w EPA
United States
Environmental Protection
Agency
velopmeril
Watershed Classification Framework
for the State of West Virginia
L
J. , *.
/ i
-------
EPA/600/R-03/141
July 2004
Watershed Classification Framework for the
State of West Virginia
WV R-EMAP Final Report
Naomi E. Detenbeck
US EPA Mid-Continent Ecology Division
National Health and Environmental Effects Research Laboratory
Duluth, MN 55804
Leslie A. Jagger1
Stacey L. Stark1
OAO Corporation
Duluth, MN 55804
Matthew A. Starry2
Computer Sciences Corporation
Duluth, MN 55804
'EPA FAIR Contract, 2EPA FAIR II Contract
CR82872001
Project Officer
Naomi Detenbeck
Mid-Continent Ecology Division
National Health and Environmental Effects Research Laboratory
Duluth, MN 55804
Mid-Continent Ecology Division
National Health and Environmental Effects Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Recyded/Recydable
Printed with vegetable-based ink on
paper that contains a minimum of
50% post-consumer fiber content
processed chl ori ne free
-------
Notice
The information in this document has been funded wholly by the U.S. Environmental Protection
Agency (US EPA), including support to U.S. Geological Survey EROS Data Center (IAG DW-
14938973) and West Virginia Department of Natural Resources (CR-82872001); and contract
support to OAO Corporation under US EPA FAIR Contract 68-W5-0065, Delivery Order #24, and
Computer Science (CSC) Corporation under US EPA FAIR II Contract 68-W01/W02-032, Task
Order #024. This document has been prepared at the US EPA National Health and Environmental
Effects Research Laboratory, Mid-Continent Ecology Division, in Duluth, Minnesota, with support
from EPA FAIR II Contract 68-W01 /W02-032 to CSC. It has been subjected to review by the US
EPA National Health and Environmental Effects Research Laboratory and approved for
publication. Approval does not signify that the contents reflect the views of the Agency, nor does
mention of trade names or commercial products constitute endorsement or recommendation for
use.
Preferred citation:
Detenbeck, N.E., L.A. Jagger, S.L. Stark, and M.A. Starry. 2004. Watershed classification framework for
the state of West Virgnia: WV R-EMAP Final Report. EPA/600/R-03/141. U.S. Environmental
Protection Agency, Office of Research and Development, National Health and Environmental
Effects Research Laboratory, Mid-Continent Ecology Division, Duluth, MN.
-------
TABLE OF CONTENTS
LIST OF ACRONYMS iv
LIST OF FIGURES v
LIST OF TABLES vii
LIST OF APPENDICES vui
INTRODUCTION I
PURPOSE OF CLASSIFICATION AND INTENDED USE OF DATA 2
Purpose of watershed classification 2
Hydrology-based classification of watersheds 4
Relationship between HUCs and watersheds 5
Use of watershed classification as a sampling and assessment framework 6
METHODS 9
USGS watershed delineation 9
Delineation and coding of hydrologic units for the state of WV 9
Watershed characterization 14
Creation of sample frame and sample design for 12-digit HUCs 17
Response threshold derivation 20
CLASSIFICATION RESULTS 30
Hydrology thresholds 30
Distribution of land-use/land-cover variables across 12-digit subwatersheds in
West Virginia 33
FUTURE UPDATES AND CLASSIFICATION REFINEMENTS 45
CONCLUSIONS 50
ACKNOWLEDGMENTS 51
REFERENCES 53
APPENDIX A: METADATA for MAPS AND DATABASES, INCLUDING DATA
QUALITY INFORMATION A-l
APPENDIX B: AMLS B-l
APPENDIX C: DATABASE TABLES C-l
111
-------
LIST OF ACRONYMS
DEM Digital elevation model
DEP Department of Environmental Protection
DNR Department of Natural Resources
EDNA Elevation Dataset with National Applications
EMAP Environmental Monitoring and Assessment Program
FGDC Federal Geographic Data Committee
GIS Geographic Information System
MAIA Mid-Atlantic Integrated Assessment
NHD National Hydrography Dataset
NWBD National Watershed Boundary Database
NWI National Wetlands Inventory
PRISM Parameter-elevation Regressions on Independent Slopes Model
R-EMAP Regional Environmental Monitoring and Assessment Program
US EPA U.S. Environmental Protection Agency
USGS U.S. Geological Survey
i\
-------
LIST OF FIGURES
Figure 1. Use of watershed classification 3
Figure 2. Definition of a subwatershed region 7
Figure 3. Generic example of Pfafstetter coding for an 8-digit HUC 8
Figure 4. USGS gaging stations and associated watershed boundaries 10
Figure 5. Map of 8-digit hydrologic cataloging units (HUCs), 10-digit watersheds, and 12-
digit subwatershed units produced for West Virginia 11
Figure 6. Automated derivation of main channel slope 18
Figure 7. Major drainage basins associated with the state of West Virginia and surrounding
states, as defined by 2- and 4-digit HUCs 19
Figure 8. Example of unequal weighting of interbasin HUCs 21
Figure 9. Map of ecoregions overlapping with West Virginia watersheds 22
Figure 10. Map of land-use covering West Virginia watersheds 27
Figure 11. Map of palustrine and lacustrine classes from National Wetlands Inventory for the
state of West Virginia 29
Figure 12. Predicted 2-year flood normalized to watershed area as a function of
watershed storage 31
Figure 13. Results of Classification and Regression Tree (CART) analysis 32
Figure 14. Map of WV 12-digit HUCs coded by fraction watershed storage 34
Figure 15. Map of WV 12-digit HUCs coded by estimated fraction impervious surface area .... 35
Figure 16. Map of WV 12-digit HUCs coded by estimated fraction surface mining area 36
Figure 17. Map of WV 12-digit HUCs coded by fraction agricultural area 37
Figure 18. Map of WV 12-digit HUCs coded by fraction forested area 38
Figure 19. Map of Omernik Level III ecoregions within West Virginia with land-use summaries. 40
Figure 20. Map of WV 12-digit subwatershed units by land-use and watershed storage cksses.. . 41
-------
Figure 21. Watershed cksses associated with 12-digit HUCs in sample population for Central
Appakchkn Pkteau and Central Ridge and Valley ecoregions in West Virginia. ... 43
Figure 22. Watershed classes associated with 12-digit HUCs in sample population for Western
Allegheny Plateau ecoregion in West Virginia 44
Figure 23. Effect of exponent term for storage on shape of relationship between watershed
storage and normalized peak flow 48
Figure 24. Effect of peak flow breakpoint, exponent term, and prediction error on correct
classification rate for high peak flow ckss or misckssification rate for low peak flow
ckss 49
VI
-------
LIST OF TABLES
Table 1. Guidelines for hydrologic unit subdivision for National Watershed Boundary
Database, according to interagency protocol (FGDC 2002) 13
Table 2. Geographic information system databases used to characterize WV watersheds 15
Table 3. Definition of sample frame for R-EMAP wadeable stream sampling in
WV, 2000-2001 20
Table 4. Potential watershed characteristics related to peak flows 24
VII
-------
LIST OF APPENDICES (CD-ROM, back pocket)
APPENDIX A: INVENTORY OF MAPS AND DATABASES A-l
APPENDIX B: AMLS B-l
APPENDIX C: DATABASE TABLES C-l
VIM
-------
INTRODUCTION
The Environmental Monitoring and Assessment Program (EMAP) is a research program to
develop the tools necessary to monitor and assess the status and trends of national ecological
resources. These tools include probability-based survey designs and indicators of biological
condition. Regional EMAP (R-EMAP) is designed to evaluate how these tools can be applied at
local and regional scales to meet the management needs of the states and regions. One of the
emerging issues in monitoring programs for the states and tribes is the need to develop designs to
meet multiple objectives outlined in different sections of the Clean Water Act. To meet the
requirements of Section 305(b), probabilistic survey designs are needed to produce estimates of
regional condition with a known level of confidence. In addition, monitoring programs must
identify impaired waters and associated causes of impairment to meet the listing requirements of
Section 303 (d) of the Clean Water Act (US EPA 2002). Ideally, monitoring programs will provide
information to allow states to more efficiently plan the next round of monitoring, in order to
identify as many impaired waters of the state as possible. One approach which can accomplish both
of these needs is to incorporate risk-based categories into sampling designs, either as strata in a
random-stratified design, or as categories to which different probability-weights are assigned. This
approach allows managers to summarize not just information about regional condition, but also
information about the risk of impairment associated with different classes of aquatic resources.
A R-EMAP project was designed for the state of West Virginia to accomplish multiple
objectives, including the evaluation of a risk-based random-stratified sampling process. An
additional goal of the WV R-EMAP project was to refine the fish index of biotic integrity (IBI)
developed through EPA's Mid-Atlantic Integrated Assessment (MALA) project to produce separate
indices for different thermal classes of streams and specific to the state of West Virginia. The
purpose of this report is to describe the methods involved in producing a preliminary watershed
classification and incorporating the risk-based watershed classification into a monitoring design for
-------
the state of West Vkginia. Validation and refinement of the watershed dassification scheme, based
on analysis of pending monitoring data, will be addressed in a subsequent report.
PURPOSE OF CLASSIFICATION AND INTENDED USE OF DATA
Purpose of watershed classification
Historically, classification systems have been used for inventory purposes, for stratifying
landscapes prior to characterizing reference or regional condition, and for facilitating
communication with natural resource managers and the public (Omernik and Gallant 1988;
Heiskary and Wilson 1990; Herlihy et al. 2000; Pan et al. 2000; Waite et al. 2000). More recently, it
has become necessary to develop classification systems that can explain differences in vulnerability
of aquatic ecosystems to stressors, aid in regionalizing water quality criteria, and predict probability
and causes of impairment of aquatic systems (US EPA 2002).
Most sources of stream impairment contained in state 303(d) listings are related to nonpoint
source pollution (US EPA 2001, 2003). To more efficiently deal with water quality management
issues, an integrated approach to small watershed assessment, diagnosis, and restoration planning is
needed (Figure 1). Current guidance from US EPA supports the development of a consolidated
assessment and listing approach to enable joint application of Clean Water Act Sections 305(b) and
303(d) (US EPA 2003). Monitoring strategies developed for 305(b) regional assessments also must
inform the 303(d) listing process. In the 2001-2003 WV Regional Environmental Monitoring and
Assessment Project (R-EMAP), use of a new watershed classification technique specific for West
Vkginia should allow the state to improve its ability to discern anthropogenic causes of impakment
from non-anthropogenic sources of variability in reference condition (Cincotta 2000; Brazner et al.
2003) by producing a series of stressor-response relationships specific to different watershed classes.
-------
threshold identification
strategies: monitoring,
modeling, restoration
i
I
cost-benefit
analyses
watershed prioritization
classification
monitoring framework
class condition and cumulative
condition assessment
screenm
threshold confirmation
[translators]
Figure 1. Use of classification in sequence of monitoring, assessment of condition,
and prioritization of watersheds for further monitoring, modeling, and restoration.
In addition, use of the watershed classification scheme within a monitoring strategy will allow
estimation of the condition of classes of watersheds and prediction of ecological risk for
unmonitored systems.
Gradients of land-use can be studied in combination with hydrology moderating factors to
establish relationships of changes in runoff, and associated water quality and biological responses,
with land-use activities (Verry 1986; Detenbeck et al. 2000). Since multiple non-point stressors
potentially are affecting biological condition, it is appropriate to use watersheds as the sampling
and/or assessment unit. In order to establish an appropriate stratified random sampling regime to
diagnose causes of impairment, watersheds within the state of West Virginia must be properly
classified. The classification developed and described in this document is based on identifying the
levels of hydrology-moderating factors such as watershed storage and watershed development at
which rapid degradation occurs. Watershed storage capacity is operationally defined as the fraction
-------
of watershed area present in lake + wetland area, and can be used to predict the magnitude of floods
(Jennings et al 1993). Peak flows are generally associated with a large fraction of nutrient and clean
sediment yields from watersheds, so that watershed storage can also be used to predict attenuation
of nonpoint-source loadings. Wetlands and lakes increase retention time within the watershed,
allowing processes such as nutrient uptake, denitrification, and sedimentation in wetlands to
improve water quality. Watershed storage can also be a good predictor of baseflow water quality
(Johnston et al. 1990; Detenbeck et al. 2000; Detenbeck et al. 2003a,b).
Hydrology-based classification of watersheds
Watershed properties such as depressional storage volume and surface-water
hydrogeomorphic types can be used to classify lakes or streams according to relative risk of impact.
Our approach to predicting stream sensitivity to nonpoint source pollution is based on the nonlinear
response of hydrologic regimes and associated loadings of non-point source pollutants to watershed
properties (Richards 1990). Selected hydrologic thresholds are related to 1) variation in levels of
watershed storage (either natural or anthropogenic) and 2) land-use activities affecting runoff and
the hydrologic regime. In common usage, the term "threshold" refers to "the point at which a
physiological or psychological effect begins to be produced" or "the point at which something
starts." We define a hydrologic thresholds a breakpoint or inflection point in a nonlinear relationship
between a watershed property and hydrologic response variable such as peak flows. For example,
peak flows per unit area often increase exponentially once the storage capacity of a watershed has
been exceeded. We have identified these thresholds based on a space-for-time substitution, or
comparison across watersheds, as inadequate historical records are available for analysis of time
series of individual watersheds.
The United States Geological Survey (USGS) has defined a series of empirical nonlinear
-------
equations relating watershed properties such as watershed area, channel slope, watershed storage,
and land-use (percent forested, percent urbanization, or percent impervious surface area) to peak
flows of given recurrence intervals (Q2, Q5,... Q100; Jennings et al. 1993). Peak flows increase
exponentially as watershed storage decreases below a given threshold (Detenbeck et al. 2000).
These thresholds form the basis for a watershed classification system.
Relationship between HUCs and watersheds
The first step in developing a classification strategy is to define the population of interest.
Theoretically, an infinite number of watersheds could be defined, starting at any random point on a
stream network (Detenbeck et al. 2003a). Benefits of defining a fixed set of watershed units include:
a) the ability to tie assessment units to management plans and actions, b) ease in mapping
classification units and classes, and c) the significance of fixed management units for stakeholders
who form watershed management organizations to facilitate local environmental protection.
The USGS is coordinating with other Federal agencies and the states to develop a National
Watershed Boundary Database (NWDB), which will sequentially divide existing 8-digit Hydrologic
Units (HUCs) into 10-digit HUCs, and 10-digit HUCs into 12-digit HUCs (Legleiter 2001). Use of a
seamless national boundary database will facilitate communication among neighboring states, and
planning and management of basins that cross interstate boundaries. In addition, creation of
seamless nationwide GIS kyers will allow development of a series of elevation, hydrography, and
derived coverages that are internally consistent with one another (Franken et al. 2001).
We chose 12-digit HUCs as the base map for our classification framework. Watersheds
associated with 12-digit HUCs span the size range of wadeable streams in West Virginia (400 -
40,000 ha), correspond to the size range of most management units in the state, and fit within a
hierarchical scheme of larger management units (i.e., 10- and 8-digit HUCs). A total of 36 8-digit
-------
HLJC units drain into the state of West Virginia; these 8-digit HUCs contain a total of 242 10-digit
HUCs (USGS watershed units) and 1427 12-digit HUCs (USGS subwatershed units). Of these 12-
digit HUCs, 883 fall predominandy within the borders of West Virginia.
Not all HUCs are equivalent to watersheds. HUCs can be defined as either basin or
interbasin. Noncoastal basin HUCs are generally equivalent with watersheds at the 12-digit scale,
while interbasin HUCs divide the main channel of the next larger size class of HUCs into segments.
Coastal HUCs can contain multiple parallel watersheds draining to a large water body (Great Lake or
ocean), but these are not an issue in West Virginia. Interbasins may include a middle segment of a
larger river that encompasses many smaller side tributaries. Interbasin HUCs can be aggregated with
upstream units to define full watersheds (Figure 2). Throughout this report, derivation of watershed
attributes for 12-digit HUCs is based on the boundary and characteristics of the full aggregated
watershed regions.
Use of Pfafstetter codes facilitates the discrimination between basin and interbasin HUCs,
and identification of all upstream or downstream HUCs. Pfafstetter codes are not the same as HUC
codes, but are produced by the USGS during the process of creating elevational derivatives such as
watershed boundaries. Pfafstetter codes are hierarchical, with a 2-digit segment added at each new
level of subdivision. The last digit of the Pfafstetter code increases consecutively from mouth to
headwater sub-basins, with even numbers associated with tributaries (basin HUCs) and odd
numbers associated with HUCs along the mainstem (interbasin HUCs) (Verdin and Verdin 1999;
Figure 3).
Use of watershed classification as a sampling and assessment framework
Once a fixed set of watersheds is identified, delineated, characterized, and classified, this set
can be used to define a sample population and survey design. At least two options exist for
-------
Subwatershed Region
HUC 050701010103
Region with National
Hydrography Data
and Shaded National
Elevation Data
Legend
| | 050701010103 Region
-/v NHD stream reaches
Map Projection Alters Equal Area Conic
Figure 2. Definition of a subwatershed region for 12-digit HUC 050701010103 through
aggregation of 05 additional upstream HUCs.
-------
Figure 3. Generic example of Pfafstetter coding for an 8-digit HUC (red outline). The base of the
main channel is coded as "1" while the headwater portion of the main channel is coded as "9".
Each of the four largest tributaries (basin HUCs) is assigned an even number between 2 and 8, in
ascending sequence from the mouth towards the headwaters. Likewise, each of the segments
draining to the mainstem between the tributaries (interbasin HUCs) is assigned an odd number
between 3 and 7, in ascending sequence from the mouth to the headwaters. Coding of subdivisions
within one of these 9 HUCs would proceed in a similar fashion, with the addition of a second digit
to the code.
-------
incorporating watershed classes into a survey design framework: 1) classes can be used as strata in a
stratified random design, or 2) classes can be assigned unequal probability weights to influence the
distribution of selected watershed units across classes (Detenbeck et al. 2003a). Both options allow
a rare but significant watershed class (e.g., one with high associated probability of impact) to be
sampled at a higher frequency than would be possible if a simple random survey design were
implemented. Monitoring data can then be assessed for classes of watersheds as well as for an entire
region, and a series of appropriate management strategies devised.
METHODS
USGS watershed delineation
To define hydrologic thresholds, gaging stations with long-term flow records and derived
peak flow statistics were identified from previous studies in the state of West Virginia (Frye and
Runner 1970; Runner 1980; Figure 4). Watersheds associated with USGS gaging stations were
delineated onscreen in Arclnfo using digital raster graphic (DRG) backdrops (1:24,000) and
National Hydrography Dataset (NHD) coverages.
Delineation and coding of hydrologic units for the state ofWV
In contrast to manual delineations for USGS gaged watersheds, hydrologic units in West
Virginia were delineated through a two or three stage process as part of the development of the
Elevation Dataset with National Applications (EDNA, Franken et al. 2001;
http://edcntsl2.cr.usgs.gov/ned-h/). The EDNA process produces HUC boundaries suitable for
inclusion in the National Watershed Boundary Dataset (Figure 5). Delineation tasks were
accomplished through an interagency agreement with the USGS EROS Data Center and
collaborative efforts of US EPA Region III staff. Stage I of the EDNA process previously had been
-------
USGS Gaging Station
Watersheds
Map Projection AJbar* Equal ATM Conic
Gagng Slatnn Ports
Gagng Station Walenhedi
Level III Ecorrgions
Name
Wteslem Allegheny PbKau
Central Appaladnam
Cenlial Appalachun Ridgn and Va
Figure 4. Location of USGS gaging stations and associated watershed boundaries used for
derivation of hydrologic thresholds in West Virginia.
10
-------
WV8,10, &12-Digit
HUCs with Contributing
8-Digit HUCs
| J 8-Digit HUCs (EDNA)
8-Digit HUCs (contributing)
10-DigitHUCs
12-Digit HUCs
Processed HUCs
Data Type
USGS1:250K
EDNA Stage II
EDNA Stage III
Map Projection. Alters Equal Area Conic
Figure 5. Map of 8-digit hydrologic catologing units (HUCs), 10-digit watersheds, and 12-digit
subwatershed units produced for West Virginia. Stage III 12-digit HUCs are shown only for three
of the 13 8-digit HUCs completed as of June 2003, i.e., those three 8-digit HUCs requiring
additional processing beyond Stage II to correct flow directionality.
11
-------
accomplished through a memorandum of understanding between the USGS and US Weather
Service. Stage I consists of an automated watershed delineation process using digital elevation
models (DEMs) with GIS to produce a nationwide coverage of catchments at the scale of 1-2 mil2.
Catchments produced at this stage reflect the predicted movement of surface water over surface
topography; elements such as karst features are not included. Stage I produces several products,
including catchment boundaries, synthetic streamlines (predicted location of streams and river
courses), and elevational derivative raster coverages such as slope, aspect, flow direction, and flow
accumulation.
During Stage II, Stage I catchments are aggregated to the scale of 10- and 12-digit HUCs,
based on a set of national guidelines (FGDC 2002; Table 1). At this point, GIS coverages are
evaluated for internal consistency and potential errors are flagged. Indicators of potential errors
either in digital elevation models or in NHD include divergence of synthetic streamlines from
mapped NHD streams and inconsistencies between HUC boundaries and other sources of mapped
watershed boundaries. During Stage III, a series of Arc View tools are used to correct small regions
of the original digital elevation models near the flagged errors to represent the actual flow of surface
water. For example, a highway overpass could be detected as a barrier to flow in an original DEM,
when in fact, water can freely flow underneath the underpass. Creating a small "notch" in the
original DEM at that neighborhood allows the stream course to be correctly predicted. After the
hydrologically-corrected DEMs have been created, boundaries for 10- and 12-digit HUCs are
regenerated, as are raster coverages for other hydrologic derivatives.
12
-------
Table 1. Guidelines for hydrologic unit subdivision for National Watershed Boundary
Database, according to interagency protocol (FGDC 2002).
Unit digits
8-digit
10-digit
12-digit
Name Size range (acres)
cataloging unit
watershed 40j000 tQ 250>m
subwatershed -, Q QQQ ^ QQQ
Subdivisions/unit
5-15
5-15
For the West Virginia R-EMAP project, we used Stage II HUCs to develop the watershed
classification framework. Currently, Stage III HUCs are only partially completed for the state, and
were only used where Stage II HUCs required substantive correction, e.g., to correct direction of
flow (Figure 5).
The assignment of Pfafstetter codes during Stages I and II of the process allows ready
identification of upstream and downstream 10- and 12-digit HUCs within 8-digit HUCs, and thus
aggregation of interbasin HUCs to define watersheds using REGION processes in Arclnfo.
Definition of Arclnfo regions is desirable considering the nested nature of watersheds. Regions are
model areal geographic features described by one or more polygons. Regions are efficient ways to
model nested or overlapping data because of the concept of shared geometry. For example, one
polygon feature can belong to multiple regions without the topology being duplicated. Two files
store the region-arc relationship and the region-polygon relationship "behind the scenes." Regions
are stored as subclasses within a polygon coverage and each region subclass has its own separate
attribute table.
The bulk of the work was completed using individual Arc Macro Language (AML) scripts
developed for each component of the characterization process. The scripts were used to create and
combine region subclasses into an integrated coverage for further geoprocessing using the
13
-------
REGIONQUERY command in Arclnfo (ESRI, Redknds CA). The REGIONQUERY command
mimics traditional oveday commands on region subclasses in an integrated coverage. So, for
example, if a single coverage had a 12-digit watershed subclass and a wetland subclass, the
REGIONQUERY command could be used to create a new subclass computed from the geometric
intersection of the two subclasses (watersheds and wetlands). The region tables keep track of which
wetland polygons belong to each watershed polygon while only physically representing the polygons
once. One table could then be created that summarized wetland type by area for each watershed
region.
Due to the size of input datasets and software/hardware limitations, 8-digit HUCs
containing 12-digit watersheds were processed individually. This presented an issue when a 12-digit
watershed required additional upstream 8-digit HUC(s) to be considered a true watershed. A
method was needed to identify diese "interbasin" watersheds and combine the output for their
required counterparts. Each interbasin watershed was given a special identifier code in the
watershed region attribute table. A table was made listing all possible interbasin codes and the 12-
digit watersheds needed to complete the full watershed. This table was imported into a Microsoft
Access database. The individual output tables for 12-digit watersheds within an 8-digit HUC were
combined into one table for each characterization process and imported into the same database. A
series of queries were performed to combine the individual output values to determine the total
output values for interbasin watersheds for each characterization process.
Watershed characterisation
Watershed regions associated with all 12-digit HUCs in the state of West Virginia and long-
term gaging stations were characterized for attributes expected to be related to generation of peak
flows (Fne and Runner 1970; Runner 1980): watershed area, main channel length, main channel
14
-------
slope, watershed storage, mean January minimum temperature, snowpack, and percent forest cover.
In addition, watersheds were characterized for other major land-use and land-cover characteristics
related to generation of runoff and nonpoint source pollution: percent mined land, percent
impervious surface area, and percent agriculture. GIS databases used to calculate watershed
characteristics are described in Table 2.
Table 2. Geographic information system databases used to characterize West Virginia watersheds. EDNA = Elevation Dataset with
National Applications, NHD = National Hydrography Database, NWI = National Wetlands Inventory, OWI = Ohio Wetlands Inventory,
NYWI = New York Wetlands Inventory, PRISM = Parameter-elevation Regressions on Independent Slopes Model.
Database
EDNA
NHD
NLCD-
mining
modified
NWI
OWI
PRISM
Layers and attributes
10-digit watershed boundaries
12-digit watershed boundaries
Pfafstetter codes
elevation grids
stream reaches
stream level
land-use, with updates for surface
mining
palustrine wetland polygons
lacustrine polygons
wetland and open water types by
grid cell
monthly snowfall
annual precipitation
minimum January temperature
Derived variables
sample units
watershed boundaries
watershed areas
upstream connections
main channel slope
main channel slope
upstream connections
percent land-cover
percent wetland + lake area
percent wetland + lake area
snowfall
mean annual precipitation
mean minimum January
temperature
Last update Scale Source
2003 1:24,000 USGS EROS Data
,.. ... Center
(30 m gnd)
1999 1:100,000 USGS
2001 1:24,000 NLCD - EPA
(30 m grid) surface mining grid -
Tennessee Valley
Authority
Thru:
US EPA Region 3,
Wheeling, WV
1981 to 1:24,000 US FWS, WV DEP
present composite coverage
based on 1:24,000 Ohio Department of
1985-1987 Natural Resources
TM scenes
1998, based 1:250,000 Climate Data Source,
on 1961-1990 Corvallis, OR
climate data,
15
-------
The PRISM (Parameter-elevation Regressions on Independent Slopes Model; Climate
Source, Corvallis, OR) GIS coverage was used to generate climatic summaries across watersheds.
PRISM is a continuous grid (raster coverage) of climatic variables generated from weather station
data and regional nonlinear regressions that take into account regional precipitation averages as well
as orographic effects associated with altitude, and proximity/influence of large surface water bodies
(Daly et all 994).
We used National Wetlands Inventory (NWI) coverages, supplemented by the Ohio
Wetlands Inventory coverages to define watershed storage. Total areal coverage of palustrine and
lacustrine polygons was added and divided by watershed area. For the Ohio Wetlands Inventory
coverage, polygons coded as 34 to 38 (wetland and open water cksses) were included. For the New
York Wetlands Inventory, cover classes 1-4 and 9 were included (all classified or unclassified
wetland polygons).
Fraction impervious surface area was derived as a function of knd-use intensity as:
fimperv = (0.55*flwintrs)+(0.90*£hintres)+fcomind,
where fimperv = fraction impervious area in watershed,
flwintrs = fraction low intensity residential area in watershed,
fhintres = fraction high intensity residential area in watershed, and
fcomind = fraction commercial, industrial, and transportation knd-use area in
watershed.
based on median of estimated constructed materials associated with different land-use cksses in
NLCD (http://www.epa.gov/mrlc/dennitions.html). These weightings are higher than those
16
-------
reported for northern Virginia (NVPDC 1980), because the latter included only effective impervious
surface area (unconnected rooftops were not included), but are consistent with classification
guidelines for NLCD.
Main channel slope was defined first by selecting main channel reaches from the NHD
coverage based on the minimum value for the reach LEVEL attribute within a watershed. Main
channel reaches were buffered by 30 meters, and this buffer was used to clip a subset of elevation
grids from the DEM coverages. Frequency analyses were used to estimate the 10th and 85th
percentile of elevation values from the stream elevation buffer files; combined with main channel
length, these could be used to calculate main channel slope (Figure 6).
Creation of sample frame and sample design for 12-digit HUCs
Although we developed the classification framework for the entire state of West Virginia, in
practice, we restricted membership in the sample frame for the 2001-03 WV R-EMAP project.
Sample frames were defined based on drainage basin, ecoregion, and 12-digit HUC watershed region
size (Table 3). The Potomac River basin was excluded from study because of species differences
related to biogeography, and because the fish IBIs developed for adjacent regions in Maryland
should be adequate for the state of WV to apply in that basin (Figure 7). In practice, we also
excluded watersheds with ill-defined drainage networks related to karst topography (e.g., no single
outlet, underground streams). Selection of 12-digit HUCs from within the sample frame each year
was based on a random-stratified design. Strata were based on watershed classes defined by
hydrologic thresholds and knd-use intensity. After defining 12-digit HUCs as single independent
watersheds (basin HUCs) or groups of hydrologically adjacent (dependent) interbasin units,
17
-------
HUC 060701010103
Region with NHD and
Shaded NED
/
c
"Ajf-WI. 1
0 Ot5 01
N
A
MUC 050701010103
Region wtm NMD nd
ShxtedNED
0.8
0.6
0.4
0.2
Derivation of main channel slope
85%
10%
472 - 436 = 36 m
/ change in elevation
36 m elev/ 29.928 km
main channel length
0
420 430 440 450 460 470 480 490 500 510
Elevation (meters above sea level)
Figure 6. Automated derivation of main channel slope. The main channel is identified based on
finding stream reaches with the minimum "LEVEL" attribute in NHD (top left), then is buffered
and intersected with a digital elevation model (top right). A frequency analysis of output elevation
grid cells yields estimates of the 10th and 85lh percentiles. Combined with main channel length,
these can be used to calculate main channel slope.
-------
Major Drainage Basins
Map Projection: Alters Equal Area Conic
| | 2-Digit HUCs
| | 4-Digit HUCs
HUC 02
02
05
HUC 04
0207
0208
0501
0502
0503
0504
0505
0507
0509
REGION NAME
Mid Atlantic Region
Ohio Region
SUBREGION NAME
Potomac
Lower Chesapeake
Allegheny
Monongahela
Upper Ohio
Muskingum
Kanawha
Big Sandy-Guyandotte
Middle Ohio
Figure 7. Major drainage basins associated with the state of West Virginia and surrounding states,
as defined by 2- and 4-digit HUCs.
19
-------
unequal probability-weighting was used to limit the chance of selecting subwatersheds directly
upstream or downstream of one another (Figure 8).
Table 3. Definition of sample frame for R-EMAP wadeable stream sampling in
West Virginia, 2000-2001.
Year Ecoregion(s) Drainage basin Watershed size (ha)
2000 Central Appalachian Pkteau inclusive 400-40000
2000 Central Ridge and Valley all excluding Potomac River 400-40000
Ecoregion basin
2001 Western Allegheny Pkteau inclusive 400 - 40000
Response threshold derivation
We carried out analyses and applied sampling designs separately by ecoregion for studies in
2000 and 2001 (Table 3, Figure 9). To define hydrologic thresholds related to natural watershed
attributes, we identified gaging stations with long-term flow records and derived peak flow statistics
from previous studies in the state of West Virginia (Frye and Runner 1970; Runner 1980; Figure 4).
We delineated watersheds for each gaging station and characterized these watersheds based on
attributes recorded in Frye and Runner (1970) and Runner (1980). We updated watershed storage
estimates from Frye and Runner's values using GIS analysis of National Wetlands Inventory
coverages and calculation of fraction watershed area covered by palustrine and kcustrine systems.
In earlier analyses conducted before GIS was widely avaikble, Frye and Runner calcukted watershed
storage by overkying a point grid on USGS topographic quadrangles and counting intersections
with wetlands, lakes, and ponds. With a rare aquatic resource such as wetknds occurring in West
Virginia or other regions with well-developed drainage systems, the accuracy of such methods is
limited by the temporal accuracy of maps, the density of grid points used, and number of grid
20
-------
Kilometers
25 50
100
A
Random-Stratified
Selection Process
Weighting Procedure
Map Projection Alters Equal Area Conic
Legend
| | Independent 12-digitHUCs
| | Dependent 12-digit HUCs
i Central Appalachians
Inclusion Probabilities
| 0.013
0.018
0026
0.053
Figure 8. Example of unequal weighting of interbasin HUCs for one watershed class in the Central
Appalachian Plateau ecoregion of West Virginia.
21
-------
Map Pro)*cton Albert Equal Aiea Cone
US EPA
Level III Ecoregions
Level III Ecoregions
Name
Wo3»m Mogfwny Fbt.au
C«n0.il AppatichKuij
Cones! AppaLxrhoo Rrioo, and Vifay^
Figure 9. Map of ecoregions overlapping with West Virginia watersheds.
22
-------
points counted per unit area. For West Virginia, percent coverage had only been recorded to a
single significant unit (as 0 or 1).
We applied two different methods to identify hydrologic thresholds, i.e., regions in plots of
area-normalized peak discharge (e.g., Q2/watershed area) showing nonlinear discrete shifts along a
gradient of watershed attributes. We obtained values for 2-year peak flows from Frye and Runner
(1970) and Runner (1980). We chose a recurrence interval of two years as an endpoint because this
is the frequency of flooding associated with development of channel morphology and sediment
delivery (Rosgen 1996). Our first method, applied in 2000, involved screening potential parameters
of interest identified by Frye and Runner (1970; see Table 4) using Linear regression analysis of log-
transformed Q2/area, with observations weighted by period of record to account for differences in
level of uncertainty (Tasker and Stedinger 1989). We chose the best number and combination of
predictive variables using Mallow's Cp statistic (SAS 1990). We identified thresholds through visual
graphical analysis, by plotting (nontransformed) Q2/area as a function of single variables or
combinations of variables and observing where nonlinearities in relationships were observed.
In 2001, we applied a second technique, based on classification and regression tree (CART)
analysis of potential variables in SYSTAT (Wilkinson 1999). Nonlinear regression analysis followed
by visual analysis of graphic bi-plots was less successful in 2001 because of collinearities in
independent variables. CART is a nonparametric technique that sequentially bisects observations
into subpopulations based on a single response variable, and identifies dependent variables
associated with breaks in the distribution. In this way, CART is able to identify nonlinearities in
relationships, as well as interactions among predictor variables, based on a minimum set of
assumptions (Wilkinson 1999).
23
-------
Table 4. Potential watershed characteristics related to peak flows, Frye and Runner (1970).
Code
A
S
St
F
L
E
S,
Units
mil2
feet -mil"1
Si
T,
Definition
watershed area
channel slope
storage (area in lakes and ponds +1)
forest area
main channel length
mean basin elevation
snowfall index
average annual precipitation - 20
precipitation intensity/24 hours, 2-year inches
recurrence interval
soil infiltration index
average minimum January temperature deg F
miles
feet above sea level
inches
inches
We identified potential thresholds of response related to land-use change from the literature
because insufficient data were available from gaged sites with significant periods of record. We
based potential land-use thresholds of response on previous analyses of data from the Maryland
Biological Stream Survey (MBSS, Boward et al. 1999). For the MBSS data set, species loss in fish
communities had been found at levels of impervious surface area above 2% of watershed area.
Nonlinear shifts in water quality were associated with levels of agricultural land-use above 25%.
Other studies have shown a change in peak snowmelt associated with loss of mature forest cover
greater than 40-60% (Verry 1986). This approach is similar to trying to bracket expected response
diresholds in bioassays for suspected toxicants.
Less information is available from the literature to detect thresholds of impact rekted to
mining activity in the Appalachian region. Scott (1984) identified percent disturbed area as the best
24
-------
predictor of changes in stream hydrology in a subset of mined watersheds of West Virginia,
presumably associated with changes in water quality (Scott 1984). However, Scott did not have long
hydrologic time series for these watersheds, but instead based his conclusions on modeled results;
his analysis showed an apparent threshold at 30% disturbed area. We chose to use a conservative
estimate of 10% for the mining threshold, similar to that shown as a threshold for percent
impervious surface area in most literature reviews (Schueler 1994), with the knowledge that the
magnitude of mining thresholds would have to be refined in the future.
We defined land-use coverage for watersheds associated with 12-digit HUCs by intersecting
watershed boundaries with a mining-modified NLCD coverage (Table 2, Figure 10). We defined
percent watershed storage by intersecting watershed boundaries with wetland inventory coverages
(Figure 11).
25
-------
Mining-Modified NLCD
Land-Use/Land-Cover
Mining-Modified NLCD
Grouped Land-Use Codes
Mining
Agriculture/Orchards
Forest
Grassland
Urban/Residential
Open Water/Wetlands
Barren
Map Projection: Albers Equal Area Conic
Figure 10. Map of land-use covering West Virginia watersheds, based on National Landcover Characterization Database, updated for mining category.
27
-------
0 25 50
Map Projection: Abets Equal Area Conic
Lakes and Wetlands
in National Wetland
Inventory (NWI)
Legend
12-digit HUCs
NWI SYSTEM
Lacustrine
Palustrine
Figure 11. Map of palustrine and lacustrine classes from National Wetlands Inventory for the state
of West Virginia.
29
-------
CLASSIFICATION RESULTS
Hydrology thresholds
In year 2000, the nonlinear equation we derived to describe 2-year peak flows (Q2) for the
Central Appalachian Plateau and Central Ridge and Valley ecoregions, was:
Q2 = Aa Pp T Snsn St5' (p < 0.05)
where Q2 2-year peak flow
A watershed area
P = average annual precipitation
T = average minimum January temperature
Sn = snowfall
St = watershed storage.
Percent forest was not included in the equation for 2-year peak flow, probably because of
a lack of significant variation in percent forest cover in these two ecoregions. When gaging station
data were combined for the full state, percent forest cover was included as a significant predictor
variable in peak flow equations for 5- and 10-year recurrence intervals, but not for 2-year recurrence
intervals. Visual graphical analysis showed that percent forest alone showed no threshold effect on
peak flows normalized to watershed area, but the product of percent forest cover, snowfall, and
minimum January temperature did show a noisy threshold with a lot of scatter. The sharpest
threshold for Q2/A identified based on visual graphical analysis was for percent watershed storage
(Figure 12), with a threshold identified at ~0.5% storage (fraction storage = 0.005).
We identified variables and associated threshold levels for Q2/A for the Western Allegheny
Plateau in 2001 through CART analysis. At each step, the CART procedure sequentially divided
observations into two categories based on the magnitude or category of the predictor variable (in
30
-------
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
Fraction storage (lake + wetland/wshd)
Figure 12. Predicted 2-year flood normalized to watershed area (cfs/mi!2) as a function of
watershed storage (lake + wetland area/watershed area) for USGS gaging station watersheds in the
Central Appalachian Plateau and Central Ridge and Valley ecoregions of West Virginia.
this case fraction storage or main channel length) that best separated the response variable,
Q2/watershed area, into low and high magnitude subpopulations. Each box in the output is
basically a frequency plot, with each observation represented by a colored dot, and each color
representing the final subclass assignment. Separation of the dataset based on combination of two
variables accomplished a 67% reduction in total variance (Figure 13). Flashy hydrologic regimes
(significantly larger Q2/A) for the Western Allegheny Plateau ecoregion occurred in low storage
watersheds (< 0.03% watershed storage) or relatively short drainage basins (main channel length <
13.5 km).
31
-------
Q2/watershed area
I
%storage<0.031
Main channel length (km)<13.5
Figure 13. Results of Classification and Regression Tree (CART) analysis of 2-year flood
normalized to watershed area (cfs/mil2) as a function of watershed attributes for USGS gaging
station watersheds in Western Allegheny Plateau ecoregion of West Virginia (percent reduction in
error = 67 %). At each step, the CART procedure sequentially divided observations into two
categories based on the magnitude or category of a predictor variable (in this case fraction storage
or main channel length) that best separated the response variable, Q2/watershed area, into low and
high magnitude subpopulations. Each box represents a frequency plot, with each observation
represented by a colored dot, and each color representing the final subclass assignment.
-------
Distribution of land-use/ land-cover variables across 12-digit subwatersheds in West Virginia
Overall, the level of watershed storage in West Virginia is very low. For all watersheds
associated with 12-digit HUCs in the 8-digit HUCs characterized (those within or overlapping with
WV state borders), the top 20th percentile of HUCs had a range of 0.8 to 18.6% watershed storage
(Figure 14). Most 12-digit HUCs in the top 20th percentile were actually located near the border or
even outside of WV state boundaries, in the headwaters of the Upper and Middle Ohio River basins.
Nonetheless storage thresholds associated with a marked change in hydrologic regime were low
enough so that both high and low storage classes could be identified within each set of ecoregions
(Figure 14).
The range of percent impervious surface area in watersheds associated with the 12-digit
HUCs characterized is also relatively low (0-18.4 %), with the top 20th percentile covering a range of
only 1.2 - 18.4 %. The spatial distribution of the top 20th percentile of percent impervious surface
area is similar to that of the top 20th percentile of watershed storage, paralleling development along
road or river corridors, and probably the creation of reservoirs (Figure 15). As for urban
development, the highest concentration of agriculture within West Virginia drainage basins (top 20th
percentile = 48-83 %) occurs outside of the state boundaries in the Upper Ohio or Potomac
drainages (Figure 16). In contrast, the highest concentration of mining activity (top 20th percentile =
1.4-16.4% watershed area) occurs within the Central Appalachian Plateau, in the southwestern
portion of the state (Figure 17). Forest cover across the state is relatively high and constant; most
instate HUCs are in the top 40th percentile of HUCs characterized (79 - 100 % forest cover; Figure
18.)
33
-------
Fraction
Watershed Storage
By 12-DigitHUC
12-Digrt HUCs
FractionStorage
0.0000-0.0010
00011 -00023
00024 - 0.0042
HH 0 0043 - 0 0082
00083-0.1860
klip Protection Albett tqujl Aiea Come
Figure 14. Map of 12-digit HUCs within 8-digit HUCs overlapping with West Virginia, coded by
fraction watershed storage ([lake + wetland area/watershed area] x 100). Classes mapped are not
equal interval classes, but represent population quintiles.
-------
Fraction
Impervious Surface
By12-DigitHUC
12-Digit HUCs
Fraction Impervious Surface
0.0000 - 0.0008
0.0009 - 0.0028
0.0029 - 0.0053
0.0054-0.0121
0.0122-0.1837
Map Projection. Abers Equal Area Conic
Figure 15. Map of 12-digit HUCs within 8-digit HUCs overlapping with West Virginia, coded by
estimated fraction impervious surface area in associated watersheds. Classes mapped are not equal
interval classes, but represent population quintiles.
35
-------
Fraction Mining
By12-DigitHUC
12-Digit HUCs
Fraction Mining
0.0000
00001-0.0012
00013-0.0064
IB} 0.0065-0.0138
Ij^B 0.0139-0.1644
Map Pto)*ctKm *Jb«f* Equal Aiea Cone
Figure 16. Map of 12-digit HUCs within 8-digit HUCs overlapping with West Virginia, coded by
estimated fraction surface mining area in associated watersheds. Classes mapped are not equal
interval classes, but represent population quintiles.
36
-------
Fraction Agriculture
By 12-DigitHUC
12-Digit HUCs
Fraction Agriculture
0.0002 - 0.0822
0.0823-0.1910
0.1911 -0.3140
. 0.3141-0.4814
^^1 0.4815-0.8315
Map Projection: AJbers Equal Area Conic
Figure 17. Map of 12-digit HUCs within those 8-digit HUCs overlapping with West Virginia,
coded by fraction agricultural area in associated watersheds. Classes mapped are not equal interval
classes but represent population quintiles.
37
-------
Fraction Forest
By 12-DigitHUC
12-Digrt HUCs
Fraction Forest
0.1400 - 0.4788
0 4789 - 0.6605
0 6606 - 0 7870
IB 07871 -08937
^H 0 8938 - 0.9989
M«p Pioftctlon Albef* Equal Ai«« Come
Figure 18. Map of 12-digit HUCs within those 8-digit HUCs overlapping with West Virginia,
coded by fraction forested area in associated watersheds. Classes mapped are not equal interval
classes but represent population quintiles.
38
-------
Because of the uneven distribution of watershed storage, watershed morphology, and land-
use across the state of West Virginia, the distribution of 12-digit HUCs is not even across watershed
classes; in fact some combinations of watershed classes are not even represented (Figures 19-20).
For example, in the Western Allegheny Plateau ecoregion, only one 12-digit HUC watershed
occurred in the high urbanization/low storage class, and no 12-digit HUC watersheds occurred with
combinations of moderate or high agriculture and low storage. Thus, not all potential watershed
classes can be compared within each ecoregion (Figures 21-22).
39
-------
Mdp Pto|*t.t)on Alb*ft Equal Ai»« Cone
Aggregated Land-Use
by Ecoregion
[ | HUC Bomdwy
Agriculture/Orchardi
BH Forest
Grasslands
Urban/fteskfartfal
H Open Water/Wetlands
Barren
Level III Ecoregions
Name
Western Allegheny Plateau
Central Appalachians
Central App. Ridges & Valeys
Figure 19. Map of Omemik Level III ecoregions within state of West Virginia, and associated
land-use summaries at ecoregion level for region within WV watersheds.
41)
-------
Western
Allegheny
Plateau
Central Appalachian
Ridges & Valleys
12-DigitHUC
Watershed Classes
Watershed Classes
Low Impact/Low Storage
j Low Impact/High Storage
High Urbanization/Low Storage
High Urbanization/High Storage
Moderate Agriculture/Low Storage
Moderate Agriculture/High Storage
High Agriculture/Low Storage
High Agriculture/High Storage
Ecoregions
Ecoregions
Level
Level
Map Projection: Albers Equal Area Conic
Figure 20. Map of 12-digit subwatershed units within West Virginia's 8-digit HUCs, classified by land-use and watershed storage classes. Storage classes are based on empirical hydrologic thresholds for
watershed storage (CA, CARV ecoregions) or combinations of storage and main channel length (WAP ecoregion).
41
-------
Map Projection: Abere Equal Area Conic
Random-Stratified
Selection of HUCs
for Sampling
Legend
| | HUCs selected for sampling
Thresholds
Reference, low storage
HI Reference, high storage
High mining, low storage
High mining, high storage
BH High agriculture, low storage
High agriculture, high storage
High impervious, low storage
i | High impervious, high storage
Level III Ecoregions
Name
| | Central Appalachians
Figure 21. Watershed classes associated with 12-digit HUCs in sample population for Central
Appalachian Plateau and Central Ridge and Valley ecoregions in West Virginia. Sample population
included 12-digit subwatersheds with associated watersheds in size range of wadeable streams and
outside of Potomac River basin.
43
-------
Watershed Classes in
Western Allegheny Plateau
Map Projection. A!b« r_ Equal Area Conic
Legend
Sample Ponts
Watershed Classes
Low Impact/Low Storage
Low Impact/High Storage
| High Urbanization/Low Storage
High Urbanization/High Storage
Moderate Agriculture/Low Storage
Moderate Agncuttvjre/High Storage
| High Agriculture/Low Storage
High Agriculture/High Storage
Level III Ecoreglons
Name
1 Western Allegheny Plateau
Figure 22. Watershed classes associated with 12-digit HUCs in sample population for Western
Allegheny Plateau ecoregion in West Virginia. Sample population included 12-digit HUCs with
associated watersheds in size range of wadeable streams.
44
-------
FUTURE UPDATES AND CLASSIFICATION REFINEMENTS
Potential future improvements to the watershed classification system for West Virginia
include the following:
Examination and documentation of differences in reference condition among low and
high storage reference ("low impact") classes based on WV R-EMAP data, EMAP data
from the MidAtlantic Integrated Assessment (MAIA), and macroinvertebrate and water
quality data collected as part of the WV Department of Environmental Protection (DEP)
rotating basin monitoring program
Examination and documentation of differences in condition between reference and
"impacted" watershed classes within storage categories based on WV R-EMAP data,
EMAP MAIA data, and macroinvertebrate and water quality data collected as part of the
WV DEP rotating basin monitoring program
Examination and adjustment of thresholds of impact associated with watershed storage
and knd-use intensity based on the analyses described above.
Potential refinement of watershed impact classes based on empirical analysis of existing
data with the addition of other watershed attributes such as development within stream
and river corridor buffers, road density, and density of stream or river road crossings.
Land-use impacts in a mountainous region might be better reflected by estimation of valley
development, as compared to land-use intensity on a whole-watershed basis.
Improved estimation of percent impervious surface area in WV watersheds. The method
used for estimation in this project was indirect. However, the Chesapeake Bay program is
currently mapping impervious surface area directly based on satellite imagery from Landsat
7, IKONOS, and MODIS platforms (http://chesapeake.towson.edu/data/), and
45
-------
additional studies are underway to calibrate percent impervious surface area estimation
equations based on actual measurements from aerial photographs (David Jennings, US
EPA, Reston, VA, personal communication).
Improved estimation of percent mined area (surficial disturbance). Current estimates were
based on interpretation of TM + SPOT satellite imagery by Tennessee Valley Authority
under contract to Canaan Valley Institute for permitted mining areas (Hope Childers, US
EPA Region III). The state of West Vkginia is currently constructing coverages of mining
activity that might eventually allow better definition of area! extent and different sources of
mining impacts (e.g., valley fill vs mountain top mining,
http://www.nrac.wvu.edu/hollowfill/, http://gis.dep.state.wv.us/data/omr.html);
however, current coverages are based on permit boundaries which do not always reflect
mining activity.
Slight adjustments in boundaries and characterizations from Stage II to Stage III 10- and
12-digit HUCs as Stage III EDNA process is completed for the state of West Vkginia.
There are several ways in which derivation of thresholds could be improved, depending on
the objectives of a monitoring project. CART analysis objectively maximizes the separation of two
subpopulations of a dependent variable (e.g., Q2/A) as a function of the best predictive variable,
e.g., fraction watershed storage. Thresholds derived from graphical analysis could be verified
through application of a statistical procedure such as CART (Wilkinson 1999). Reanalysis of USGS
data for the Central Appalachian ecoregions using CART yielded an apparent threshold of fraction
storage = 0.001, or 0.1%, very close to the threshold of 0.5% derived through visual examination).
46
-------
Alternatively, thresholds could be chosen by optimizing the chance of correctly identifying
watersheds with high normalized peak flows, i.e., those with greater potential for bank erosion and
transport of nonpoint source pollutants, while minimizing the chance of misidentifying watersheds
expected to have low normalized peak flows. If watershed storage is used as the indicator of
normalized peak flows, then the probability of the two types of misclassification error will depend
on the strength of the relationship between normalized peak flows and watershed storage, the error
associated with the relationship between the two variables, and the underlying distribution of
watersheds across the range of watershed storage. The absolute value of the exponent term in the
equation relating storage and peak flows is a measure of the "strength" of the relationship. An
exponent of zero would indicate that there is no relationship between storage and normalized peak
flows, while a highly negative value indicates a rapid rate of increase in peak flows as storage is
reduced (Figure 23).
Several examples are shown below to illustrate the combined effect of the magnitude of the
exponent (-2 versus -0.01), the magnitude of a desired breakpoint in Q/A relative to its underlying
distribution (median versus 95th percentile), and the effect of differing levels of error of prediction
(standard error of estimate = 0 to 100%; Figures 24a-h). In these simulations, an even distribution
of watersheds across storage cksses was used. In all cases there is a tradeoff between maximizing
the proportion of correct predictions for high normalized peak flows (Figures 24a,c,e,g) and
minimizing the proportion of incorrect predictions for the low normalized peak flow class (Figures
24b,d,f,h). As we reduce the chosen threshold for percent watershed storage, our ability to predict
watersheds with high normalized peak flows improves, but the misclassification error rate for the
low peak flow ckss also increases (e.g., compare Figure 24a with Figure 24b). Our chances for
47
-------
Effect of exponent term for storage
STA-5
STA-2
STA-1
STA-0.5
STA-0.2
STA-0.1
STA-0.01
STA0
20 40 60 80 100 120
% watershed storage
Figure 23. Effect of magnitude of exponent term for storage on shape of
relationship between watershed storage and normalized peak flow (Q/area =
ST).
making correct high flow predictions and minimizing errors in predicting low flow watersheds arc
generally better the stronger the relationship is between storage and peak flows (e.g., compare Figure
24a with Figure 24c and Figure 24b with Figure 24d). The greater the error in the relationship
between storage and normalized peak flows (higher SEE), the worse our predictions will be.
(Compare different color lines within Figure 24a or 24b.) Flowever, prediction error appears to
have little influence if we wish to distinguish only those watersheds with the most extreme peak
flows (> 95lh percentile of Q/A; Figures 24e and f). In this example, our ability to avoid making
misclassificanon errors for low peak-flow watersheds is relatively small if our breakpoint for Q/A is
high (Figures 24f and h). If at the same time, the strength of the relationship is weak (e.g., exponent
= -O.U1), however, we also have little chance of correctly identifying the high peak-flow watersheds
(l-igurc 24g).
48
-------
a)
c)
e)
g)
Fraction LO storage dass with Q/A > median
. s s e s - s
Eft
I c<
1 in
4 «
| 1
i»
1"
| 0.4
S 0.2
I o
S.
Effect of standard error of estimate on
correct prediction of Q/A > threshold
in LO storage class, exponent = -2
K^V^
- SEE = 0%
- SEE = 1%
- SEE=10%
SEE=50%
- SEE=7S%
- SEE=100%
> 20 40 SO 80 100 120
Chosen threshold, % watershed storage
ect of standard error of estimate on
>rrect prediction of Q/A > threshold
LO storage class, exponent = -0.01
)hf^~~^ ^
SEE=60%
- SEE=7S%
- SEE=100%
0 20 40 60 80 100 120
Chosen threshold, % watershed storage
| Eff
£ CC
< in
a «
1 i
1«
r
§04
I"
o
fo
| Eff
1 «
» in
i«
1 '
§0.8
D
|«
£0.4
2 0.2
1
ect of standard error of estimate on
irrect prediction of Q/A > threshold
LO storage classes, exponent = -2
V
- SEE = 0%
- SEE = 1%
- SEE=10%
SEE=SO%
- SEE=75%
- SEE=100%
9 20 40 60 80 100 120
Chosen mreshold.% watershed storage
ect of standard error of estimate on
rrect prediction of Q/A > threshold
LO storage class, exponent = -0.01
SEE=SO%
- SEE=76%
- SEE=100%
1 °
£ 0 20 40 60 SO 100 120
Chosen threshold, % watershed storage
b)
d)
f)
Effect of standard error of estimate on
c incorrect inclusion of Q/A > threshold
I in HI storage class, exponent = -2
§ 0.7
f 0.6
S 0.6
I**
|0.3
V 0.2
V
- SEE = 0%
- SEE = 1%
- SEE=10%
SEE=60%
- SEE=76%
- SEE=100%
£ 0 20 40 60 60 100 120
Chosen threshold, % watershed storage
Eff
c in<
TO .
1 ln
f 0.8
5 °-7
|0.6
S 0.6
CD
« °A
i 0.2
ts o
C
ect of standard error of estimate on
correct inclusion of Q/A > threshold
HI storage class, exponent = -0.01
__^A
SEE=60%
- SEE=75%
- SEE=100%
9 20 40 60 80 100 120
Chosen threshold, % watershed storage
S Eff
8 in(
1 0.07
£ 0.06
1 0.05
§ 0.04
| 0.03
£ 0.02
* 0.01
ect of standard error of estimate on
correct inclusion of Q/A > threshold
n HI storage class, exponent = -2
I
- SEE = 0%
- SEE = 1%
- SEE=10%
SEE=SO%
- SEE=7S%
- SEE=100%
2 0 20 40 60 80 100 120
Chosen threshold, % watershed storage
. Efl
g in
? in
< 0.07
1 0.06
S 0.04
f 0.03
a 0.02
S 0.01
I °
ect of standard error of estimate on
correct inclusion of Q/A > threshold
HI storage class, exponent = -0.01
/V\A
SEE=50%
- SEE=76%
- SEE=100%
0 20 40 60 80 100 120
Chosen threshold, % watershed storage
Figure 24. Effect of peak flow breakpoint (Q/A > median [a-d] versus Q/A >95th percentile [e-h]),
magnitude of exponent term in relationship, Q/area = STa (a = -2 [a,b,e,fj versus a = -0.01[c,d,g,h])
and prediction error on correct classification rate for high peak flow class [a,c,e,g] or
misclassification rate for low peak flow class [b,d,f,h].
49
-------
CONCLUSIONS
In some regions such as West Vkginia, the watershed classification framework for
monitoring will be limited initially by available regional data and information on relationships
between land-use and hydrologic or biological response. However, the process is iterative and
classification schemes can be validated and improved upon based on initial monitoring results. In
previous approaches to watershed classification, we have developed thresholds based on land-use
attributes calculated as a fraction of watershed area. Land-use variables described as a fraction of
watershed area might be less useful in montane regions where development is concentrated in
valleys; in these cases it might be more appropriate to express land-use as a fraction of stream buffer
zone area.
Watershed storage was one to two orders of magnitude lower in West Vkginia than in Great
Lake regions for which we have derived hydrologic thresholds for storage. However, the hydrologic
threshold for watershed storage was also an order of magnitude lower for WV as compared to the
upper Midwest, suggesting that storage is still a critical component in determining hydrologic
regime. Both removal and addition of watershed storage elements is affected by human activities.
Thus, it is possible that not all combinations of land-use and watershed storage classes will occur
within a region, as was found for the Western Allegheny Plateau ecoregion. If examination of all
potential interactions between hydrologic and land-use classes is necessary or desired, we will need
to include data from adjacent states with a wider gradient of watershed conditions.
The use of nonlinear regression analysis followed by visual analysis of bi-plots to identify
variables of interest for hydrologic thresholds is less objective and will be of limited use in cases of
collinearity of independent variables, or where classes have alternative definitions, e.g., Condition B
< X OR Condition C < Y, as was observed for the Western Allegheny Plateau ecoregion. In these
50
-------
cases, CART analysis should be a more powerful approach to define thresholds. It is possible,
however, that graphical analysis will be more useful when interactive terms (Condition B < X and
Condition C < Y) are important in denning thresholds. In the current analysis, visual graphical
analysis was used because of the form of the equation relating peak flows and watershed variables;
no inflection point can be defined in an exponential curve because the second derivative is constant.
Graphical analysis can be followed by CART analysis to confirm the placement of the threshold in
an unbiased fashion.
ACKNOWLEDGMENTS
We gratefully acknowledge contributions by Federal and contract staff of the USGS EROS
Data Center (Sue Greenlee, Kris Verdin, Jay Kost, Dean Tyler) under interagency agreement DW-
14938973 in developing Stage II and Stage III tools used for catchment aggregation and creation of
hydrologicaUy-corrected digital elevation models, without which this project would not have been
possible. We also gratefully acknowledge the contributions of Sharon Batterman (US EPA MED-
Duluth, IAG Project Officer) and GIS staff of US EPA Region III, in particular, Don Evans and
Carmen Constantine, in helping to create Stage II and Stage III 10- and 12-digit HUC boundaries
for the state of West Virginia; and of Wendy Bkke-Coleman, US EPA Office of Environmental
Information, for facilitating this arrangement. We thank Hope Childers, contract staff at US EPA
Region III, Wheeling, WV office for providing the mining-modified NLCD land-use coverage for
West Virginia. We thank US EPA staff at Region III- Philadelphia, PA and Fort Meade, MD (Tom
DeMoss, Tom Pheiffer) and Wheeling, WV offices (Maggie Passmore, Frank Borsuk); state of WV
DNR and DEP staff (Dan Cincotta, WV DNR; Pat Campbell, Dave Montali, John Wilts, WV
DEP); and staff of the Canaan Valley Institute (Tom DeMoss, Ron Preston) for consultation on all
51
-------
aspects of this project We thank Debra Taylor at the US EPA MED-Duluth for help in data entry
and USGS gaging station watershed delineations. We gratefully acknowledge the GIS support
provided by OAO Corporation under US EPA FAIR Contract 68-W5-0065, Delivery Order #24;
and Computer Sciences Corporation under US EPA FAIR II Contract 68-W01/W02-032, Task
Order #024 in watershed characterization and map production. We also thank Peg Pelletier and
Jonathan Smith for providing helpful review comments on an earlier draft of this report.
-------
REFERENCES
Boward, D.M., P.P. Kazyak, S.A. Stranko, M.K. Kurd, and T.P. Prochaska. 1999. From the
Mountains to the Sea: The State of Maryland's Freshwater Streams. EPA/903/R-99/023.
Maryland Department of Natural Resources, Monitoring and Nontidal Assessment
Division, Annapolis, MD.
Brazner, J.C., D.K. Tanner, N.E. Detenbeck, S.L. Batterman, S.K. Stark, L.A. Jagger, and V.M.
Snarski. 2003. Landscape influences on fish assembkge structure and function in western
Lake Superior tributaries. Submitted to Environmental Management.
Cincotta, D. 2000. A Small Watershed Characterization, Classification, and Assessment for West
Virginia Utilizing EMAP Design and Tools. R-EMAP proposal submitted to US EPA
Mid-Continent Ecology Division, Duluth, MN and US EPA Region III, Philadelphia, PA.
Daly, C., R.P. Neilson, and D.L. Phillips. 1994. A statistical-topographic model for mapping
climatological precipitation over mountainous terrain. Journal of Applied Meteorology,
33:140-158.
Detenbeck N.E., D. Cincotta, J.M. Denver, S.K. Greenlee, and A.R. Olsen. 2003a. Watershed-
based survey designs. Submitted to Environmental Monitoring and Assessment (accepted with
minor modifications).
Detenbeck N.E., C.M. Elonen, D.L. Taylor, L.E. Anderson, T.M. Jicha, and S.L. Batterman.
2003b. Effects of hydrogeomorphic region, watershed storage and mature forest on
baseflow and snowmelt stream water quality in second-order Lake Superior Basin
tributaries. Freshwater Biology 48(5):911-27.
Detenbeck, N.E., S.L. Batterman, V.J. Brady, J.C. Brazner, V.M. Snarski, D.L. Taylor, J.A.
Thompson, and J.W. Arthur. 2000. A test of watershed classification systems for ecological
risk assessment. Environmental Toxicology and Chemistry 19(4):1174-1181.
FGDC. 2002. Federal Standards for Delineation of Hydrologic Unit Boundaries. Federal
Geographic Data Committee. http://www.ftw.nrcs.usda.gov/huc_data.html
Franken, S.K., D.J.Tyler, and K.L. Verdin. 2001. Development of a National Seamless Database
of Topography and Hydrologic Derivatives, Paper 730 in Proceedings, 2001 ESRI
International User's Conference, San Diego, CA.
Frye, P.M. and G.S. Runner. 1970. A Proposed Streamflow Data Program for West Virginia, U.S.
Dept. of the Interior, Geological Survey Water Resources Division, Open-file report,
Charleston, WV.
Heiskary, S.A. and C.B. Wilson 1990. Minnesota lake water quality assessment report. Minnesota
Pollution Control Agency, St. Paul, MN.
53
-------
Herlihy A.T., D.P. Larsen, S.G. Paulsen, N.S. Urquhart, and BJ. Rosenbaum. 2000. Designing a
spatially balanced, randomized site selection process for regional stream surveys: The
EMAP Mid-Atlantic pilot study. "Environmental Monitoring and Assessment, 63:95-113.
Jennings, M.E., W.O. Thomas, Jr., and H.C. Riggs. 1993. Nationwide Summary of U.S. Geological
Survey Regional Regression Equations for Estimating Magnitude and Frequency of Floods
for Ungaged Sites, 1993. U.S Geological Survey, Reston,VA. Water Resources
Investigations Report WRI 94-4002.
Johnston, C.A., N.E. Detenbeck, and GJ. NiemL 1990. Cumulative impacts of wetland loss on
stream water quality and quantity. Biogeochemistry 10:105-141.
Legleiter, KJ. 2001. Interagency Development of National Watershed and Subwatershed
Hydrologic Units, Paper 492 in Proceedings, 2001 ESRI International User's Conference,
San Diego, CA.
NVPDC (Northern Virginia Planning District Commission). 1980. In Schueler, R.T. 1984. The
importance of imperviousness. Watershed Protection Techniques 1(3):100-111.
Omemik, J.M. and A.L. Gallant. 1988. Ecoregions of the upper Midwest states. EPA/600/3-
88/037, U. S. Environmental Protection Agency, Environmental Research Laboratory,
Corvallis, OR.
Pan, Y., R.J. Stevenson, B.H. Hill, and A.T. Herlihy. 2000. Ecoregions and benthic diatom
assembkges in Mid-Atlantic Highlands streams, USA. Journal of the North American
Benthological Society, 19:518-540.
Richards, R.P. 1990. Measures of flow variability and a new flow-based classification of great lakes
tributaries. J. Great Lakes Res. 16:53-70.
Rosgen, D. 1996. Applied river morphology. Wildland Hydrology, Pagosa Springs, CO.
Runner, G.S. 1980. Hydrologic Data for Runoff Studies on Small Drainage Areas, West Virginia
Department of Highways, Research Project 16, Open-File Report 80-560, Charleston,
WV.
SAS. 1990. SAS/STAT Users Guide. Version 6, Fourth edition. SAS Institute, Gary, NC.
Schueler, T.R. 1994. The importance of imperviousness. Watershed Protection Techniques 1 (3):100-111.
Scott, A.G. 1984. Analysis of characteristics of simulated flows from small surface-mined and
undisturbed Appalachian watersheds in the Tug Fork basin of Kentucky, Virginia, and
West Virginia. U.S. Geological Survey, Reston, VA. Water Resources Investigations
Report WRI 84-4151.
54
-------
Tasker, G.D., and J.R. Stedinger. 1989. An operational GLS model for hydrologic regression.
Journal of Hydrology 111:361-375.
US EPA. 2001. The National Costs of the Total Maximum Daily Load Program (Draft Report).
U.S. Environmental Protection Agency, Office of Water, Washington, DC EPA/841/D-
01/003.
US EPA. 2002. Consolidated Assessment and Listing Methodology: Toward a Compendium of
Best Practices. First Edition. U.S. Environmental Protection Agency, Office of Wetlands,
Oceans, and Watersheds. Washington, DC
(http://www.epa.gov/owow/monitoring/calm.html).
US EPA. 2003. National 303(d) List Fact Sheet.
(http: / /oaspub.epa.gov/waters/national_rept. control).
Verdin, K.L., and J.P. Verdin. 1999. A topological system for delineation and codification of the
earth's river basins. /. Hydrol. 218:1-12.
Verry, E.S. 1986. Forest harvesting and water: The Lake States experience. Water Resources Bulletin
22:1039-1047.
Waite, I.R., A.T. Herlihy, D.P. Larsen, and D.J. Klemm. 2000. Comparing strengths of geographic
and nongeographic classifications of stream benthic macroinvertebrates in the
Mid-Atlantic Ffcghlands, USA. Journal of the North American Onthological Society, 19:429^-41.
Wilkinson, L. 1999. Systat 9.0 Statistics I. SPSS Inc. Chicago, IL.
55
-------
SEPA
United States
Environmental Protection
Agency
Please make all necessary changes on the below label,
detach or copy, and return to the address in the upper
left-hand comer.
If you do not wish to receive these reports CHECK HEREfJ ;
detach, or copy this cover, and return to the address in the
upper left-hand corner.
PRESORTED STANDARD
POSTAGE & FEES PAID
EPA
PERMIT No. G-35
------- |