods. RDisJX is not directly comparable, as it incorporates both the extent and
intensity of human activities in its calculation. However, disturbance extent and intensity
are correlated, and Table A-12 below shows the regression association of RD/s_/Xwith
hifpAnydrca in the NLA dataset. Whittier et al's 25 and 50% thresholds roughly
correspond with RDisJX thresholds set at 0.34 and 0.54.
Table A-12. Regression association of RDisJX with hifpAnydrca
hifpAnydrca
0.00
0.25
0.50
0.75
0.80
1.00
RDIS_
IX
0.13
0.34
0.54
0.74
0.78
0.94
Thresholds of RDisJX (0.20 and 0.75) used in the NLA correspond to hifpAnydrca
values of 0.08 and 1.00. The NLA thresholds are more stringent at both ends of the
disturbance spectrum than those of Whittier et al. (2002), in the sense that they identify
lakes with both lesser and greater amounts of shoreline disturbance. Lakes with
RDisJX <0.20 have very low levels of lake and near-lake disturbance, and those with
RDisJX >0.75 have very high levels of disturbance.
Lakeshore Habitat Index
Indices of riparian cover and complexity were based on visual estimates of
vegetation cover and structure in three vegetation layers at 10 evenly-spaced 15 x 15 m
plots adjacent to the lake shore. Field data used to calculate indices of riparian cover
and complexity included cover-class estimates of large (>0.3m dbh) and small diameter
(<0.3m dbh) tree cover in the >5m high vegetation layer, woody and non-woody
vegetation in the mid-layer (0.5 to 5 m), and woody, non-woody, inundated, and barren
37
-------
classes in the ground cover layer (<0.5 m) of the 10 lakeshore plots. These vegetation
classes are virtually identical to those described by Kaufmann et al. (1999) for streams,
as is the procedure for converting cover class data to estimates of mean cover for all
the types and combinations of vegetation. For each vegetation cover type, field crews
estimated areal cover in five classes: absent (0), sparse (0-10%), moderate (10-40%),
heavy (40-75%), and very heavy (>75%). Based on cover estimation techniques
described by Daubenmire (1968), lakeshore vegetation metrics were calculated by
assigning cover class midpoint values (i.e., 0%, 5%, 25%., 57%, and 87.5%) to each
plot's observations and then averaging those cover values across all 10 stations.
Three RVegQ index formulations were used in the NLA, and were assigned by
aggregated ecoregion (ECOWSA9). To calculate lake Riparian Vegetation Cover-
Complexity (RVegQ) indices, sub-metrics were scaled from 0-1, using dataset
maximum values. The summary indices were calculated as the mean of their weighted
subcomponents, so also vary from 0 to 1.
RVegQ_2 sums the woody cover in three lakeside vegetation layers and includes
inundated vegetation as a positive characteristic. This is appropriate for moist
ecoregions (NAP,SAP,UMW,CPL) where tree vegetation can be expected in relatively
undisturbed locations.
RVegQ_2 = {( rviwoodv 12.5) + rvfcGndInundated}
2
RVegQ_7 accommodates lack of tree canopy in ref sites by summing only lower
two layers of woody vegetation, where rviLowWood = rvfcGndWoody + rvfcUndWoody.
It also includes inundated vegetation as a positive characteristic. This index is
appropriate for plains ecoregions (NPL, SPL) where tree canopy may not be expected
in the absence of human activities, or where presence of tree canopy or enhanced tree
canopy cover around lakes may be associated with human activities (e.g.
TPL,NPL,SPL).
RVegQ_7= l( rviLowWood 11.75) + rvfcGndlnundated}
2
RVegQ_8 sums the woody cover in three lakeside vegetation layers and includes
inundated vegetation as a positive characteristic. In contrast to RVegQ_2, this index
also includes the presence of large diameter trees and accommodates bedrock and
boulders as natural shoreline. Sub-metric ssiNATBedBId is an index of natural rock
shoreline that precludes vegetation. It is calculated as ssiNATBedBId = ssfcbedrock +
sfcboulders, but the value of ssiNATBedBId is set to 0 in lakes that have a substantial
amount of anthropogenic seawalls and revetment (i.e., ifhipwWalls >0.10). This index is
appropriate for ecoregions that have potential to grow large diameter trees, are
relatively arid, or lack vegetated lake shorelines at high elevations (WMT, XER).
38
-------
RVegQ_8 = Urviwoodv/2.5} + rvfpCanBici + rvfcGndlnundated + ssiNATBedBId}
4
The formulation of RvegQ is assigned by aggregate ecoregion (ECOWSA9):
For ECOWSA9 = NAP, SAP, UMW, CPL: RVegQ = RVegQ_2
For ECOWSA9 = NPL, SPL, TPL: RVegQ = RVegQ_7
For ECOWSA9 = WMT, XER: RVegQ = RVegQ_8
Shallow Water Habitat Index
Indices of littoral cover and complexity were based on field visual estimates of
the areal cover of 10 types of littoral cover features within each of ten littoral plots (15 x
10 m) adjacent to the shoreline, and spaced evenly around the periphery of each lake.
Field data included cover-class estimates of Woody snags >0.3m diameter, Wood and
brush <0.3m diameter, inundated live trees >0.3m diameter, inundated aquatic and
herbaceous vegetation, overhanging vegetation <1m above water surface, rock ledges,
boulders, and human structures, plus a separate estimation of floating, emergent, and
submergent aquatic macrophytes. The cover classes used by the field crews are
identical to those described above for riparian vegetation and applied by Kaufmann et
al. (1999) to streams, as are the procedures for converting cover class data to estimates
of mean cover for all the types and combinations of fish concealment and aquatic
macrophyte cover. For each vegetation cover type, field crews estimated areal cover in
five classes: absent (0), sparse (0-10%), moderate (10-40%), heavy (40-75%), and very
heavy (>75%). Littoral cover metrics were calculated by assigning cover class midpoint
values (i.e., 0%, 5%, 25%., 57%, and 87.5%) to each plot's observations and then
averaging those cover values across all 10 stations. Mean cover values for various
types of cover were summed to yield combined cover metrics.
Three Shallow Water Habitat (LitCvrQ) index formulations were used in the NLA,
and were assigned by lake type class ("CLUS"). To calculate LitCvrQ indices, sub-
metrics were scaled from 0-1 using dataset maximum values. The summary indices
were calculated as the mean of their weighted subcomponents, so also vary from 0 to 1.
LitCvr_B includes all types of aquatic macrophytes along with the other natural
biotic and abiotic cover structure elements included in fciNatural. Its added emphasis
on snag cover, but lack of additional emphasis on floating and emergent aquatic
macrophytes is appropriate for Clus D (warm, low conductivity lakes, mostly in CPL).
LitCvr_B = I fciNatural + ( fcfcSnaa / 0.2875 ))
2
LitCvr_C includes submerged and other types of aquatic macrophytes in
fciNatural, but its emphasis on floating and emergent forms in addition to snags is
39
-------
appropriate for CLUS A (SAP reservoirs), where presence of submerged aquatic
macrophytes indicates water clear enough (low turbidity) for submergent vegetation.
LitCvr_C = fciNatural + ( fcfcSnaa 10.2875 ) + l( amfcEmeraent + amfcFloatina ) I
1.515)
3
LitCvr_D is the appropriate littoral cover-complexity index formulation for most
lake types in the NLA, where increases in submerged aquatic macrophytes is typically
associated with nutrient inputs from human disturbances. It excludes submerged
aquatic macrophytes, but increases the weighting of floating and emergent macrophytes
in addition to snags.
LitCvr_D = ( SomeNatCvr/'\.5) + ( fcfcSnaa 10.2875) + {(amfcFltEma ) 11.515!
3
Where: SomeNatCvr = (fcfcBoulders + fcfcBrush + fcfcLedges +
fcfcLivetrees + fcfcOverhang)
amfcFltEmg = (amfcEmergent + a mfcFloafing)
The formulation of regional LitCvrQ used for each lake was assigned according
to lake type ("CLUS"), a geographically-constrained multivariate classification. Within
each of three broad geographic areas (Eastern Highlands, Plains and Lowlands, Xeric
and Mountain West), lakes were grouped by their size, depth, morphology, depth,
elevation, temperature, precipitation, calcium concentration, latitude, and longitude.
Assignments were as follows:
For CLUS = B, C, E, F, G: LitCvrQ = LitCvr_D.
For CLUS = A: LitCvrQ = LitCvr_C.
For CLUS = D: LitCvrQ = LitCvr B.
Physical Habitat Complexity Index
The Physical Habitat Complexity index (LitRipCvQ) is simply the arithmetic mean
of the respective values for the Riparian and Littoral Cover Complexity indices RVegQ
and LitCvrQ:
LitRipCvQ = (RVeaQ + LitCvrQ)
2
For example, lakes in the NPL and in the SPL lakes that are within lake type
"CLUS B" use the Riparian formulation RVegQ_2 and the Littoral formulation
LitCvrQ_D. Their formulation for combined Littoral-Riparian Cover Complexity
40
-------
LitRipCvQ is LRCvQ_2D = (RVegQ_2 + LitCvrQ_D)/2. Formulations were assigned by
Lake Class (CLUS) and Ecoregion (ECOWSA9) as follows:
For CLUS A:
For CLUS B:
For CLUS C:
in UMW, CPL:
in NPL, SPL, TPL:
For CLUS D:
in CPL:
in SPL, TPL:
For CLUS E:
in UMW:
in NPL, SPL, TPL:
For CLUS F:
For CLUS G:
LitRipCVQ = LRCVQ_2C
LitRipCVQ = LRCVQ_2D
LitRipCVQ = LRCVQ_2D
LitripCVQ = LRCVQ_7D
LitRipCVQ = LRCVQ_2B
LitripCVQ = LRCVQ_7B
LitRipCVQ = LRCVQ_2D
LitripCVQ = LRCVQ_7D
LitRipCVQ = LRCVQ_8D
LitRipCVQ = LRCVQ_8D
Physical Habitat Index Precision and its Interpretation
Physical habitat measurements were repeated at a stratified random subset of 91
NLA sample lakes during the summer 2007 index sampling period. These repeat
samples allow an assessment of the within-season repeatability of lake habitat metrics.
Table A-13 shows the precision of the four physical habitat condition indicators used in
the NLA. The basic measure of repeatability is RMSrep, the Root Mean Square of
repeat visits. The RMSrep is a measure of the absolute (unsealed) precision of
measurement, and incorporates both measurement and short-term temporal variability.
Table A-13. Precision and distribution characteristics of the four Physical Habitat
condition indicators applied in the National Lakes Assessment — Calculated for
the survey as a whole, and separately for three combined ecoregions: the Eastern
Highlands, Plains and Lowlands, and the West.
Metric
NLA: (df=91)
RDis IX
RVegQ
LitCvrQ
LitRipCvQ
EHigh:(df= 21)
RDis IX
RVegQ
LitCvrQ
LitRipCvQ
RMS™,,
0.115
0.058
0.059
0.043
0.096
0.052
0.060
0.042
S/N
4.8
2.9
2.7
3.9
7.0
2.6
1.6
2.4
Mean/Med
0.48/0.49
0.17/0.16
0.12/0.09
0.15/0.13
0.42/0.42
0.21/0.20
0.14/0.12
0.18/0.17
B&*
0-1
0-1
0-1
0-1
0-1
0-1
0-1
0-1
Rgnhs
0 - 0.947
0-0.558
0-1.0(0.79)
0-0.588
0-0.932
0-0.489
0.002-0.630
0.011-0.457
Rgobs/RMSre
D
8.2
9.6
16.9(13.4)
11.6
9.7
9.4
10.5
10.6
41
-------
PlnNLow (df=49)
RDis IX
RVegQ
LitCvrQ
LitRipCvQ
West: fdf=23)
RDis IX
RVegQ
LitCvrQ
LitRipCvQ
0.129
0.064
0.061
0.043
0.096
0.046
0.054
0.041
3.5
1.9
3.4
4.3
7.1
6.5
0.7
3.3
0.52/0.54
0.16/0.14
0.13/0.09
0.14/0.12
0.42/0.41
0.16/0.14
0.079/0.062
0.12/0.11
0-1
0-1
0-1
0-1
0-1
0-1
0-1
0-1
0-0.936
0-0.558
0-1.0(0.79)
0-0.588
0 - 0.947
0-0.491
0 - 0.423
0-0.421
7.3
8.7
16.4(13.0)
13.7
9.9
10.7
7.8
10.3
The RMSrep for a metric (column 1 of Table A-13) is an estimate of the average
standard deviation of that metric if measurements were repeated at all lakes, and
standard deviations for each lake were averaged across lakes. It is often scaled by
comparing it to some magnitude of variation that is of interest. Alternative sealers might
be the magnitude of expected change or the magnitude of an ecologically important
difference. It is often difficult to define such a change for a broad survey region. Useful
and relevant alternatives are to compare RMSrep to the potential (theoretical) range
(Rgpot in Table A-13) or the observed range (Rg0bs in Table A-13) of the metric in a
survey such as the NLA.
The ratio of Rg0bs/RMSrep for metric is an expression of its potential for
discerning differences among lakes. The last column of Table A-13, shows that the
ratio Rg0bs/RMSrep ranged from 7.3 to 18.8 the four Physical Habitat metrics used in the
NLA. These results show good potential for these variables to discern lake differences
over the ranges observed nationally and in the major subregions.
Another way of scaling the precision of habitat metrics to the "job at hand" is to
examine their components of variance. The ratio of variance among lakes to that due to
measurement (or temporal) variation within individual lakes has been termed a "Signal-
to-noise" ratio, (S/N shown in column 2 of Table A-13). One can think of S/N as the
ability of the metric to discern differences among lakes in this survey context. If the
among-lake variance in the region or nation is a meaningful variation in lake condition,
then the S/N is a measure of the ability of a metric to discern lake condition. This
variance-partitioning approach is explained in Kaufmann et al. (1999) and Faustini and
Kaufmann (2007), where the authors referred to RMSrep as RMSE and evaluated S/N in
stream physical habitat variables. In those publications, the authors generally
interpreted precision to be high relative to regional variation if S/N >10, low if S/N <2.0,
and moderate if in-between. The NLA physical habitat metrics have mostly moderate
precision in this set of data (S/N 2.7 to 4.8 in the national dataset), which means that
there can be a substantial, but not crippling influence of measurement "noise" in our
classification, regression, plots, and distributions. Larsen et al. (2004) examined the
effects of measurement imprecision on the ability of stream physical habitat metrics and
sampling designs to detect temporal trends. Kaufmann et al. (1999) and Faustini and
Kaufmann (2007) discuss the effect of various levels of S/N on classification, regression
and population estimates.
42
-------
Within subregions of the NLA, one or another of the metrics showed low S/N.
Although measured with approximately the same RMSrep in all three regions, the S/N
ratio of RVegQ ranged from 1.9 in the Plains-Lowlands to 6.5 in the West. This finding
suggests that low S/N in the Plains-Lowlands is largely an indication of low variation in
riparian vegetation as measured by RVegQ, rather than poor measurement precision.
However, we may want to explore more sensitive indicators of riparian vegetation
condition in the Plains-Lowlands to reformulate RVegQ for these regions. Similarly,
LitCvrQ was measured with approximately the same absolute precision (RMSrep) in all
three regions, but S/N ranged from 0.7 in the West to 3.4 in the Plains-Lowlands - this
time indicating lack of variation in littoral complexity in the West compared with the other
regions, rather than low absolute precision in the West. Again, however, we may want
to explore more sensitive indicators of littoral complexity in lakes of the West.
Reference Site Screening
The general NLA strategy was to base most indicator metric expectations on the
distribution of indicator values observed in least-disturbed reference sites. These
expectations were either site-specific predictions derived by modeling the influence of
important non-anthropogenic environmental factors, or were simple statistics concerning
the central tendency and distribution of metric values in reference sites within an
appropriate region or class of lakes. For biotic and habitat indicators, lakes were first
classified into one of seven types, CLUS A through G, based on a geographically-
constrained multivariate classification. Within each of three broad geographic areas
(Eastern Highlands, Plains and Lowlands, Xeric and Mountain West), lakes were
grouped by their similarity in size, depth, morphology, depth, elevation, temperature,
precipitation, calcium concentration, latitude, and longitude (see previous sections of
this appendix for details). Within each lake type (CLUS), we attempted to obtain
approximately 20 or more reference lakes by choosing the least disturbed lakes on the
basis of chemical variables and direct observations of agricultural and non-agricultural
human disturbances along the lake margin. For each group (CLUS), a series of
reference threshold concentrations were established. These varied by group to account
for regional variations in water chemistry and littoral-riparian disturbances. Any lake
sampled in the survey was considered to be reference if it met every threshold
established for the relevant group. Screening parameters were: total phosphorus; total
nitrogen; chloride; total sulfate; acid neutralizing capacity, dissolved organic carbon;
dissolved oxygen in the epilimnion; proportion of lakeshore with non-agricultural
disturbances; proportion of lakeshore with agricultural disturbances; and the relative
extent and intensity of human influences of all types together. Following this process,
170 reference lakes (CLASS = Yes) were identified for the entire survey, of which 160
had useable field physical habitat data.
For use in setting lake physical habitat expectations, we modified reference
CLASS (used for biological indicators throughout the NLA), by excluding lakes with level
control structures in addition to evidence of large lake level fluctuations visible in aerial
43
-------
photos and field measurements (bfxhorizdist >10m). This lake level fluctuation screen
(CLASSP) dropped 11 reference sites from the set used by biological indicators,
including 6 in WMT, 1 in XER, 1 in SPL, 3 in SAP; but none in NAP, NPL, TPL, and
CPL, and UMW. With one exception in the UMW, all reference sites with the
bfxhorizdist >1 Om also had level-control structures and extensive "bathtub rings" visible
in aerial photographs (e.g. Cle Elum Lake in Washington). The one site in the UMW
ecoregion that met the bfxhorizdist >10m criterion was not dropped as a reference site
because it was a natural lake on very flat topography, and had no level control
structures. The revised reference screen yielded 149 least-disturbed reference lakes
(CLASSP = Yes), of which 107 were in the UMW, WMT, and NAP (Table A-14). It was
difficult to find sites of minimal reference quality in the NPL, SPL, TPL, and XER
ecoregions.
Habitat Indicator Expectations
The paucity of reference sites necessitated some analytical strategies for
deriving expectations for the various physical habitat indices. First was to coalesce
reference sites from relatively similar regions (NPL, SPL, TPL combined into
"CENPLN"), or regions where sites varied as much within regions as between regions
(WMT and XER combined into "WEST"). The second strategy was to attempt to
normalize reference distributions and estimate their percentiles rather than use simple
non-parametric statistics (e.g., 5th, 25th percentiles); we did this by log-transforming to
approximate normal distributions and to calculate the logarithmic (geometric mean) and
logarithmic standard deviations to estimate the percentiles of the reference distributions.
The third strategy was to alter the habitat metrics to take into account subregional
differences (see above discussion of revisions in RVegQ, LitRipCvrQ to accommodate
different regional expectations concerning tree canopy vegetation in the Central Plains
regions: NPL, SPL and TPL). Lastly, we calculated site-specific expectations for all
lakes in the WEST (MTN and XER), based on their latitude, elevation, and subregion.
For the Western United States (ECOWSA9= WMT or XER), expected values of RVegQ,
LitCvrQ, and LitRipCvrQ were modeled by multiple linear regression predictions based
on lake elevation, latitude, and subregion (WMT versus XER) in all sites of the WEST
excluding highly disturbed lakes (Table A-15). It was necessary to define the controlling
effects of these natural factors on a larger set of lakes than only the relatively small
number of reference lakes. However, calculating ratio of observed/expected values of
the metrics based on moderately disturbed lakes in addition to reference lakes
necessitated refining the 0/E expectations. We did this by examining the distribution of
0/E values in Western reference sites — the expectation was the geometric mean 0/E
in reference sites. In all the other regions of the NLA, the geometric mean of the
distribution of RVegQ, LitCvrQ, and LitRipCvrQ values in regional reference sites were
set as the expected values for sites within their respective regions. Table A-16
summarizes the reference expectations for RVegQ, LitCvrQ, and LitRipCvrQ in the nine
ecoregions of the NLA.
44
-------
Table A-14. Number of regional reference sites used for setting Physical Habitat
Expectations and Condition
Ecoregion (ECOWSA 9) Reference sites (CLASSP = Yes)
NAP
SAP
UMW
CPL
CENPLN:
TPL
NPL
SPL
WEST:
MTN
XER
28
14
41
13
12
41
(5)
(1)
(6)
(38)
(3)
149
Setting Condition Criteria
Recall that the high (>0.75) and low (<0.20) disturbance criteria for the direct field
quantification of lakeshore human disturbances (RDisJX) were set independently of
reference site distributions. By contrast, reference site distributions were used to set
condition criteria for RVegQ, LitCvrQ, and LitRipCvrQ. The metrics we examined were
observed/expected values (0/E) for each metric, calculated by dividing observed values
for each lake by its appropriate expected value. In the WMT and XER, the expectations
were lake-specific values calculated from the regression models in Table A-15. For the
other regions, the expectations were the region-specific geometric mean values of the
respective regional reference lake distributions shown in Table A-16.
We recognize that there is natural variability in expectations that is not captured
by our modeling in the WEST, and certainly not by the single regional geometric mean
values that serve as expected values in the other regions. Consequently, we compared
each lakes 0/E value with the statistical distribution of 0/E values in the set of regional
reference sites. Lakes with 0/E values lass than the 5th percentile of the reference
distribution were scored in "poor" condition, whereas those above the 25th percentile
were scored in "good condition. Those with 0/E values between the 5th and 25th
Table A-15. Multiple Linear Regression Models used for calculating reference
expectations for Physical Habitat in Western Region (MTN, XER) Lakes
Riparian Vegetation:
Expected Log(RVegQ_8) = -1.2108 - (0.000037*ELEV_PT) + (0.0126*FLD_LAT) + (0.1112*WMT)
Rsq=0.2364 RMSE=0.1715 p<0.0001 Model=3 Error=122 CTotal=125 — excludes CLASSP=Trash,
bfxhorizdist>=10m, and 16 low outliers outside of bounds of ref residuals;
45
-------
Littoral Cover:
Expected Log(LitCvrQ_D) = -0.9738 - (0.000073*ELEV_PT);
Rsq=0.0500 RMSE=0.3145 p=00075, Model=ldf Error=140 CTotal=141 — excludes CLASSP=Trash,
bfxhorizdist>= 1 Om;
Littoral-Riparian Cover:
Expected Log(LitRipCvQ_8D =
-1.0751-(0.000038*ELEV_PT)+(0.0083*FLD_LAT)-(0.000079*XER_X_ELEV);
Rsq=0.1927 RMSE=0.1869 p<0.0001 Model=3 Error=129 CTotal=132 — excludes CLASSP=Trash,
bfxnorizdist>=10m, and 9 low outliers outside of bounds of ref residuals;
percentiles of the reference distribution were rated in "fair" condition (Table A-17). Note
that because the reference lake sample sizes were small, the Log(0/E) mean (i.e.,
geometric mean) and its standard deviation were used to estimate these percentiles.
PHAB Metric Performance
Examination of the lakeshore human disturbance metric RDisJX across all nine
NLA ecoregions (unweighted sample statistics) shows a wide distribution in all
ecoregions, but notably higher disturbance in the NPL and lowest disturbance in the
WMT and NAP (Figure A-7). Similarly, Riparian Vegetation (RVegQ) metric values
were highest in the WMT and NAP and lowest in the NPL (Figure A-8A). Although less
pronounced, this pattern persisted after RVegQ was transformed into 0/E values
(Figure A-8B).
Littoral Habitat Cover and Complexity (LitCvrQ) was notably higher in the Coastal
Plain (CPL) than in any of the other regions (Figure A-9A). Lakes in the Northern
Plains, by contrast had the lowest LitCvrQ, although sample distributions of LitCvrQ
were relatively low throughout the other inland plains ecoregions (SPL, TPL). After
scaling LitCvrQ as the 0/E variable LitCvr_OE, many ecoregions (e.g., WMT, CPL,
NAP) had sample median 0/E near 1.0 (Figure A-9B). The Northern Plains still had the
lowest sample median.
Table A-16. Reference expectations for RVegQ, LitCvrQ, and LitRipCvrQ in the
nine ecoregions of the NLA.
Ecoregion Metric Metric Basis Expected value
NAP
SAP
UMW
CPL
RVegQ
RVegQ
RVegQ
RVegQ
Log RVegQ 2
Log RVegQ 2
Log RVegQ 2
Log RVegQ 2
0.268
0.235
0.252
0.290
46
-------
TPL
NPL
SPL
MTN
XER
RVegQ Log RVegQ 7 0.176
RVegQ Log RVegQ_7 0.176
RVegQ Log RVegQ_7 0.176
RVegQ Log RVegQ 8 Lake-specific/(Elevation, Latitude, ECOWSA9)
RVegQ Log RVegQ 8 Lake-specific/(Elevation, Latitude, ECOWSA9)
NAP
SAP
UMW
CPL
TPL
NPL
SPL
MTN
XER
NAP
SAP
UMW
CPL
TPL
NPL
SPL
MTN
XER
LitCvrQ
LitCvrQ
LitCvrQ
LitCvrQ
LitCvrQ
LitCvrQ
LitCvrQ
LitCvrQ
LitCvrQ
LitRipCvrQ
LitRipCvrQ
LitRipCvrQ
LitRipCvrQ
LitRipCvrQ
LitRipCvrQ
LitRipCvrQ
LitRipCvrQ
LitRipCvrO
Log LitCvrQ D
Log LitCvrQ C(D)
Log LitCvrQ D
Log LitCvrQ B(D)
Log LitCvrQ D(B)
Log LitCvrQ D
Log LitCvrQ D(B)
Log LitCvrQ D
Log LitCvrQ D
Log LitRipCvrQ 2D
Log LitRipCvrQ 2C(D)
Log LitRipCvrQ 2D
Log LitRipCvrQ 2B(D)
Log LitRipCvrQ 7D(B)
Log LitRipCvrQ 7D
Log LitRipCvrQ 7D(B)
Log LitRipCvrQ 8D
Loa LitRipCvrO 8D
0.147
0.191
0.169
0.299
0.114
0.114
0.114
Lake-specific /(Elevation)
Lake-specific /(Elevation)
0.214
0.231
0.220
0.305
0.153
0.153
0.153
Lake-specific /(Elevation, Latitude, ECOWSA9 x Elev)
Lake-specific /(Elevation, Latitude, ECOWSA9 x Elev)
The Coastal Plain (CPL) had the highest combined Littoral-Riparian Cover
(LitRipCvrQ), followed by the NAP, UMW, SAP and WMT (Figure A-10). The NPL had
the lowest median LitRipCvrQ, with low medians also in the XER and the other inland
plains ecoregions. Once scaled as the 0/E variable (LitRipCvr_OE), the WMT, NAP,
CPL and UMW had median values approaching 1, but lakes in the NPL remained
substantially below 0.5 (Figure A-1 OB).
There were clear, usually progressive, and often substantial declines in habitat
cover and complexity (RVegQ, LitCvrQ, and LitRipCvrQ and their 0/E transforms) from
minimally disturbed to highly disturbed lakes in most of the nine NLA ecoregions
(Figures A-11). While scaling these variables as 0/E values masked natural differences
in cover and complexity among regions, it facilitated comparisons of condition and
impairment across regions (see especially Figure 6). The weakest relationships to
disturbance were generally seen in the littoral cover index (LitCvrQ and LitCvr_OE),
especially in the CPL and the two western ecoregions (Figures A-12 and -14). Very
strong contrasts in RVegQ (and its 0/E transform) were seen in many regions,
especially CPL, UMW, WMT (Figures A-12 and -14). LitRipCvrQ and its 0/E transform
showed strongest declines with disturbance in the UMW, SPL and WMT (Figures A-12,
-13, -14).
47
-------
Table A-17. Condition criteria for rating lake condition as good, fair and poor.
The 5th and 25th percentiles of the reference O/E distributions, respectively, were
set as the upper bounds for poor and fair condition. These percentiles were
estimated, respectively, the log mean minus 1.65 and 0.67 times the log standard
deviation of the reference distribution of the habitat metric shown. They are
expressed as antilogs of those values, i.e., as O/E fractions
Ecoregion
CENPLN
WEST
NAP
SAP
CPL
UMW
CENPLN
WEST
NAP
SAP
CPL
UMW
CENPLN
WEST
NAP
SAP
CPL
UMW
Metric
Rveg_OE
Rveg_OE
Rveg_OE
Rveg OE
Rveg_OE
Rveg_OE
LitCvr OE
LitCvr OE
LitCvr OE
LitCvr OE
LitCvr OE
LitCvr OE
LitRipCvr_OE
LitRipCvr_OE
LitRipCvr_OE
LitRipCvr_OE
LitRipCvr OE
LitRipCvr_OE
O/E 5th
0.548864321
0.573040082
0.616062821
0.548655832
0.52679907
0.590622517
0.277411845
0.271963111
0.469158944
0.338649158
0.46826856
0.415609054
0.510252334
0.578149294
0.668243895
0.64137939
0.586374398
0.634351542
O/E 25th
0.783803142
0.866209615
0.821438398
0.783682231
0.770851994
0.807491575
0.594121132
0.591858651
0.735420954
0.644243433
0.734853894
0.700104711
0.76092701
0.860819309
0.84901038
0.834981782
0.805128129
0.831254482
48
-------
3_NPL 4_SPL 5_TPL 6_CPL 7_UMW 8_NAP 3_SAP
ECO 9
Figure A-7. Comparison of Lakeshore Disturbance (RDisJX) across nine NLA
ecoregions
49
-------
rf..
1_V*1T 2 _XER 3 _N PL
4_SPL 5_TPL
ECO 9
S_CPL_ 7_UMW 8_NAP 9_SAP
y
e
g
0
E
•
•
I
•
•
III li n B
H fn B ill H ff It
fflfffflf
1_V*1T 2_XER 3_NPL 4_SPL E_TPL 6_CPL 7_UMW 8_NAP 9_SAP
EC09
Figure A-8. Comparison of A) Observed, and B) Expected , Lakeshore Riparian
Vegetation Structure and Cover (RVegQ) across nine NLA ecoregions
50
-------
I
2^
-------
M
2JXER 3_NPL 4_SPL 5_TPL 6_CPL
EC09
S_NAP 9_SAP
Figure A-10. Comparison of A) Observed, and B) Expected, Littoral and Riparian
Habitat Structure and Cover (LitRipCvrQ) across nine NLA ecoregions
52
-------
REF SO-SO TOASH REF SO-SO TRASH
NAP NAP NAP SAP SAP SAP
Q 0.2-
REF SO-SO TRASH REF SO-SO TRASH
NAP NAP NAP SAP SAP SAP
p 0. 2
C
REF SO-SO TRASH REF SO-SO TRASH
REF SO-SO TRASH REF SO-SO TRASH
NAP NAP NAP SAP SftP SAP
REF SO-SO TRASH REF SO-SO TRASH
NAP NAP NAP SAP SAP SAP
REF SO-SO TRASH REF SO-SO TRASH
NAP NAP NAP SAP SAP SAP
Figure A-11. Contrasts in scaled and unsealed Physical Habitat metrics among
reference (Ref), moderately disturbed (So-So), and highly disturbed (Trash) lakes in the
Eastern Highlands (Northern and Southern Appalachians)
T
REF SO-SO TRASH REF SO-SO TRASH
CPL CPL CPL UMW UMW UMW
REF SO-SO TRASH REF SO-SO TRASH
CPL CPL CPL UMW UMVJ UMW
REF SO-SO TRASH
REF SO-SO
UMW UMW
REF SO-SO TRASH REF SO-SO TRASH
CPL CPL CPL UMW UMW UMW
REF SO-SO TRASH REF SO-SO
CPL CPL CPL UMW UMW
REF SO-SO TRASH REF SO-SO TRASH
CPL CFL CPL UMW UMW UMW
Figure A-12. Contrasts in scaled and unsealed Physical Habitat metrics among
reference (Ref), moderately disturbed (So-So), and highly disturbed (Trash) lakes in the
Coastal Plain and the Upper Midwest
53
-------
REF SO-SO WASH RET SO-SO TPAIH KEF
HPL HFL WL m W. IFL TPL
. • - - {1
)"SD TPJkStl *£' 50-50 WASH D.EF 5O-S) Tft«5H
-------
REF SO-SO TRASH REF SO-SO TRASH
VA1T VWIT WMT XER XER XER
REF SO-SO TRASH REF SO-SO TRASH
WMT WIT WMT XER XER XEP
I
REF SO-SO TRASH REF SO-SO TRASH
WMT WIT WT XER XER XER
REF SO-SO TRASH REF SO-SO TRASH
SO-SO TRASH
Figure A-14. Contrasts in scaled and unsealed Physical Habitat metrics among
reference (Ref), moderately disturbed (So-So), and highly disturbed (Trash) lakes in the
Central Plains (Northern, Southern, and Temperate Plains)
NLA Index Precision and Interpretation
Lake condition indicators were repeated at a stratified random subset of 36 to 96
NLA sample lakes during the summer 2007 index sampling period (dfrep +1 in Table A-
15). These repeat samples allow an assessment of the within-season repeatability of
these metrics. Table A-15 shows the precision of a selection of lake condition indicators
used in the NLA. The basic measure of repeatability is RMSrep, the Root Mean Square
of repeat visits. The RMSrep is a measure of the absolute (unsealed) precision of the
whole measurement and analytical process, incorporating also short-term temporal
variability within the summer sampling period. One can envision RMSrep for a metric is
an estimate of its average standard deviation if measured repeatedly at all lakes, and
standard deviations for each lake were averaged across lakes. For Log transformed
variables, one can view the antilog of the RMSrep as a proportional standard deviation;
something like a . The antilog of 0.179 is 1.51. Then, for example the, RMSrep of 0.179
for Log10(PTL+1) means that the +/- error bound on a measurement in a lake is the
measured value times 1.51 and divided by 1.51. So, the +/-1 StdDev error bounds on
a PTL measurement of 10 ug/L during the index period is (10 -r-1.51) to (10 x 1.51) or
6.6 to 15.1.
55
-------
RMSrep is often scaled by comparing it to some magnitude of variation that is of
interest. Alternative sealers might be the magnitude of expected change or the
magnitude of an ecologically important difference. It is often difficult to define such a
change for a broad survey region. Useful and relevant alternatives are to compare
RMSrep to the potential (theoretical) range or the observed range (Rg0bs in Table A-15)
of the metric in a survey such as the NLA. The ratio of Rg0bs/RMSrep for metric is an
expression of its potential for discerning differences among lakes. The last column of
Table A-15 shows that the ratio Rg0bs/RMSrep ranged from 7.6 to 25.9 for the 11
selected NLA metrics. These results show good potential for these metrics to discern
lake differences over the ranges observed nationally.
Another way of scaling the precision of metrics to the "job at hand" is to examine
their components of variance. The ratio of variance among lakes to that due to
measurement (or temporal) variation within individual lakes has been termed a "Signal-
to-noise" ratio, (S/N shown Table A-15). One can think of S/N as the ability of the
metric to discern differences among lakes in this survey context. If the among-lake
variance in the region or nation is a meaningful variation in lake condition, then the S/N
is a measure of the ability of a metric to discern lake condition. This variance-
partitioning approach is explained in Kaufmann et al. (1999) and Faustini and Kaufmann
(2007), where the authors referred to RMSrep as RMSE and evaluated S/N in stream
physical habitat variables. In those publications, the authors generally interpreted
precision to be high relative to regional variation if S/N >10, low if S/N <2.0, and
moderate if in-between. When S/N is over about 10, the effect of measurement error on
most interpretations is nearly insignificant within the national context; between 6 and 10
these effects are minor. Between S/N of 2 and 5, the effects of imprecision should be
acknowledged, examined and evaluated. From 2 to 4 they are usually adequate to
make good-fair-poor classifications, but there is some distortion of CDFs and a
significant limitation of the amount of variance that can be explained by approaches
such as multiple linear regression (The magnitude of the within-lake variance
component limits on the amount of among-lake variance that can be explained by
multiple linear regression using single visit data).
S/N for TPL and NPL had high precision (S/N>10) in the national survey. Secchi
depth, turbidity, riparian disturbances, and the diatom IBI determined from the top of the
sediment cores all had moderately high precision (S/N 4.8-7.1). Chlorophll-a, Riparian
and Littoral habitat cover and complexity, and the sediment core bottom diatom IBI had
moderate precision in this set of data (S/N 2.0 to 3.9 in the national dataset), which
means that there can be a substantial, but not crippling influence of measurement
"noise" in classification, regression, plots, and distributions based on those variables.
Larsen et al. (2004) examined the effects of measurement imprecision on the ability of
stream physical habitat metrics and sampling designs to detect temporal trends.
Kaufmann et al. (1999) and Faustini and Kaufmann (2007) discuss the effect of various
levels of S/N on classification, regression and population estimates.
56
-------
Table A-15 Precision and distribution characteristics of diatom IBI and indices of
diatom assemblage integrity, nutrient concentrations, trophic status, water
clarity, shoreline human disturbance, and lakeshore physical habitat applied in
the National Lakes Assessment
Metric
DiatomlBI
LDC_ADJ(Top)
LDC_ADj '(Bottom)
Chem-Nutrients
Log(1+PTL)
Log(NTL)
Log(1+CHLA)
Log(SECMEAN)
Log(TURB)
PHab:
RDis IX
RVegQ
LitCvrQ
LitRipCvQ
dfrep
93
35
95
95
95
91
95
90
89
88
87
RMS rep
5.37
8.30
0.179
0.132
0.389
0.164
0.216
0.115
0.058
0.059
0.043
S/N
5.8
2.0
12.1
11.3
3.6
7.1
7.1
4.8
2.9
2.7
3.9
Mean/Median
-8.8 / -8.5
-9.7 / -9.0
1.51 71.41
2.79/2.76
1.04/0.93
0.111/0.130
0.619/0.563
0.48 / 0.49
0.17/0.16
0.12/0.09
0.15/0.13
RQobs
-50 - +38
.47 _ +45
0.00-
3.69
1.00-
4.42
0.028-
2.97
-1.40-
1.56
-0.833-
2.76
0-0.947
0-0.558
0-1.0
0-0.588
RQobs/
RMS rep
16.4
11.1
20.6
25.9
7.6
18.0
16.6
8.2
9.6
16.9
11.6
Quality Assurance Summary
The NLA has been designed as a statistically valid report on the condition of the
Nation's lakes at multiple scales, i.e., ecoregion (Level II), and national, employing a
randomized site selection process. The NLA is an extension of the EMAP methods for
assessing lakes, similar to the 1997 Northeastern Lakes Assessment; therefore, it uses
similar EMAP-documented and tested field methods for site assessment and sample
collection as the Northeast Lakes Assessment. The NLA collected data on
phytoplankton, zooplankton, sediment diatoms, water chemistry and physical habitat.
Key elements of the Quality Assurance (QA) program include:
• Quality Assurance Project Plan - A Quality Assurance Project Plan (QAPP)
was developed and approved by a QA team consisting of staff from EPA's
Office and Wetlands Oceans and Watersheds (OWOW) and Office of
57
-------
Environmental Information (OEI) and a Project QA Officer. All participants in
the program signed an agreement to follow the QAPP standards. Compliance
with the QAPP was assessed through standardized field training, site visits,
and audits. The QAPP addresses all levels of the program, from collection of
field data and samples and the laboratory processing of samples to
standardized/centralized data management.
Field training and sample collection - EPA provided training sessions
throughout the study area (with at least one instructor in each session) for all
field crew members of each field crew team. All field teams were audited on
site within the first few weeks of fieldwork. Adjustments and corrections were
made on the spot for any field team problems. To assure consistency, EPA
supplied standard sample/data collection equipment and site container
packages. 1,028 random site, reference site, and repeat site samples were
collected.
Water chemistry laboratory QA procedures - NLA used the same single
lab all water chemistry samples. The Western Ecology Division (WED) was
responsible for QA oversight in implementing the NLA QAPP and lab
standard operating procedures (SOPs) for sample processing.
Zooplankton laboratory QA procedures - NLA used four labs, all four were
audited for adherence to the NLA QAPP/SOP for benthic sample processing.
This included internal quality control (QC) checks on sorting and identification
of zooplankton and the use of the Integrated Taxonomic Information System
for correctly naming species collected, as well as the use of a standardized
data management system. Independent taxonomists were contracted to
perform QC analysis of 10% of each labs samples (audit samples).
Phytoplankton laboratory QA procedures - NLA used one lab, this lab was
audited for adherence to the NLA QAPP/SOP for benthic sample processing.
This included internal quality control (QC) checks on sorting and identification
of phytoplankton and the use of the Integrated Taxonomic Information System
for correctly naming species collected, as well as the use of a standardized
data management system. Independent taxonomists were contracted to
perform QC analysis of 10% of each labs samples (audit samples).
Sediment Diatom laboratory QA procedures - NLA used four labs, all four
were audited for adherence to the NLA QAPP/SOP for benthic sample
processing. This included internal quality control (QC) checks on sorting and
identification of sediment diatoms. Independent taxonomists were contracted
to perform QC analysis of 10% of each labs samples (audit samples).
Entry of field data - NLA used a standardized data management structure,
i.e., the same standard field forms for data collected in the field, with
centralized data entry through scanning in to electronic data files. Internal
error checks were used to confirm data sheets were filled out properly.
58
-------
Records management - These records include (1) planning documents,
such as the QAPP, SOPs, and assistance agreements and (2) field and
laboratory documents, such as data sheets, lab notebooks, and audit records.
These documents are ultimately to be maintained at EPA. All data are
archived in the STORET data warehouse at www.epa.gov/STORET.
59
-------
References
Barbour, M.T., J. Gerritsen, B.D. Snyder, and J.B. Stribling. 1997. DRAFT Revision to
Rapid Bioassessment Protocols for Use in Streams and Rivers. 841-D-97-002.
U.S. Environmental Protection Agency, Office of Water, Washington, DC.
Barbour, M.T., J. Gerritsen, B.D. Snyder, J.B. Stribling. 1999. Rapid Bioassessment
Protocols: For Use in Streams and Wadeable Rivers: Periphyton, Benthic
Macroinvertebrates, and Fish. EPA841-B-99-002. U.S. Environmental Protection
Agency, Washington, D.C.
Bauer, S.B., and T.A. Burton. 1993. Monitoring Protocols to Evaluate Water Quality
Effects of Grazing Management on Western Rangeland Streams. EPA 910/9-91-
001. U.S. Environmental Protection Agency, Region X, Seattle, WA. 166 p.
Benson, B.J. and J.J. Magnuson. 1992. Spatial heterogeneity of littoral fish
assemblages in lakes: relation to species diversity and habitat structure.
Canadian Journal of Fisheries and Aquatic Sciences 49:1493-1500.
Blocksom, K. A. 2003. A Performance Comparison of Metric Scoring Methods for a
Multimetric Index for Mid-Atlantic Highlands Streams.B.A. Markert, A.M. Breure
and H.G. Zechmeister (ed.), Bioindicators and Biomonitors, Chapter14.
Environmental Management. Springer-Verlag, New York, NY, 31(5):670-682.
Birks, H.J.B. 1998. Numerical tools in paleolimnology- Progress, potentialities, and
problems. Journal of Paleolimnology. 20: 307-332.
Charles, D.F., J.P. Smol, and D.R. Engstrom. 1994. Paleolimnological approaches to
biological monitoring. In: Biological monitoring of aquatic systems, edited by L. L.
Loeb and A. Spacie, CRC Press:Boca Raton, p. 233-293.
Christensen DL,Herwig BR, Schindler DE, Carpenter, SR. 1996. Impacts of lakeshore
residential development on coarse woody debris in north temperate lakes.
Ecological Applications 6: 1143-1149.
Cumming, B.F., Smol, J.P., Kingston, J.C., Charles, A.F., Birks, H.J.B., Camburn, K.E.,
Dixit, S.S., Uutala, A.J., and Selle, A.R. 1992. How much acidification has occurred
in Adirondack region (New York, USA) lakes since pre-industrial times? Canadian
Journal of Fisheries and Aquatic Sciences 49: 128-141.
Daubenmire, R. 1968. Plant Communities: A Textbook of Plant Synecology. Harper &
Row, New York. 300 p.
60
-------
Dixit, S.S., J.P. Smol, D.F. Charles, R.M. Hughes, S.G. Paulsen, and G.B. Collins.
1999. Assessing lake water quality changes in the lakes of the northeastern United
States using sediment diatoms. Canadian Journal of Fisheries and Aquatic
Sciences 56: 131-152.
Dunne, T., and L.B. Leopold. 1978. Water in Environmental Planning. W.H. Freeman,
New York. NY. 818 p.
Dufrene, M., and P. Legendre. 1997. Species assemblages and indicator species: the
need for a flexible asymmetrical approach. Ecological Monographs 67:345-366
Faustini, J.M. and P.R. Kaufmann. 2007. Adequacy of Visually Classified Particle Count
Statistics from Regional Stream Habitat Surveys. Journal of the American Water
Resources Association 43(5): 1293-1315.
Halliwell, D. 2007. Lake habitat measures. New England Lake News - New England
Chapter of the North American Lake Management Society 2(2): 1-4.
Halliwell, D. 2008. Lake habitat measures. Out of the Blue -A Newsletter of the Water
Quality Division,Vermont Agency of Natural Resources Department of
Environmental Conservation. 33:8-9.
Hatzenbeler, G.R., J.M. Kampa, M.J. Jennings, and E.E. Emmons. 2004. A comparison
offish and aquatic plant assemblages to assess ecological health of small
Wisconsin lakes. Lake and Reservoir Management 20:211-218.
Hill, A. B. (1965). "The environment and disease: association or causation."
Proceedings Royal Society Medicine 58: 295-300.
Hynes, H.B.N. 1972. Ecology of Running Waters. Univ. of Toronto Press, Canada.
555p.
Jennings, M.J., M.A. Bozek, G.R Hatzenbeler, E.E. Emmons, and M.D. Staggs. 1999.
Cumulative effects of incremental shoreline habitat modification on fish
assemblages in north temperate lakes. North American Journal of Fisheries
Management 19:18-27.
Jennings, M.J., E.E. Emmons, G.R. Hatzenbeler, C. Edwards, and M.A. Bozek. 2003.
Is littoral habitat affected by residential development and land use in watersheds of
Wisconsin lakes? Lake and Reservoir Management 19:272-279.
Juggins, S. 2003. C2 User guide. Software for ecological and paleoecological data
analysis and visualization. University of Newcastle, Newcastle upon Tyne, UK.
69pp.
61
-------
Karr, J.R. 1993. Defining and assessing ecological integrity beyond water quality.
Environmental Toxicology and Chemistry 12:1521-1531.
Kaufmann, P.R., P. Levine, E.G. Robison, C. Seeliger, and D.V. Peck. 1999.
Quantifying Physical Habitat in Wadeable Streams. EPA 620/R-99/003.
Environmental Monitoring and Assessment Program, U.S. Environmental
Protection Agency, Corvallis, OR.
Kaufmann, P.R. and T.R. Whittier. 1997. Habitat Assessment. Pages 5-1 to 5-26 In:
J.R. Baker, D.V. Peck, and D.W. Sutton (Eds.). Environmental Monitoring and
Assessment Program -Surface Waters: Field Operations Manual for
Lakes.EPA/620/R-97/001. U.S. Environmental Protection Agency, Washington,
D.C.
Larsen, D.P, P.R. Kaufmann, T.M. Kincaid, and N.S. Urquhart. 2004. Detecting
persistent change in the habitat of salmon-bearing streams in the Pacific
Northwest. Canadian Journal of Fisheries and Aquatic Sciences 61:283-291.
Merrell, K. et al. (2008 in Capen et al. 2008) Merrell, K., E. Howe and S. Warren. 2009.
Examining Shorelines, Littorally. Lakeline 29:1.
Radomski, P. and T.J. Geoman. 2001. Consequences of human lakeshore
development on emergent and floating-leaf vegetation abundance. North American
Journal of Fisheries Management 21:41 -46.
Rowan, J.S., J.Carwardine, R.W.Duck, O.M.Bragg, A.R. Black, J.E.J. Cutler, I. Soutar,
and P. J. Boon. 2006. Development of a technique for lake Habitat Survey (LHS)
with applications for the European Union Water Framework Directive. Aquatic
Conservation 16(6):637-657. Doi: 10.1002/aqc.786 .
Smol, J.P. 1992. Paleolimnology: An important tool for effective ecosystem
management. Journal of Aquatic Ecosystem Health 1:49-58.
Smol J.P., Wolfe A.P., Birks H.J.B., et al. 2007. Climate-driven regime shifts in the
biological communities of arctic lakes. Proceedings of the National Academy
Sciences 102: 4397-4402.
Stoermer E.F. and J.P. Smol. 1999. Applications and uses of diatoms: prologue. In
Stoermer E.F. and Smol J.P. (eds) The diatoms: Applications for the
Environmental and Earth Sciences, Cambridge, Cambridge University Press, pp 3-
8.
Taillon, D. and M. Fox. 2004. The influence of residential and cottage development on
littoral zone fish communities in a mesotrophic north temperate lake.
Environmental Biology of Fishes 71: 275-285
62
-------
ter Braak, C.J.F. and van Dam, H. 1989. Inferring pH from diatoms: a comparison of old
and new calibration methods. Hydrobiologia 178: 209-223.
ter Braak, C. J. F., and S. Juggins. 1993. Weighted averaging partial least squares
regression (WA-PLS): an improved method for reconstructing environmental
variables from species assemblages. Hydrobiolgia: 269/270: 485-502.
ter Braak, C.J.F., and P. Smilauer. 2002. CANOCO reference manual and CanoDraw
for Windows. User's guide: software for canonical community ordination. Version
4.5. Microcoputer Power, Ithaca, NY, USA.
USEPA2007 NLA Field Methods USEPA. 2007. Survey of the Nation's Lakes. Field
Operations Manual. EPA841-B-07-004. U.S. Environmental Protection Agency,
Washington, DC.
Wagner, T., A. Jubar, M. Bremigan. 2006. Transactions of the American Fisheries
Society 135:843-852.
WhittierTR, Halliwell DB, Paulsen SG. 1997. Cyprinid distributions in northeast USA
lakes: Evidence of regional-scale minnow biodiversity losses. Canadian Journal of
Fisheries and Aquatic Sciences 54: 1593-1607.
Whittier, T.R., S.G. Paulsen, D.P. Larsen, S.A. Peterson, AT. Herlihy, and P.R.
Kaufmann. 2002. Indicators of ecological stress and their extent in the population
of Northeastern lakes: a regional-scale assessment. Bioscience 52(3):235-247.
Van Sickle, J. J.L. Stoddard, S.G. Paulsen and A.R. Olsen. 2006. Using relative risk to
compare the effects of aquatic stressors at a regional scale. Environmental
Management 38:1020-1030.
Van Sickle, J. and S.G. Paulsen. 2008. Assessing the attributable risks, relative risks,
and regional extents of aquatic stressors. Journal of the North American
Benthological Society 27:920-931.
63
------- |