4>EPA
United States
Environmental Protection
Agency
Guidelines to Assessing
Regional Vulnerabilities
RESEARCH AND DEVELOPMENT
-------
-------
EPA/600/R-08/XXX
April 2008
www.epa.gov
Guidelines to Assessing
Regional Vulnerabilities
by
Elizabeth R. Smith1, Megan H. Mehaffey1, Robert V. O'Neill2,
Timothy G. Wade1, J. Vasu Kilaru1, and Liem T. Tran3
1U.S. Environmental Protection Agency
Office of Research and Development
National Exposure Laboratory
Research Triangle Park, NC
2TN and Associates
Oak Ridge, TN
3University of Tennessee
Knoxville, TN
Notice: Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official
Agency policy. Mention of trade names and commercial products does not constitute endorsement or
recommendation for use.
U.S. Environmental Protection Agency
Office of Research and Development
Washington, DC 20460 073cmb08
-------
11
-------
Notice
The U.S. Environmental Protection Agency (U.S. EPA), through its Office of Research and
Development (ORD), funded and managed parts of the research described here under contract
EP-D-06-072 with TN and Associates, Inc. It has not yet been subjected to the Agency's peer
and administrative review and has not yet been approved for publication as an EPA document.
Final Draft iii
-------
IV
Final Draft
-------
Acknowledgments
Many people contributed to the Regional Vulnerability Assessment methods development
described in this report. Specifically, we would like to acknowledge the following:
Rochelle Araujo, EPA, National Exposure Research Laboratory
Tim Johnson, EPA, National Risk Management Research Laboratory
K. Bruce Jones, U.S. Geological Survey
Rick Linthurst, EPA, Office of Research and Development
Peter McKinnis, Waratah Corporation
Michael O'Connell, Waratah Corporation
Roger Tankersley, Tennessee Valley Authority
Paul Wagner, EPA, National Exposure Research Laboratory
Lisa Wainger, University of Maryland
Dennis Yankee, Tennessee Valley Authority
Preferred Citation:
Smith, E.R., Mehaffey, M.H., O'Neill, R.V., Wade, T.G., Kilaru, J.V., and Tran,
L.T. 2008. Guidelines to Assessing Regional Vulnerabilities.
Final Draft v
-------
VI
Final Draft
-------
Executive Summary
Environmental decision-makers today are faced with declining budgets, lack of problem-
focused monitoring data, and issues that range from subtle and slow (such as changes in
species composition) to conspicuous and immediate (e.g., catastrophic events). At the
same time, there is greater recognition that environmental decisions that are made today
are likely to impact human well-being in the future. Thus, there is a growing desire to
evaluate potential decisions with regard to their future implications. Further, in
attempting to reach a decision, an environmental decision-maker can quickly become
overwhelmed by the huge amounts of disparate types of data and information that are
available on resources, conditions, and stressors within a region. The EPA's Regional
Vulnerability Assessment (ReVA) program was designed to deal with these problems:
ReVA methodology establishes a platform that can help environmental decision-makers
target limited resources and enable proactive decision-making.
ReVA has a broad spatial perspective, uses existing data, and applies an integrated
approach to assessment; it can incorporate large, disparate sources of available spatial
data on resources, environmental conditions, and stressors, and then visually express
these conditions (or combinations of these conditions) in map form. ReVA methods also
allow users to prepare "what if scenarios; these scenarios permit inspection of likely
future changes in environmental vulnerabilities, given user-determined inputs on
anticipated regional changes in factors such as population growth, economic conditions,
land use, transportation infrastructure, etc. ReVA can improve the environmental
decision-making process by permitting more realistic inputs for environmental decision-
making and by expressing results of multiple factors at a regional spatial scale.
Since 1998, much of the research effort within the ReVA program has focused on the
mechanics of how data and model results can be integrated into meaningful indices
designed to address specific assessment questions posed by environmental decision-
makers. The approach developed by the ReVA program allows decision-makers to
evaluate current conditions and vulnerabilities through the use of indices. This approach
allows an evaluation of net change, so that the user can visualize how both positive and
negative changes affect future conditions and vulnerabilities.
ReVA's approach as presented in these guidelines includes the following steps:
• Acquisition of spatially explicit data
• Data processing
• Metric selection and integration
• Development/selection of spatially explicit models
• Creation of alternative scenarios
• Synthesis
• Results communication
Final Draft vii
-------
Vlll
Final Draft
-------
Table of Contents
Notice iii
Acknowledgments v
Executive Summary vii
List of Figures xi
List Abbreviations and Acronyms xiii
Section 1 - Introduction and Background 1
Why Look at the Broad Scale? 1
Vulnerability versus Risk 2
The Need for an Integrated Approach 2
Section 2 - Data Used in ReVA 5
Types of Spatial Data 5
Data Inputs for ReVA 5
National Data Sources 5
Regional Data Sources 8
Local Data Sources 9
Data Quality Considerations 9
Section 3 - Data Processing in ReVA 11
Database Management 11
Reporting Units 11
Data Reapportionment 12
Missing Data 14
Metric Selection and Preparation for Integration 16
Section 4 - Spatially Explicit Models 21
Why Models? 21
Types of Models 21
Empirical Models 21
Process Models 22
Examples of Spatial Models Used in ReVA 22
Nitrate and Sulfate Deposition Modeling 22
Mercury Deposition Modeling 22
Invasive Species Modeling 23
Nitrate in Ground Water Modeling 24
Forecasting Drivers of Change 24
Land-Use Change Models as a Component of ReVA 25
Models that Use Land Cover/Land Use as Input 27
Resource Extraction 27
Pollutants/Changes in Water Quality 28
Spread of Invasive Non-Indigenous Species 28
Models that Do Not Include Land Use/Land Cover as Input 28
Final Draft ix
-------
Section 5 - Creating Alternative Future Scenarios 31
Building Alternative Scenarios for Analyzing Future Trends 31
Scale Considerations in Proj ective and Prospective Modeling 31
Spatial Scale 32
Temporal Scale 32
Anticipating Responses to Policies 32
Using Other Spatial or Monitoring Data to "Spatialize" Scenarios 33
Section 6 - Synthesis 35
Available Information and Data Preparation 35
Methods for Integrating Variables 36
Simple Sum and PCA Sum 36
Best and Worst Quintiles 37
State Space Method 38
Criticality Analysis 38
Overlay Method 38
Stressor-Resource Matrix 39
Moving to Smaller Scales 39
Uncertainty in ReVA Analyses 40
Section 7 - Results Communication 41
Audience and Assessment Needs 41
Visualization 42
Glossary 51
References 59
Final Draft
-------
List of Figures
Figure 1. Schematic depicting differences in the spatial (X axis) and temporal (Y axis)
scales of ecosystem responses to various types of stressors 2
Figure 2. Schematic of steps in the ReVA approach. ReVA's Environmental Decision
Toolkit (EDT) is used for synthesis, scenario analysis and communication of
results by visual representations 3
Figure 3. Example of simple apportionment of data by area-weighting. Hydrologic Unit
Code 1 (HUC 1) contains 20% of the population designated by the shaded
area, while HUC 2 contains 80% of the population. The boundary between
HUC 1 and HUC 2 is represented by the dashed line 13
Figure 4. Example of apportioning population data for small reporting units (represented
by squares) for an urban area (shaded area) that occurs in two HUCs. Values
shown in the counties represent percentage of the urban area's population;
thus, they sum to 100 14
Figure 5. Graph (hypothetical X and Y axes) showing measured data (solid circles), an
interpolated point, and an extrapolated point 15
Figure 6. Graphic depicting 2000 distribution of Giant Salvinia (Salvinia molesta) and
estimated 2020 distribution using the Genetic Algorithm for Rule-set
Prediction (GARP) model 23
Figure 7. Graphic showing NAWQA sample sites (map on left) and results of logistic
regression model that estimates probability of exceeding a threshold of nitrate
concentration in shallow ground water aquifers across the Mid-Atlantic region.
24
Figure 8. Graphic depicting current land use/land cover based on the National Land
Cover Database (NLCD) 1992 (left) and estimated 2020 land use/land cover
using the Slope, Land use, Exclusion, Urban, Transportation, Hillshading
model (SLEUTH) in combination with planned roads and permitted mines
(right) 27
Figure 9. Graphic depicting an example of variables and indices produced for different
levels of users within the Sustainable Environment for Quality of Life (SEQL)
project 42
Figure 10. Graphic depicting a scatter plot comparing two variables, nonpoint source
nitrogen loadings as estimated by the model LTHIA and percent crop
agriculture along streams within a 60-m buffer 43
Figure 11. Graphic depicting a histogram of number of watersheds and percent crop
agriculture within a 60-m buffer along streams 43
Figure 12. Graphic depicting a box plot of nonpoint source nitrogen with percent crop
agriculture within a 60-m buffer along streams 44
Final Draft xi
-------
Figure 13. Graphic depicting a screenshot from a ReVA Environmental Decision Toolkit
(EDT) with a radar plot for a displayed 8-digit HUC. In a radar plot, each
spoke of the wheel represents an individual variable and the amount of green
represents the relative rank of that variable in relation to the same variable in
all other 8-digit HUCs across the region (green represents good conditions, not
green represents poor conditions) 44
Figure 14. Graphic showing the percent of forest cover for every 8-digit HUC across
EPA Region 5 as displayed using quintiles as the binning method 45
Figure 15. Graphic showing the percent of forest cover for every 8-digit HUC across
EPA Region 5 as displayed using equal intervals as the binning method 46
Figure 16. Graphic showing the percent of forest cover for every 8-digit HUC across
EPA Region 5 as displayed using natural breaks as the binning method 46
Figure 17. Graphic depicting linked micromaps 48
Figure 18. Graphic depicting the comparison between two future alternative scenarios
(upper maps) with a difference map highlighting both individual watershed
differences as well as overall regional differences 49
xii Final Draft
-------
List of Abbreviations and Acronyms
ATtlLA
BASINS
BIOCLIM
CMAQ
OEMs
EOT
FGDC
FWS
GARP
GIS
HUC
ICLUS
IDW
LANDSAT
LTHIA
MAIA
NASS
NATA
NAWQA
NCDC
NED
NHD
NTS
NLCD
NOAA
NRCS
NWI
PCA
ReVA
RGDT
RUSLE
SAB
SEQL
Analytical Tools Interface for Landscape Assessments
Better Assessment Science Integration Point and Nonpoint Sources
Bioclimatic prediction system
Community Multiscale Air Quality
Digital Elevation Models
Environmental Decision Toolkit
Federal Geographic Data Committee
Fish and Wildlife Service
Genetic Algorithm for Rule-set Production
Geographic Information System
Hydrologic Unit Code
Integrated Climate Land Use Scenarios
Inverse Distance Weighting
Land Remote Sensing Satellite
Long-Term Hydrologic Impact Assessment
Mid-Atlantic Integrated Assessment
National Agricultural Statistical Survey
National Air Toxics Assessment
National Water-Quality Assessment Program
National Climatic Data Center
National Elevation Dataset
National Hydrography Database
Non-Indigenous Species
National Land Cover Dataset
National Oceanic and Atmospheric Administration
National Resources Conservation Service
National Wetlands Inventory
Principle Components Analysis
Regional Vulnerability Assessment
Regional Growth Decision Tool
Revised Universal Soil Loss Equation
Science Advisory Board
Sustainable Environment for Quality of Life
Final Draft
Xlll
-------
SERGoM
SETAC
SLEUTH
SSURGO
STATSGO
TIGER
U.S. EPA
USCB
USDA
USGS
USLE
XML
Spatially Explicit Regional Growth Model
Society of Environmental Toxicology and Chemistry
Slope, Land use, Exclusion, Urban, Transportation, Hillshading
model
Soil Survey Geographic database
State Soil Geographic database
Topologically Integrated Geographic Encoding and Referencing
U.S. Environmental Protection Agency
U.S. Census Bureau
U.S. Department of Agriculture
U.S. Geological Survey
Universal Soil Loss Equation
Extensible Markup Language
xiv
Final Draft
-------
Section 1
Introduction and Background
Decision-makers today face increasingly complex environmental problems that require
integrative and innovative approaches for analyzing, modeling, and interpreting various
types of information. ReVA acknowledges this need and is designed to evaluate methods
and models for synthesizing diverse kinds of available information on the distribution of
stressors and sensitive ecological resources. As with any study, the first and probably
most important step is to establish a clear goal. For ReVA, the goal is to develop and
demonstrate approaches that use existing data to evaluate current and future conditions
and vulnerabilities of valued resources (native biodiversity, water quality, forest
productivity, etc.) resulting from ecological drivers of change1 and later, management
alternatives.
Why Look at the Broad Scale?
ReVA is designed to help decision-makers use existing data and model results at a broad
scale, allowing insights into (1) where problems are likely to occur in the future, (2) what
environmental stresses are likely to be of most concern, and (3) how alternative
management decisions might play out in terms of trade-offs across the region. The
broad-scale approach is important for several reasons. First, by stepping back and
assessing landscape (regional scale) characteristics and the distribution of resources and
stressors, spatial relationships become apparent. Over time, land use and invasive species
may change and can be expected to move across the landscape. Identifying where these
things are currently occurring can provide insights as to when and where these issues will
occur in the future. Second, many of the drivers of ecological change occur at the
regional scale over fairly long time periods (Figure 1). Thus, a broad-scale approach is
necessary to capture these changes, for they could be easily overlooked at a finer scale.
Third, evaluating projected changes at a broad scale enables strategic management
responses by considering what is best for the region overall, even while managing finer-
scale risks or problems that are unavoidable or are part of the trade-offs that come with
any environmental decision.
Drivers of ecological change are generally accepted as including land-use change, invasive non-indigenous
species, resource extraction, pollution and pollutants, and climate change (Chambers et al., 2007).
Final Draft
-------
CO
c
o
I
E
to
CO
o
8
2,
Invasive
Species
Climate Change
I
Resource and Land Use
Extreme
Natural
Events
Pollution
Local Regional National
Geographic Scale of Ecosystem Response
Figure 1. Schematic depicting differences in the spatial (X axis) and temporal (Y axis) scales of
ecosystem responses to various types of stressors.
Vulnerability versus Risk
As its name implies, ReVA is based on vulnerability assessment; it examines a broad
range of information across a region and attempts to identify areas where as-yet-
unidentified endpoints might be vulnerable. ReVA accomplishes this objective by
applying environmental indicators (or descriptive metrics) to represent important
endpoints and examines the co-occurrence of valued resources and stressors to represent
vulnerability of sensitive endpoints to potential harm. The techniques used to examine
how stressors and resources combine seek to reveal threats that are often not clearly
identifiable or quantifiable, and allow the users of ReVA output to explore complex
interdependences of related issues (cf. Liotta, 2005; Liotta and Miskel, 2004).
The Need for an Integrated Approach
In addition to taking a broad spatial perspective, ReVA stresses an integrated approach to
assessment. This emphasis imparts greater realism to environmental decision-making by
presenting problems simultaneously to permit the decision-maker a broader perspective
in identifying the most vulnerable resources within a region. In considering all resources,
Final Draft
-------
conditions, and stressors, decision-makers typically confront huge amounts of data which
results in a challenge to make the information meaningful. These difficulties are
addressed by the ReVA approach (Figure 2), which allows decision-makers to evaluate
current conditions and vulnerabilities using indices. The use of indices permits the
decision-maker to evaluate how positive and negative changes affect future conditions
and vulnerabilities.
•U
Extrapolation / interpolation
Model development / forecasting
EOT
Synthesis
-U
Scenario analysis
43-
Visualization / communication / access to information
Figure 2. Schematic of steps in the ReVA approach. ReVA's Environmental Decision Toolkit (EOT)
is used for synthesis, scenario analysis and communication of results by visual
representations.
Final Draft
-------
Final Draft
-------
Section 2
Data Used in ReVA
Types of Spatial Data
Geospatial data typically have two basic components: (1) the location or geographic
context and (2) attributes of that location or area. The geographic context, in turn, falls
into several categories: point, line, polygon, and grid. A point is simply a discrete
location of an entity, designated by x and y coordinate values, such as an air monitoring
station or a soil sampling site; a line abstraction is used to represent linear objects such as
roads or streams; and polygons represent areas such as political borders or water bodies.
A grid is a special raster-based (cell-based) geography where all the cells in the grid are
square and equal in area and each cell contains only one value for the variable of concern.
Typical gridded datasets are used for variables such as elevation (e.g., a digital elevation
model) and land-cover classification. Points, lines, and polygons may have any number
of attributes associated with a single element. For example, a polygon that represents a
county may have attributes such as area, perimeter, population, and per capita income.
Data Inputs for ReVA
The data required to perform ReVA-type analyses (see Figure 2) may be acquired from
many sources. The area or region of concern, existing resources, types of stressors, and
the questions and concerns about the region determine the data requirements. The main
requirement is that the data used in any analysis must be collected consistently across the
region of concern. Data are available from various sources at the national, regional,
State, and local levels. National sources include many federal agencies such as the
United States Geological Service (USGS) and the United States Environmental
Protection Agency (U.S. EPA). An example of a regional source of information might be
the U.S. EPA Chesapeake Bay Program Office. States, too, have geographic data
holdings, but the extent and quality of these datasets can vary widely. Finally, at the
finest scale, counties and local municipalities have geographic data at very local scales.
These datasets can include land-parcel data and zoning information. The data in these
local-scale datasets often vary greatly among local areas in terms of level of detail and
quality, making combination across boundaries difficult.
National Data Sources
ReVA uses a number of datasets that are available for at least the conterminous states.
With these base layers, numerous landscape and environmental metrics can be computed
for most areas in the nation. One of the most useful Web sites for obtaining data at a
national scale is operated by the USGS. This Web site can be accessed at:
Final Draft
-------
http://seatnless.usgs.gov/website/seatnless/viewer.php. The following are examples of
datasets that can be downloaded from this site.
• NLCD 1992 and 2001 - The National Land Cover Dataset (NLCD) is a gridded
dataset that contains consistently (within a year) collected and processed imagery with a
land-cover classification scheme for the entire U.S. The 1992- and 2001-era data are
nationally available and can be downloaded from the NLCD Web site:
http://www.mrlc.gov/index.asp. Significant changes were made to the processing
methodology of Land Remote Sensing Satellite (LANDS AT) imagery which makes
direct comparison of NLCD 1992 and NLCD 2001 difficult.
• NED - The National Elevation Dataset (NED) is a gridded dataset that contains
elevation values for each grid cell; such datasets are referred to as digital elevation
models (OEMs). These data are available at several scales. Typically the 30-meter (or
1/3 arc-second) data are used; this dataset is available nationally. The 10-meter (or 1/9
arc-second) data also are available for some areas. Due to their fine scale, the 30- and
10-meter elevation datasets are large. For some applications, it may be acceptable to use
a larger-scale grid, such as the 100-meter (1 arc-second) dataset.
• National Atlas - Data from the USGS National Atlas are also available at the USGS
"seamless" server site. However, for better descriptions and access to metadata, it is
helpful to visit the site at: http://nationalatlas.gov/pros.html. The National Atlas contains
spatial datasets on diverse variables, including: the 2002 Census of Agriculture, breeding
bird survey locations, invasive species, forest fragmentation estimates, vegetation growth,
West Nile virus surveillance, wildlife mortality, and other variables, encompassing
geology, climate, environment, transportation, and water.
• NHD - The National Hydrography Dataset (NHD) is a 1:100,000-scale digital
representation of the nation's streams and rivers. NHD is very useful in many landscape
analyses, especially in conjunction with OEMs and land cover. It also is useful for
hydrologic modeling and is populated with various attributes that allow analysis of flow
networks. The NHD is available at: http://nhd.usgs.gov/data.html.
• NWI - Maintained by the Fish and Wildlife Service (FWS), the National Wetlands
Inventory (NWI) is a digital spatial dataset of the wetlands in the U.S. and is available at:
http://wetlandsfws.er.usgs.gov/NWI/download.html.
• TIGER/Line 2000 - The Census 2000 TIGER/Line shapeftles were created from the
Topologically Integrated Geographic Encoding and Referencing (TIGER) database of the
United States Census Bureau (USCB). The shapeftles contain data on the following: line
features such as roads, railroads, hydrography, and transportation and utility lines;
boundary features such as statistical (e.g., census tracts and blocks), government (e.g.,
places and counties), and administrative (e.g., congressional and school districts); and
boundaried and landmark features such as point (e.g., schools and churches), area (e.g.,
parks and cemeteries), and key geographic locations (e.g., apartment buildings and
factories). A number of vendors offer value-added products that improve on the USCB's
Final Draft
-------
version of the data. Freely available USCB data can be accessed at:
http://arcdata.esri.com/data/tiger2000/tiger download.cfm.
• Census 2000 - The USCB also administers the decadal census. While numerous
products are available, the more detailed demographic data can provide useful
information about housing, income, education, race, age, gender, and other socio-
economic indicators. Like other U.S. government products, many vendors offer value-
added products that build on the basic data collected by the USCB. For more
information, visit: http://www.census.gov/main/www/cen2000.html.
• Soil data - The State Soil Geographic database/Soil Survey Geographic database
(STATSGO/SSURGO) are geographic databases maintained by the Natural Resources
Conservation Service (NRCS) that contain generalized soil types. The datasets were
created by generalizing more detailed soil survey maps. For STATSGO, where more
detailed soil survey maps were not available, data on geology, topography, vegetation,
and climate were assembled, together with LANDS AT images. Soils of like areas were
studied, and the probable classification and extent of the soils were determined. Map unit
composition was determined by transecting or sampling areas on the more detailed maps
and expanding the data statistically to characterize the whole map unit.
The STATSGO dataset consists of geo-referenced vector digital data and tabular digital
data. The map data were collected in 1- by 2-degree topographic quadrangle units and
merged into a seamless national dataset. It is distributed in state/territory and national
extents. The soil map units are linked to attributes in the tabular data, which give the
proportionate extent of the component soils and their properties.
The tabular data contain estimated and measured data on the physical and chemical soil
properties, soil interpretations, and static and dynamic metadata. Most tabular data exist
in the database as a range of soil properties, depicting the range for the geographic extent
of the map unit. In addition to low and high values for most data, a representative value is
also included for these soil properties. For more information, see:
http://www.ncgc.nrcs.usda.gov/products/datasets/statsgo/data/index.html.
The STATSGO database is being updated and renamed to the Digital General Soil Map
of the United States. The updated version will be available for download from the Soil
Data Mart: http://soildatamart.nrcs.usda.gov/.
The STATSGO database is designed primarily for regional, multistate, river basin, state,
and multicounty resource planning, management, and monitoring. It is not detailed
enough for analyses at the county level or finer-scale. The SSURGO dataset is much
more detailed than STATSGO. It is designed primarily for farm and ranch,
landowner/user, township, county, or parish natural resource planning and management.2
Pennsylvania State University Cooperative Agriculture Extension, November 2007.
http://lal.cas.psu.edu/software/tutorials/soils/st diff.html
Final Draft
-------
Currently, plans are for the digital data for SSURGO to be completed in 2008. For more
information on SSURGO, see: http://soildatamart.nrcs.usda.gov/SSURGOMetadata.aspx.
• Omernik Ecoregions - Ecoregions are areas of the landscape that are classified into
regions on the basis of geology, physiography, vegetation, climate, soils, land use,
wildlife, and hydrology. For more information and access to data that can be
downloaded, visit: http://www.epa.gOv/wed/pages/ecoregions/level_iii.htm.
• Climate Data - The National Oceanographic and Atmospheric Administration's
(NOAA's) National Climatic Data Center (NCDC) collects and disseminates climate data
that includes such parameters as temperature, precipitation, and wind speeds. These data
are available for download at: http://lwf.ncdc.noaa.gov/oa/climate/climatedata.html.
At least three other national-scale sources for environmental data can be accessed for use
in ReVA:
• NOAA Geophysical Data - NOAA provides access to a wide variety of geophysical
data. These can be accessed at: http://www.ngdc.noaa.gov/ngdcinfo/onlineaccess.html.
• A site operated by Collins Software (Houston, Texas) contains links to various GIS
data: http://www.collinssoftware.com/freegis_by_region.htm
• Digital Watershed - This site, maintained by Michigan State University, includes
spatial data and models similar to those found in EPA's Better Assessment Science
Integrating Point & Nonpoint Sources (BASINS) program (see:
http://www.epa.gov/waterscience/basins/). The Digital Watershed can be found at:
http://www.iwr.msu.edu/dw/.
Regional Data Sources
Because ReVA focuses on regions, regional data sources can be well-suited for ReVA
applications. The types of spatial data available at regional scales obviously vary with
the region of interest. To date, ReVA has used regional datasets from the following
sources:
• The Chesapeake Bay watershed has long been an area of concern and widely studied.
The Chesapeake Bay Program databases can be queried based upon user-defined inputs
such as geographic region and date range. Each query results in a downloadable, tab- or
comma-delimited text file that can be imported to programs such as SAS, Excel, or
Access for further analysis. Chesapeake Bay Program databases can be found at:
http://www.chesapeakebay.net/data/. GIS data for the Chesapeake Bay monitoring
program are available at:
http://www.chesapeakebay.net/data/data_desc.cfm?DB=CBP_GIS
• The Mid-Atlantic Integrated Assessment (MAIA) encompasses the Chesapeake Bay
watersheds, but extends farther, including the Mid-Atlantic states. Due to its high
population density and rapid growth in population, the Mid-Atlantic region has been
8 Final Draft
-------
studied intensively. Data for this region can be obtained at:
http ://www. epa. gov/maia/html/data.html
• The Southeastern Ecological Framework is a comprehensive set of spatial data on
ecological resources and habitat for the Southeastern United States (U.S. EPA Region 4).
These data can be found at: http://www.geoplan.ufl.edu/epa/connectivity.
Local Data Sources
With the spread of GIS technology for integrating and managing municipal functions,
many cities, towns, and counties now generate and manage spatial data at the local scale.
Local datasets include information on variables such as school and fire district maps, and
zoning and land-use maps. Examples of local datasets for Wake County, North Carolina,
can be reviewed at: http://www.lib.ncsu.edu/gis/wake.htmltflayers.
Data Quality Considerations
The usefulness of any data depends upon their quality. For geospatial data, the Federal
Geographic Data Committee (FGDC) has established metadata standards. Many spatial
datasets now come with metadata files in text, html, or XML formats. These metadata
describe the nature of the data, its lineage, the procedures that were used in processing
and generating the data, and the potential uses and limitations of the data.
Two main data-quality elements are of concern when using spatial data: locational
accuracy and attribute accuracy. Locational accuracy refers to the accuracy of
information about the spatial location. For example, if the location of a soil-sampling site
is given in latitude and longitude, the associated metadata should reflect the accuracy of
that measurement (i.e., within 10 meters). Attribute accuracy refers to the accuracy
measurement of the variable of interest at the location. Again, using the same soil-
sampling example, the accuracy of a measured constituent in the soil (such as cadmium)
at the location of interest might be plus or minus 5 parts per million. Frequently, further
processing or generalization of the data may introduce additional uncertainties that
should also be documented and considered.
Final Draft
-------
10
Final Draft
-------
Section 3
Data Processing in ReVA
Once all of the individual core datasets are acquired, they must be assembled into a single
GIS database and one or more spatial units must be selected for reporting final results
(Figure 2). Some data may need to be reapportioned if their collecting or enumeration
unit differs from those of the reporting unit. Additionally, missing data may need to be
estimated by interpolation or extrapolation to complete a dataset. Then, metrics can be
calculated or modeled (modeling is discussed in the next chapter). Finally, appropriate
metrics can be identified from the full suite of variables for integration using ReVA
integration tools.
Database Management
It is good practice to choose a single projection and datum (reference point) for storing all
spatial data before generating metrics. In the Mid-Atlantic study, for example, two raster
datasets (NED and NLCD) were used to prepare several metrics. The native projection
for both of these datasets was standard U.S. Albers, NAD83. Projecting raster data
requires resampling the data, and should be avoided if possible. Therefore, U.S. Albers,
NAD83 was chosen for the Mid-Atlantic study, and all data in other projections were
converted to this projection before further processing.
Reporting Units
Descriptive metrics must be summarized and reported for specific areas. These areas,
called reporting units, need to be of appropriate scale and relevant to the study. Some
examples of commonly used reporting units are political boundaries (such as counties),
naturally-defined areas (such as watersheds or ecoregions), or equal-sized cells in a
square or hexagonal grid. Each type of reporting unit has advantages and disadvantages.
Watersheds are a good choice for reporting units for water-quality studies: for many
watersheds, data are available for various factors at the watershed outlet. Further, most
stressors and resources that can affect the sample data are contained within the reporting
unit. In the ReVA Mid-Atlantic study, the 8-digit Hydrologic Unit Code (HUC) was one
of the reporting units used. HUCs are advantageous in that they can be scaled in size,
from 2-digit (the largest) to 12-digit (the smallest). Currently, HUCs that are smaller than
8 digits are not available for the entire United States, but are available for some areas.
An advantage of counties as a reporting unit is they often represent the decision-maker's
area of interest. Further, information in one of the core datasets, census data, is collected
by county, or even by smaller units that nest within county boundaries. Disadvantages of
Final Draft 11
-------
counties as reporting units are that county boundaries may not correlate well with natural
boundaries, and they cannot be scaled.
Grid cells as reporting units have several advantages. They are all the same size, which
can facilitate comparisons between areas. Grid cells also can make it easier to notice
patterns in indicator maps. Finally, grid cells can be scaled, meaning that the user may
select any size for the cells. Unfortunately, grid cells do not match either decision-
making boundaries or natural boundaries, which is a significant drawback. Further, grid
placement is arbitrary, so shifting the grid may substantially change some indicator
values in some cells.
Data Reapportionment
Some data, such as socio-demographic or economic data from the USCB, are collected
by specific areas; these data are generally enumerated by county. When the collecting or
enumerating area and reporting unit boundaries do not match, data must be apportioned
from one area to the other. The easiest way to do this is by area-weighting. For example,
if 20% of a county is located in HUC 1 and 80% is located in HUC 2, then 20% of the
population for the county would be assigned to HUC 1 and 80% would be assigned to
HUC 2 (Figure 3). An area-weighting method involves the assumption that values (the
number of people, in the current example) are evenly distributed across the county, which
is obviously incorrect in many or all cases.
12 Final Draft
-------
\ HUC2
HUC1
Figure 3. Example of simple apportionment of data by area-weighting. Hydrologic Unit Code 1 (HUC
1) contains 20% of the population designated by the shaded area, while HUC 2 contains
80% of the population. The boundary between HUC 1 and HUC 2 is represented by the
dashed line.
For census data, a better approach for reapportionment makes use of the fact that
population and housing units are enumerated by smaller block groups within counties.
County-level variables, such as the number of children under five years of age, can be
apportioned to block groups based on proportion of the county population contained in
the block group. If, for example, a county has 1,000 children under five, and block group
1 contains 2% of the county's total population, then 20 children can be assigned to that
block group. This apportionment method involves the assumption of an even distribution
of demographic and economic conditions across the county - a more realistic possibility,
compared to assuming an even spatial distribution of people. An example of population
reapportionment by small reporting units located within two HUCs is given in Figure 4.
Final Draft
13
-------
Figure 4. Example of apportioning population data for small reporting units (represented by
squares) for an urban area (shaded area) that occurs in two HUCs. Values shown in the
counties represent percentage of the urban area's population; thus, they sum to 100.
Continuing with the Mid-Atlantic ReVA example, block groups were then intersected
with HUCs (although some other reporting unit could be used, as noted previously) and
values were apportioned using the area-weighted method described above. In this
process, using the much smaller block groups, rather than counties, was expected to
mitigate much of the error introduced by the assumption of even spatial distribution.
Values from each of the block groups in the HUC, partial or whole, were then summed to
determine the overall estimated value of the 1990 population total for the HUC.
Missing Data
Missing data can be estimated using various interpolation or extrapolation methods.
These techniques are meant to be used only with continuous data, such as elevation; they
are not appropriate for categorical data, such as land cover.
Extrapolation is the process of using known data to predict values for areas or times
beyond the spatial or temporal extent of the known data (Figure 5). An important
assumption of extrapolation is that observed patterns or trends are consistent in space and
time. Therefore, extrapolation is usually more reliable over short distances or time
14
Final Draft
-------
intervals. This method of estimating missing data becomes progressively more suspect
when applied to larger distances or longer time intervals.
Interpolation is the process of estimating values between two or more known values
(Figure 5). As with extrapolation, data can be interpolated over time and space. Linear
interpolation is the most straightforward method of estimating values, but other functions
can be used for interpolation. Common spatial interpolation methods include Inverse
Distance Weighting (IDW), splining, trend surface analysis, and kriging.
Extrapolated
point
Interpolated
point
Figure 5. Graph (hypothetical X and Y axes) showing measured data (solid circles), an interpolated
point, and an extrapolated point.
Kriging is a method of spatial interpolation that minimizes the variance of estimation
error. It is a linear, unbiased, least-squares method that uses spatial covariance to help
estimate values at locations that have not been sampled. Kriging is often used with point
data, such as air quality samples, to create a surface map where every cell has a value.
An excellent reference on kriging is Cressie (1993).
Final Draft 15
-------
Metric Selection and Preparation for Integration
After datasets for individual variables are assembled and documented, the variables must
be examined carefully for relevance, consistency, and interdependence. Then comes the
hard part: variables (or metrics), or combinations of variables, must be selected for use in
integration. In an earlier document (Smith et al., 2003), we reviewed 11 methods for
integrating metrics into indices for use in ReVA. That review provides details on how
each of the 11 indices are calculated and it provides discussion on each method's
advantages and disadvantages. Our objective in this section is more basic: we note that
while some ReVA metrics are developed from models (described in more detail in
Section 4 below), others are simpler and can be calculated without the use of models.
Examples of metrics that do not depend on the use of models are percent of forest cover
and road density within the reporting unit. Percent forest cover simply involves
overlaying land cover on the reporting unit and dividing forest area by total area, then
expressing the proportion as a percentage. Road density is a similarly simple metric and
is estimated by overlaying roads on reporting units using standard GIS tools to determine
the sum of road length within each unit. In the Mid-Atlantic study, many of the
indicators related to land cover were generated using an Arc View extension called the
Analytical Tools Interface for Landscape Assessment (ATtlLA
http://www.epa.gov/nerlesdl/land-sci/attila/index.htm).3
As a first step in preparing metrics for ReVA, it is important to examine the relevance of
each variable to the assessment being performed (Smith et al., 2003). Expert judgment
often is required in this process. One might decide, for example, to include both total fish
biomass and biomass of a fish species known to be highly sensitive to water pollution.
One might decide to include numbers of people employed in forestry as a resource
variable vulnerable to urbanization, and yet exclude numbers employed in financial
industries as marginally related to the current assessment. Including variables that are not
of immediate relevance can bias the integrated estimates of vulnerability across the
region.
The second step in index development is to examine the frequency distribution of each
variable across the region. The examination can reveal outlier data that need to be
explained. In the Mid-Atlantic dataset, for example, several watersheds had values for
sedimentation that were nearly an order of magnitude greater than elsewhere. Close
inspection revealed that the high sedimentation values had been derived from an
independent study that had estimated sedimentation values using a linear regression
between landscape variables and sedimentation. The watersheds that had unusually large
sedimentation values had landscape values that were well outside the range used in the
original regression model. No other method for modeling sedimentation was available so
it was necessary to eliminate the sedimentation variable from the dataset. In other types
of study, it might be acceptable to simply truncate the frequency distribution and
eliminate the outliers. In the ReVA approach, eliminating the outlier values means
3ATtILA is a free software application developed by EPA's Landscape Sciences Program. It is used to
calculate many of the landscape metrics used in ReVA-type assessments and can be applied to any type of
land-use/land-cover data (i.e., any scale).
16 Final Draft
-------
eliminating those watersheds from any further analysis, because there must be a value for
every variable that is used, for every watershed.
Because many of the integration methods in ReVA are statistically based, it is necessary
to examine the frequency distributions for discrete variables that can violate the
assumptions of the statistical methods. Discrete variables may enter environmental data
sets because presence/absence data are common. The result of presence/absence data is a
frequency distribution with peaks at 0 and 1, and no intermediate values. In the original
Mid-Atlantic dataset, presence/absence data were available for many individual species.
This problem was solved by aggregating the presence/absence data into continuous
variables representing the number of terrestrial and aquatic species within a watershed.
Sometimes it is not feasible to aggregate variables to overcome the problem of
presence/absence data. Then, the variable should be eliminated before using ReVA
integration methods to prepare regional estimates of vulnerability. The variable can still
be retained for some specific analyses, such as mapping regional patterns of presence and
absence.
The third step in index development is to examine the candidate variables for
mathematical dependence. Mathematical dependence means that some variable "X" is
simply a mathematical combination of other variables. For example, one cannot include
native forest acres, nonnative forest acres, and total forest acres in an index, because the
third variable, total forest acres, is simply the sum of the other two. This type of problem
is solved by eliminating any one of the three variables.
Many of the integration methods in the ReVA approach assume that the variables are
statistically independent. Therefore, the fourth step of index development is to examine
the dataset of variables for statistical interdependences. The simplest way to do this is to
search the variance-covariance matrix for all variables across all watersheds for unusually
high correlations.
In a variance-covariance matrix, high covariance values (i.e., those near 1.0) may indicate
that the two variables are essentially measurements of the same stressor or resource. For
example, "number of families below the poverty line" and "low annual household
income" are two very similar measures of a social group that might be vulnerable to
environmental degradation. One of the variables should be dropped from the dataset,
carefully choosing and retaining the variable that is more relevant to the current
assessment objective. As a rule of thumb, variable pairs that have covariance values
above -0.95 may need to be considered closely for the possibility of inappropriate
redundancy.
In other cases, high values of covariance may represent subtle mathematical
dependencies. Two variables which logically appear to be independent may actually be
mathematically related. This can occur, for example, with landscape cover metrics
attempting to measure contiguous habitat on the watershed. High values of covariance
between calculated values in this case may indicate that the different equations have
Final Draft 17
-------
converted to the same measure of contagion, at least on the watersheds under
consideration. If this is discovered to be the case, one of the variables should be dropped
from the dataset.
In considering the covariance matrix, high values of covariance between two stressor
variables may mean that the two are measures of the same underlying stress on the
ecological system. In this case, one of the two variables should be eliminated, as noted
previously. However, the high values for covariance between two stressor variables also
may be due to other factors. Stressors such as air pollution and water pollution may co-
occur in watersheds, even though the two Stressors originate from different sources and
have independent mechanisms. The co-occurrence of two independent Stressors in this
case means that the stress on the ecological system is significantly increased. Thus, both
Stressors are appropriate for assessing environmental vulnerability in this case and both
variables should be retained.
Significant correlations also can occur between a resource variable (such as biodiversity)
and a stressor variable (such as forest fragmentation). When these correlations result
from an underlying cause-effect relationship, both variables should be retained in the
dataset.
To facilitate combining the variables into integrated measures of condition and
vulnerability, the data are normalized. Normalization is used to ensure that all variables
have the same numerical range and can thus be compared. In the Mid-Atlantic dataset,
for example, a range of 0 to 1.0 was chosen, where 0 represents the "best" value of a
variable across the region and 1 represents the "worst" value. Having found the "worst"
value for a variable in the region, the values of that variable in the remaining watersheds
can be divided by the "worst" value to standardize the values between 0.0 and 1.0.
Variables also must be "direct!onalized" before integration. This ensures that all
variables that represent a negative or positive condition are aligned such that high values
mean the same thing, and that low values have the same meaning. For variables that are
clearly resources (a positive attribute with a normalized value tending toward 0) and
Stressors (a negative attribute, with normalized values tending toward 1), this is simple
and may require only a change in sign before normalization. However, for other
variables, such as socioeconomic data or other descriptive data, it may not be as clear
how to directionalize. In some cases, for example, a variable (e.g., population density
within the watershed) is considered as a stressor on the ecological system and thus is
normalized with a value that tends toward 1. In other cases, a variable (e.g., number of
threatened and endangered species) is considered a resource, but one that renders the
ecosystem more vulnerable to additional stress. In this case, the variable might be coded
such that its normalized variable tends toward 1. In previous ReVA applications, we
have generally evaluated if the value of the variable increases the overall sensitivity of
the reporting unit to additional stresses. If so, the variable is considered to move
condition and vulnerability in a negative direction, so directionalization should tend
toward 1. In short, careful judgment must be exercised in such cases and the direction of
18 Final Draft
-------
variable standardization may need to be adjusted depending upon the assessment
question.
The highest value of a stressor such as human population growth is considered to be the
"worst" value, and thus is assigned a value of 1.0. Conversely, the highest value of a
resource (such as native aquatic fauna) is considered to be the "best" value, and thus is
assigned a value of 0.0. This method of coding is advantageous in that it allows the
assessor to quickly evaluate all variables for a watershed. A watershed that has many
variables with scores near zero is considered to be in relatively good environmental
condition, because the low scores mean that resources are high and stressors are low.
The method of variable normalization and direct!onalization used in the ReVA Mid-
Atlantic dataset provides a relative estimate of "best" and "worst," because the limits are
chosen as the extremes within the region. This coding strategy has the advantage of
spreading the data across the extremes within the region, which simplifies the task of
distinguishing between watersheds. But the approach has drawbacks, too: the coding
strategy does not use objective criteria of "good" or "bad." The result is that watersheds
might be considered to be in relatively good condition within the region, even though all
of the watersheds within the region might be in poor condition if judged against an
objective standard. Unfortunately, the present state of knowledge does not allow
objective criteria to be developed for most variables, so the analyses are limited to
evaluating relative vulnerability.
A better method of standardizing the variables would be based on thresholds established
by statute or scientific study. Individual variables then could be coded by the extent to
which variable values were above or below the specified threshold. Thresholds may be
available as ecotoxicological ECx values (ECx refers to a concentration above which an
associated adverse effect occurs, for "X" percent of the individuals in a population).
Thresholds also may be based on an expert opinion or on a societal consensus as
expressed in statutes that limit human activities. An extensive literature review was
conducted in an attempt to find thresholds for the variables used in the Mid-Atlantic
region. This review revealed that thresholds existed for only a small percentage of the
variables and could not be used as the basis for standardization in this application.
A final factor that must be considered before integrating the variables into useable
indicators is whether there is an imbalance in the dataset between different factors
influencing vulnerability. For example, a dataset may have five measures of stressors on
the aquatic system (e.g., riparian vegetation, agriculture on steep slopes, inputs of
pesticides and herbicides, and roads crossing streams), but only one measure of the biotic
community (e.g., the number of native aquatic species). In this case, if one were to
calculate an integrated measure by summing the coded variables, one would be assigning
five times more weight to the stressors than to the single resource. To avoid an
imbalance between the stressors and the resource, one can average the five stressors, then
use this average to represent a single composite stressor.
Final Draft 19
-------
In general, the best approach to an imbalance between stressor and response variables is
to first categorize the dataset into groups of discrete factors, such as terrestrial stressors
and terrestrial resources. Then one can average within the groups before combining the
data to assign equal weights to the different factors. The need to balance and the exact
groupings needed to achieve balance is determined by the purpose(s) of the assessment.
If, for example, one wishes to determine which of the individual aquatic stressors is most
important, then averaging the several stressors would not be appropriate. Alternatively, if
one wants to assess the relative condition of watersheds across the region, then balancing
the dataset is generally appropriate. The easiest way to do this is to give equal weight to
each of the factors determining condition. The choice is a matter of judgment and the
answer may differ for different analyses done on the same regional dataset.
20 Final Draft
-------
Section 4
Spatially Explicit Models
Why Models?
Spatially explicit data are required to compare risk across a region (Hunsaker et al.,
1992). Typically, data for regional assessments include infrastructure (e.g., roads),
stressors (e.g., atmospheric deposition, chemical inputs), landscape features (e.g.,
geology, elevation, vegetative cover), sensitive resources (e.g., wetlands), and ecological
endpoints (e.g., avian biodiversity). Unfortunately, in many cases these data are not in a
format that can easily be incorporated into a regional analysis. For example, consistent
monitoring data for surface water and ground water are usually only available at
relatively fine scales and these data are unevenly distributed across a region. To use
these types of data, models are needed to estimate values for points where data are not
available. For this reason, models are an important part of the overall ReVA process.
However, it is critical to keep in mind that models are only tools to guide the researcher
to further inquiries about the nature of the system under evaluation. Models are an
abstraction or simplification of a more complex system: they are not truth but "the lie that
helps us see the truth" (Fagerstram, 1987).
Types of Models
Mathematical models translate our understanding about relationships (e.g., cause-effect
processes) into equations. Such models help reduce the vast quantities of available data
and facilitate the generation of useful hypotheses. Further, data which appear to be
"outliers" based on the model are more evident, which makes it easier to determine if the
outliers are really outliers or if the model needs to be adjusted. The two classes of
mathematical models that are most commonly used during any assessment are empirical
models and process models.
Empirical Models
Empirical models are used to examine the relationships between single and multiple
variables without incorporating the underlying mechanisms responsible for the
relationship. In ReVA, for example, statistical (empirical) models relate land cover to a
dependent variable, such as nutrient load, pollutant deposition, or bird migration. The
relationship between land cover and a dependent variable can be linear, exponential,
bimodal, or any of a large number of other forms; the relationship only needs to be
simple and statistically strong (assuming a large geographic area is used to capture a
broad range of variability). The simplified structure of an empirical model is both its
strength and its weakness. Empirical modeling allows an investigator to evaluate large
Final Draft 21
-------
quantities of data, but it does not provide information on the fundamental cause of the
observed relationships.
Process Models
Process models, as the name implies, include known processes or mechanisms in nature.
In the case of ReVA, a model such as AQUATOX could be used to predict changes in
biological and ecological endpoints such as the abundance of phytoplankton, the
abundance of game and bottom fish, the concentrations of nutrients and dissolved
oxygen, or even the percentage of organic matter in sediments in response to toxic
organic chemicals. The predicted conditions could then be expressed spatially and thus
be incorporated as changes in resources. The major drawback to process models is the
extensive effort needed to "fit" the model with reasonable values for its constituent
parameters (e.g., current populations, population growth rates, land use, pesticide
application rates, lake dimensions, etc.) which are needed to operate the model.
Examples of Spatial Models Used in ReVA
When doing an assessment at a broad regional or national scale, the lack of or uneven
distribution of monitoring sites often requires the development of spatial models for
filling in areas not covered. Several examples of the types of models and model output
ReVA has used to meet assessment needs are given below as "thumbnail" examples. The
models range from regression to Bayesian analyses to more complex combinations of
statistical and mathematical algorithms.
Nitrate and Sulfate Deposition Modeling
Nitrate and sulfate deposition estimates used by ReVA came from an empirical model
developed by Grimm and Lynch (2000). This model addressed the sparseness of the
National Atmospheric Deposition Program monitoring sites by using a multiquadric
equation developed by Hardy (1971) which provides the density needed for use in a
spatial-weighted linear least-squares regression algorithm. This model yields deposition
estimates as a function of latitude, longitude, elevation, slope, and topographic aspect.
The elevation, slope, and aspect parameters all are derived from USGS DEM datasets.
Mercury Deposition Modeling
Bayesian statistical methods were used to develop models which derive interpolated
maps of weekly mercury deposition. Data on monitored samples of mercury deposition
were supplied by the National Atmospheric Deposition Network - Mercury Deposition
Network (http://nadp.sws.uiuc.edu/mdn/). However, due to the small number of
monitoring sites available, additional data were needed. By including nitrate, sulfate, and
precipitation, all of which correlate with mercury, we were able to supplement the
amount of spatial information. Using these related sources of information, we developed
a space-time model that provided spatial predictions of nitrate, sulfate, mercury, and
precipitation, as well as their associated uncertainties, including spatial and temporal
22 Final Draft
-------
misalignment among the networks. Since the depositions of these constituents occur in
response to precipitation, we also modeled a spatial field of the probability of
precipitation for the area of interest. Depositions and probabilities of precipitation, in
turn, are jointly modeled through time-varying linear models of co-regionalization
(Banerjee et al., 2004). The mercury-deposition model therefore provides a constructive
specification of the cross-covariance function allowing for non-stationarity and
dependence among the fields.
Invasive Species Modeling
Genetic Algorithm for Rule-set Production (GARP) modeling was used to create spatial
maps of the potential distributions of invasive species within the region of interest (see
figure 6). GARP uses occurrence data for a species within its current range to predict the
species' likely distribution within the area of interest. Input data include spatial data on
species occurrence and environmental factors such as temperature, precipitation, solar
radiation, snow cover, and frost-free days. GARP uses multiple rule types including
BIOCLIM, logistic regressions, and a genetic algorithm (an artificial intelligence
application) to generate a set of "IF... THEN" rule statements that describe the
relationships between the species and the environmental conditions. The output from
GARP can then be projected onto a "new" landscape to visualize the species' potential
distribution. The distribution also can be projected onto areas of an actual or potential
invasion/introduction under different land cover and climatic conditions (Peterson et al.,
2003).
Figure 6. Graphic depicting 2000 distribution of Giant Salvinia (Salvinia molesta) and estimated 2020
distribution using the Genetic Algorithm for Rule-set Prediction (GARP) model.
Final Draft
23
-------
Nitrate in Ground Water Modeling
Data on ground water quality obtained from USGS National Water-Quality Assessment
Program (NAWQA) studies were used in association with geographic data to develop
logistic-regression equations to predict the probability of nitrate exceeding a specified
management concentration threshold (Greene et al., 2005). The geographic data included
land cover, soil permeability, soil organic matter, depth of soil layer, depth to water table,
clay content of the soil, silt content of the soil, and hydrologic groups within a specified
area of influence. The relationship of these factors with nitrate concentrations above a
threshold was determined using logistic regression. Since well data were not uniformly
distributed across the study area, the coefficients calculated from the significant
geographic features were used to create a surface map of the likelihood for exceeding
acceptable levels of nitrate across the study area (figure 7).
pUSGS
sciaace tors changing work)
PREDICTED PROBABILITY OF EXCEEDING 1.5mg/LGF NITRATE CONCENTRATION IN GROUND WATER.
RaVA
NAWQA SAMPLE SITES
PREDICTED PROBABILITY
Figure 7. Graphic showing NAWQA sample sites (map on left) and results of logistic regression
model that estimates probability of exceeding a threshold of nitrate concentration in
shallow ground water aquifers across the Mid-Atlantic region.
Forecasting Drivers of Change
ReVA relies on various models to evaluate current and future ecological conditions and
vulnerabilities at a broad spatial scale. At the regional scale, a number of drivers of
24
Final Draft
-------
ecological change operate to effect changes that are observable at this broad scale and
over a temporal scale that may span decades. Changes may not be observable at a local
scale until they reach some threshold, yet may have irreversible consequences if not
anticipated and addressed strategically. Regional-scale drivers of change include the
following:
• Land-use change
• Resource extraction (e.g., over-fishing, timber harvest, mining)
• Changes in pollutants (e.g., nonpoint source pollution, agricultural runoff) and
pollution (e.g., changes in atmospheric deposition)
• Spread of invasive, non-indigenous species (e.g., pests and pathogens, introduced
species)
• Climate change (e.g., changes in weather patterns)
Of these drivers of change, land use is probably the most important, because land cover
and land-use pattern affect every other driver of change. Thus, land use is often one of
the most important parameters in any model that estimates a future distribution of
stressors related to resource extraction, changes in pollutants and pollution, the spread of
invasive species, and even changes in local weather patterns.
Resource extraction often follows development, as roads are constructed to access remote
areas where resources have not yet been exploited. Models of nonpoint source or
agricultural runoff specifically include land use as input parameters; these can represent
the amount of chemical applications for farmland and sediment loading in areas that lack
riparian buffers. Models of air deposition include mobile-source estimates, as well; thus,
the pattern of road networks has implications for regional air quality. Many invasive
non-indigenous species are transported by people and the spread of such organisms is
generally facilitated by transportation networks. Similarly, land cover provides habitat
for invasive species, which relates to the range of their spread. And finally, regional-
scale models of climate change can include land-use/land-cover information as inputs,
because local weather patterns are influenced strongly by surface roughness and
reflectance, in addition to shading afforded by vegetative cover.
Land-Use Change Models as a Component of ReVA
Land-use change models are an important component of ReVA. A particular problem is
posed by projecting land-use changes caused by population growth - that of
apportionment. For example, population growth can result in conversion of land to
residential and agricultural uses (Wheeler et al., 1998). Distributing these changes
spatially is critical to projecting changes in stressors such as aquatic nonpoint source
pollution (e.g., percent impervious surface or agriculture on steep slopes) and forest
productivity. Land-use changes also can directly alter estimates of resources (e.g.,
wildlife habitat, wetlands, etc.). To identify the most appropriate model for forecasting
land-use change in the Mid-Atlantic region, ReVA reviewed and evaluated several land-
use change models (Wagner et al., 2006). These models ranged from simply
documenting plans for highway construction and new employment centers to estimating
Final Draft 25
-------
land demand from state census projections to customizing applications of a traditional
resource economics model (Hardie and Parks, 1997) to a state-of-the-art cellular model of
urban growth (Clarke et al., 1997).
In the Mid-Atlantic region, ReVA chose to use output from the Slope, Land use,
Exclusion, Urban, Transportation, Hillshading (SLEUTH) model in combination with
other sets of land-use data (figure 8). SLEUTH uses a cellular automata simulation
approach (Clarke et al., 1997) to illustrate future urbanization based on historic patterns
of land transition. We chose SLEUTH because it distributes change spatially, employs
more complex rules than those of a typical cellular automata simulation method, and uses
numerous data sources (including topography, road networks, and settlement
distributions) to accumulate probabilistic estimates based on Monte Carlo methods
(Jackson et al., 2004). It is, however, an urban growth model and thus may not
effectively represent regional land-use change processes, such as changes in rural land
use or the creation of new urban centers.
For forecasting, SLEUTH assigns each 1-km cell a probability of being developed in any
given time frame. We chose 50% as the threshold and created a binary map of
developed/not developed in 2020. SLEUTH does not address any other land-cover
changes (e.g., conversion of forest to agriculture).
The following steps were used to create the final future land-use map for use in regional
analysis.
1. Begin with NLCD 1992 (30-meter resolution).
2. Add new urban areas, based on outputs of the SLEUTH model. SLEUTH
produced 1-km raster output of projected areas of urban growth. Areas predicted
to have a 50% or greater probability of being developed were "burned in" as
urban cover on the NLCD map.
3. Planned roads and road expansions were overlaid with NLCD and "burned in" as
new developed cover in 2020.
4. Mining permits were obtained from Virginia, West Virginia, and Pennsylvania.
All permitted areas were assumed to be mined in 2020; each of these areas was
also "burned in" as mined area in 2020.
5. Areas where mines and urban were coincident were coded as "mines."
6. Areas that did not coincide with new urban area, roads, or mining retained their
1992 land cover status.
26 Final Draft
-------
Figure 8. Graphic depicting current land use/land cover based on the National Land Cover Database
(NLCD) 1992 (left) and estimated 2020 land use/land cover using the Slope, Land use,
Exclusion, Urban, Transportation, Hillshading model (SLEUTH) in combination with
planned roads and permitted mines (right).
Models that Use Land Cover/Land Use as Input
As future scenarios are developed, the challenge is to translate the projected scenarios
into spatial changes in stressors and resources. In most cases, the changes can be
extrapolated using the same models that are used in assessing current conditions. Since
population growth and urbanization results in changes to land-use proportions, it is
simply a matter of applying the model to the new land cover.
Resource Extraction
Many ecological resources are considered vulnerable, yet the use of these resources
provides society with valued goods, services, and other benefits. Such benefits can
involve resource extraction (e.g., forests and minerals), recreation (e.g., hiking and
fishing), waste treatment, and nutrient recycling. Vulnerable ecological resources in this
category are critical because damage to them can impact society immediately. In the
East, which was the focus of our initial ReVA studies, forests are one of the largest
resources of concern. Forests provide numerous goods and services, including recreation,
economic timber harvest, and species habitat. Forests are vulnerable to urban growth,
fragmentation caused by timber harvests and accompanying roads, and introduction of
exotic pests and pathogens. Mineral extraction is a driver of ecological change largely
due to the impacts to other resources (e.g., water quality, habitat, etc.).
The USDA Forest Service's Forest Inventory and Analysis data were used to estimate
current and future forest conditions at the watershed scale (Schaberg and Abt, 2004). A
timber economic forecasting model (Subregional Timber Supply Model; see Prestemon
and Abt, 2002) was used to project trends in timber harvest and forest sustainability into
Final Draft
27
-------
the future. This model included information on projected land-use change, because we
expected that much of the timber resource extraction would follow new developments.
As mentioned in the section under land use, we used available state mining to predict
where mining would likely occur in the future. This was reflected as a change in land use
for our future land-use/land-cover map.
Pollutants/Changes in Water Quality
The susceptibility of a landscape to erosion is estimated by semi-empirical models such
as the Universal Soil Loss Equation (USLE). The USLE has been widely used to
estimate average annual soil loss (mass per unit area) according to known erosion
mechanisms: rainfall, soil type, slope, vegetative cover, and agricultural management
practices. By incorporating the new values developed for future land cover, it is possible
to get an estimate of potential future erosion. Modifications of the USLE, such as the
RUSLE (Revised USLE), make use of meteorological data to estimate soil erosion with
temporal responses for specific time periods or rainfall events. Although many soil
transport mechanisms have been characterized for small watersheds, existing models for
estimating sediment delivery to surface waters require an extensive local calibration. As
a result, these models lack utility at scales suitable for regional assessments.
Nutrient loading models such as the LTHIA (Long-Term Hydrologic Impact Assessment;
Harbor and Grove, 1997), Reckhow's (1980) model (which was incorporated into the
Arc View extension, ATtlLA), or land-cover-based regression models from pour-point
samples can be modified to use forecasted land-use change data. With LTHIA, the land-
cover grid or the percentages of the land-use types can be used, depending on the model
(Pandey et al., 2000). For the Reckhow model and other statistically based models, the
percentages of each land-use type are used in conjunction with a set of land-use
coefficients to calculate the overall nutrient load for a watershed. However, the ATtlLA
extension will convert percentage values and allow the user to set coefficients based on
regional knowledge (U.S. EPA, 2004).
Spread of Invasive Non-Indigenous Species
Future scenarios of invasive non-indigenous species also were created using the projected
land-use/land-cover map as input. GARP modeling (see section on examples of spatial
models) was used to estimate the future distributions of several species of concern. The
projected distributions were based on habitat requirements, which includes land cover
and land use.
Models that Do Not Include Land Use/Land Cover as Input
Air pollution modeling is complex and requires multiple layers of data, including
estimates of emissions from stationary and mobile sources. These data are needed to
predict pollutant loadings to the landscape from the atmosphere. Land use and land cover
are not extremely important in this case and these conditions generally are included at the
regional traffic-demand modeling phase for mobile source emissions. Urban growth also
28 Final Draft
-------
may be considered when estimating increased numbers or demand from stationary
sources (i.e., Energy Generating Units). However, other factors, such as topography and
meteorology, are very important factors in predicting air pollution. Two regional
pollutant datasets used by ReVA include the National Air Toxics Assessment (NATA;
see http://www.epa.gov/ttn/atw/O and the Community Multiscale Air Quality model
(CMAQ; see http://www.epa.gov/asmdnerl/CMAQ/).
Final Draft 29
-------
30
Final Draft
-------
Section 5
Creating Alternative Future Scenarios
Alternative scenarios are not intended to predict the future. Rather, they present a series
of plausible future states that are likely to include: (1) a mixture of modeled projections
of current trajectories and/or prospective forecasts, (2) alternative policy and/or
management options that will likely affect ecological goods and services, and (3) various
spatial and monitoring data that are used to describe both the baseline and the alternative
scenarios in terms of landscape characteristics and associated ecosystem services.
General guidelines for creating scenarios are provided by Liu et al. (2007), Pandey et al.
(2000), and Weingand (1995). These guidelines suggest the following: (1) scenarios that
include "best-case," "worst-case," and a "most-likely" case are both useful and
informative, (2) scenarios should be distinct, or at least different enough to discern
changes over space and time, (3) scenarios should explore the bounds of what is feasible,
and (4) scenario creation should have a clear focus, purpose, or direction, thereby
ensuring that the number of scenarios created, analyzed, and assessed is kept to a
minimum (fewer are better than many). ReVA follows these guidelines in developing
scenarios and recommends ReVA users to do the same.
Building Alternative Scenarios for Analyzing Future Trends
Alternative future scenarios can be prepared either by creating a set of static future
scenarios for a specified time in the future or by projecting trends in a series of time steps
until a specific time period has been accommodated. Either method must include some
projections of past trends (e.g., population growth, corn prices, etc.) that the ReVA user
may want to combine with conditions that differ significantly from those that have been
observed in the past.
Scale Considerations in Projective and Prospective Modeling
Regardless of the mix of protective (extension of past trends) and prospective
(significantly different from past trends) modeling, the appropriate scale of the model and
the geographic extent of the region being modeled must be considered. This is
particularly true of land-use change models, which in the past have been developed to
represent urban growth trends, but only rarely have captured regional growth processes
which include the development of new urban centers and rural to exurban land
conversions.
Final Draft 31
-------
Spatial Scale
A good reference for local-scale growth models is U.S. EPA (2000). Examples of
regional land-use change models include the Resource Economics Model (Parks et al.,
2000) (this model is not spatially explicit), the Spatially Explicit Regional Growth Model
(SERGoM; see Theobald, 2005), and the Integrated Climate Land Use Scenarios
(ICLUS) project (e-mail from Britta Bierwagen, U.S. EPA Global Climate Change
Program, to Elizabeth Smith, U.S. EPA National Exposure Laboratory, dated May 2007),
which is being developed by the Global Change Program.
Various spatial scale issues are associated with climate change. If climate change is
included as part of the scenario creation, for example, then a prospective model of
projected changes in weather patterns at a suitable scale (regional or subregional) will be
more appropriate than one at a national or broader scale.
Temporal Scale
Alternative future scenarios can be created in ReVA either by projecting conditions for
multiple time steps that build on one another or by creating a future scenario independent
of intermediate conditions that is based on some vision of the future. The environmental
decision-maker should consider the type of end product or decision tool that is envisioned
and how the future scenario will feed into that product or tool. For example, if the goal is
to display changes over time in response to user input, then fine-scaled time steps for the
forecast models may be needed to create dynamic responses. These short time steps
might also be needed to represent processes important for assessing changes in ecosystem
services. Alternatively, if the objective is to compare a suite of discrete scenarios that
cannot be altered by the user, then coarser time steps (e.g., decadal, as in ICLUS) may
suffice. Generally, for large geographic regions, fine-scale temporal detail is not feasible
because their inclusion can greatly increase the need for computational resources.
Temporal detail also may be less critical at broad spatial scales because changes in
ecological services generally take time to become evident. In other words, resolution will
be coarser when the goal is to represent the broad spatial scale.
Anticipating Responses to Policies
Beyond projecting change by continuing a current trend, alternative scenarios are used to
explore futures that involve trade-offs of ecosystem services through alternative
decisions, policy levers, or incentives. Determining what these policies are likely to
"look like" can be done by obtaining input from experts (i.e., from EPA Program
Offices). Alternatively, a forward-looking estimate of future policies can be citizen-
driven. That is, planners and developers can provide information on what commercial
and residential densities are feasible for certain sections of an urban area, or a group of
stakeholders could envision a future they would like to see.
Five types of issues are associated with incorporating input from experts or citizens/
stakeholders: (1) future scenarios with too many details (clients who want a "perfect"
prospective future - really more of a prediction), (2) too many scenarios (trying to please
32 Final Draft
-------
everyone), (3) bias (i.e., listening to only a few sources of input), (4) plausibility, and (5)
trouble converting the input into a spatial model. The first four issues can be dealt with
by managing expectations, working iteratively with stakeholders, and clearly conveying
the capabilities of the regional approach. Approaches to the fifth issue are discussed in
the next section.
Using Other Spatial or Monitoring Data to "Spatialize" Scenarios
Alternative future scenarios must be spatially explicit to effectively represent effects on
ecosystem services and the trade-offs associated among the alternative scenarios.
However, not all of the information used to create the alternative future scenarios will be
in this format. Therefore, available spatial data and GIS decision rules are used to
"spatialize" nonspatially explicit model results and other features of the scenarios, such
as possible policy alternatives. Here is an example of why "spatialization" is needed, and
how it can be accomplished. In the Future Midwestern Landscapes project, ReVA uses
an economic projection of crop plantings (acreages) based on prices for corn and other
crops (dollars per acre). This projection is used to determine how much land is planted as
feedstock for ethanol (used as a biofuel). The results of this model must be expressed
spatially, using information such as SSURGO soil data, National Agricultural Statistical
Survey (NASS) crop data, and spatial representations of tillage practices, streams, roads,
protected areas, etc. For policy alternatives, GIS decision rules are needed to develop the
models to reflect these alternatives.
It is possible that additional point or monitoring data (e.g., air deposition data) may be
used to create baseline or alternative landscapes. To do this, it is necessary to create a
surface from these points, using some form of extrapolation, interpolation, or other type
of model that predicts conditions at specific locations, such as spatial statistical
approaches.
Final Draft 33
-------
34
Final Draft
-------
Section 6
Synthesis
Once an extensive dataset that covers many aspects of environmental quality and
vulnerability is assembled, it is necessary to synthesize or integrate the information. If
the information is not integrated, it is difficult to evaluate the overall effectiveness of
environmental policy. For example, restoration of riparian vegetation may improve
stream water quality, but if agriculture on steep slopes, roads crossing streams, wetland
loss, and urbanization are extensive on the watershed, then planting trees along the
stream bank may not improve in-stream water quality.
Combining individual variables into an integrated indicator is inevitably controversial
(Andreasen et al., 2001). Researchers, for example, may have a sophisticated
understanding of the interplay of environmental variables and frequently will disagree on
the overall impact of these variables. As a result, the scientific community rarely is
content with a single integrated value; they generally will want to examine the original
data and debate the implications. Few decision-makers, on the other hand, possess the
scientists' sophistication in data interpretation, but decisions must be made and limited
resources must be allocated, even if scientists disagree. Similarly, stakeholders will want
to know if actions taken in the past have actually made the environment "better," and
they may not agree as to which aspects of the environment should be prioritized. So the
question is not whether or not to synthesize and integrate the data - instead, the challenge
is to develop and test innovative approaches to integration.
Available Information and Data Preparation
An important limitation on the ReVA approach is the quality and extent of the available
data. In the case studies that have been examined by ReVA to date, information has been
limited to variables measured in other programs (Smith et al., 2004). Indeed, one of the
original motivations for the ReVA program was to synthesize the multiple physical,
chemical, and biological datasets being gathered by disparate programs within EPA.
Resources do not exist within the ReVA program to perform field measurements. Thus,
the analyses and the conclusions drawn from the analyses are limited by the available
data. In the Mid-Atlantic study, for example, adequate information was available on
remotely sensed land cover, but relatively little information was available on biodiversity
across the region. Data on the numbers of native, non-indigenous, and threatened and
endangered species were only available for relatively small number of taxa. Therefore,
the study could not examine or represent important aspects of biodiversity.
In some cases, variables can be calculated using models developed by other researchers.
Examples range from well-studied air quality models for nitrate and sulfate deposition to
regression models relating watershed land cover to stream water quality (Jones et al.,
Final Draft 35
-------
2001). To date, resources have not existed within ReVA to develop complex simulation
models requiring testing and validation, such as exposure models or dose-response
models. Instead, the program relies on testing and validation operations performed by the
originators of the models.
To apply the ReVA methodology, it is important to understand the relationship between
the data and the objectives of the analysis. In applications such as the Mid-Atlantic
study, the objective was to assemble the extensive available data, place the data into a
regional spatial framework, and explore the possibilities of integrating the data to locate
potentially vulnerable watersheds. The study did not begin with a problem; it began with
the objective of locating spatial patterns of environmental quality and identifying
potential problem areas that might not be identified by other methodologies. Other
applications may well begin with more specific objectives, such as relating spatial
patterns of development to air and water quality.
When the study involves a specific goal, the objective will determine the data needed. In
some cases, assessment questions may involve smaller scales such as a single 8-digit
HUC. However, the more common case will be that the data to address these questions
simply do not exist. If data needed to address the assessment questions do not exist, the
ReVA methodology cannot be used to address the assessment questions.
In many assessment projects, available data and models have been supplemented by the
use of expert opinion. Expert opinion is often qualitative or, at best, can be described by
principles of Fuzzy Logic (Klir and Bo, 1995). This presents major but surmountable
problems for integrating expert opinion with measured data or modeled variables. Often,
expert opinion is the only available option for supplying information required for a
specific assessment. ReVA will likely be using expert opinion in future projects and the
use of expert opinion will require developing the appropriate analytical tools for its
integration.
Methods for Integrating Variables
Simple Sum and PCA Sum
The simplest method for integrating the variables is to sum their normalized values. This
is referred to as the Simple Sum. Because the summation method contains no prior
assumptions about relative importance of the variables, this approach is easily
understood. The purpose is to provide an overview of the spatial pattern of
environmental quality by combining stressors, resources, and socioeconomic factors.
Because the Simple Sum does not account for the correlation structure of the regional
dataset, ReVA also developed an integrating method referred to as the Principle
Components Analysis (PCA) Sum. The PCA Sum method accounts for correlations by
weighting variables by principal components. Details of the method can be found in
Smith et al. (2004).
36 Final Draft
-------
After some experience with using the Simple Sum and PCA Sum methods, we
recommend that these two methods always be used together. The PCA Sum method
removes potential bias if many stressors co-occur in space. On the other hand, the
Simple Sum method might be more useful if the co-occurring stressors act
synergistically.
The two summation methods are visualization tools that allow one to see all of the spatial
patterns of all of the variables in a holistic or synthetic manner. Because these methods
provide such a generalized picture, they should not be used for providing answers to
assessment questions. Rather, they should be used simply to visualize the spatial pattern
of potential environmental problems across a region.
A potential problem that arises with the two summation methods is an imbalance. This
can occur, for example, if one has many measures of water quality and only one measure
of land-use change. The resulting sum is heavily weighted toward the aquatic. To avoid
this problem, one can average within categories (i.e., aquatic or terrestrial) and sum the
averages.
Best and Worst Quintiles
To calculate the Best and Worst Quintiles, variables are ranked and subdivided into
quantiles with the same number of watersheds. Each watershed is then evaluated in terms
of the number of its scores that fall in the best and worst quantiles. Watersheds are then
depicted in quantiles again, based on these counts. This method must be used with
caution as it is not a very sophisticated analytical technique. It can, however, be used to
highlight where favorable and unfavorable conditions tend to cluster within the region.
A Monte Carlo uncertainty analysis was performed for this approach using the Region 3
dataset (Tran et al., 2007a). The results of this analysis showed that data errors had little
impact on which watersheds appeared in the "Best Quintile" (i.e., the 20% of watersheds
in the best ecological condition) or the "Worst Quintile" (i.e., the 20% of watersheds in
the worst ecological condition). Watersheds in intermediate positions often shifted
quintiles when error was randomly applied across the variables. We concluded that the
Best and Worst Quintiles could be reliably estimated, but there was significant
uncertainty about the positions of intermediate watersheds.
The explanation for this uncertainty pattern appears to reside in the regional dataset.
Watersheds in the Best Quintile tended to be mountainous and inaccessible. In these
watersheds, the resources were abundant and there were few human stressors, so all
variables tended to have values near the "good" end of the spectrum. Random errors
changed the value of the individual variables but did not change the sum, because all
variables indicated good ecological condition. Conversely, the watersheds in the Worst
Quintile tended to be urban; they had relatively few natural resources and numerous, co-
occurring human stresses. Therefore, these watersheds tended to remain in the Worst
Quintile even when random error was introduced.
Final Draft 37
-------
State Space Method
The State Space Method measures the distance between two points (i.e., two watersheds)
in multivariate space. The distance measure used by ReVA (Tran et al., 2006) avoids the
potential bias in distance measures such as the Mahalanobis distance (see De
Maesschalck et al., 2000).
The State Space Method is very versatile and can be used for various assessment
applications. It can be used, for example, to measure the overall distance of each
watershed in the region from a reference point. The reference point might be a nearly
pristine area, such as a national park, in which case the distance is a measure of
degradation from this pristine state. Alternatively, the reference point might be the most
vulnerable watershed in the region, in which case the distance is a measure of resilience.
In the current implementation, the user can choose the reference point and see how far the
other watersheds deviate from this reference.
The State Space Method is particularly valuable in analyzing the results of scenario
studies. In scenario studies, additional stressors, such as climate change or invading
species, are imposed on the region. Conversely, the scenario may be designed to evaluate
the regional improvement resulting from particular mitigation or restoration activities.
The distance measure then indicates the degree of degradation or improvement on each
watershed. This multivariate analysis is necessary because, for example, restoration
activities alone may have little impact on overall quality in the region if all other stressors
remain or worsen.
Criticality Analysis
Criticality Analysis is similar to the State Space method in that it measures distance from
a reference state. But in this case, the reference state is a postulated prehuman or totally
nondisturbed state. The logic is that this measures how far an ecological system has been
disturbed away from the state under which it evolved. The greater this distance is, the
greater the probability that the system will pass a critical stability point and change to a
new state. The theoretical justification for this idea can be found in Smith et al. (2004).
Because Criticality Analysis measures a distance in multivariate space, any of several
distance measures could be used. In the Region 3 study (Smith et al., 2004), ReVA used
a fuzzy distance measure (Tran and Duckstein, 2002). This measure was chosen because
the pre-disturbance state could not be defined with precision, so we estimated the pre-
disturbance distributions of variables using fuzzy logic. While this choice seems
reasonable, other measures of distance might be chosen in future applications.
Overlay Method
The Overlay Method attempts to identify watersheds where important resources still exist
but the remaining resources are under significant stress. Such watersheds are vulnerable
in the sense that further stress, e.g., from additional development, could result in the loss
38 Final Draft
-------
of valued resources. Thus, the Overlay Method provides a direct measure of regional
vulnerability.
The Overlay Method first divides a dataset into stressors and resources. In the current
implementation, the coded variables are summed within stressor and resource classes.
The method then classifies watersheds by comparing the number of resources with the
number of stressors. When resources and stressors are both high, the watershed is likely
to be highly vulnerable.
Stressor-Resource Matrix
In any regional analysis in which mitigation is a potential policy option, there is a need to
identify the stressor(s) having the greatest impact on the valued resources and to identify
the resources that are most intensively stressed. This has led the Society of
Environmental Toxicology and Chemistry (SETAC) to develop a matrix methodology
that uses expert opinion to identify the greatest stressor (Foran and Ferenc, 1999; Ferenc
and Foran, 2000). The ReVA approach permits an explicit analysis based on regional
data rather than on expert opinion.
The ReVA method constructs a matrix that blocks stressors and resources and connects
the blocks with a vector. By raising the matrix to a large power, the influence of all
stressors on all resources is captured in the vector. Mathematical details can be found in
Tran et al. (2007). The largest vector element then indicates the most influential stressor.
A similar matrix can be constructed to determine the resource receiving the greatest
stress.
Moving to Smaller Scales
While the Integration Methods are general and not limited to any specific scale, the
ReVA methodology was designed for regional assessments and it is recommended that it
be used only for that scale. The problem with applying the ReVA approach to smaller
scales lies with the data. In general, the information available across the region cannot be
directly applied to smaller scaled problems. For example, monitoring stations scattered
across a region can be reasonably averaged upward to provide estimates at larger-scale
watersheds. However, choosing smaller watersheds would result in missing data. The
missing data can be supplied by spatial interpolation, but interpolation assumes that the
monitoring stations adequately represented maxima, minima, and spatial trends - a
condition that is rarely or ever the case. The result is that using interpolation to scale to a
finer resolution typically introduces far greater error than averaging up to larger scales.
The greatest power of the ReVA methodology lies in assessing spatial patterns of
vulnerability across large regions. Over large regions, remotely sensed land-use data,
GIS technology, and advances in landscape ecology provide a powerful means for
combining and analyzing spatial information. At larger scales (across topographic
gradients, soil types, ecoregions, and human development patterns), the spatial pattern
Final Draft 39
-------
and the differences among subregions can most clearly be shown and analyzed. At
smaller scales (such as a single state or about twenty 8-digit HUCs), the patterns can be
less interpretable and the statistical power of the integration methodology is diminished.
Uncertainty in ReVA Analyses
All measurements have associated uncertainty, referred to as measurement error. At least
two sources of measurement error are important to ReVA. First, the value assigned to the
spatial unit has some associated uncertainty. Second, even if the metric value is known
perfectly, there is uncertainty associated with the impact of the stressor on a response
variable within that spatial unit. These two types of uncertainties are discussed in
Wickhametal. (1997).
If the greatest strength of ReVA lies in integrating available data, its greatest danger of
misapplication lies in assuming that the available data are sufficient. Implicit in the
ReVA methodology is the possibility of false negatives. That is, based on available data,
a given watershed may appear to be in reasonable ecological condition and not vulnerable
to further stresses. In this assessment scenario, the watershed would not be given high
priority for managerial action. However, the watershed may in fact be highly vulnerable
due to stressors that are unknown at the time of analysis. For example, illegal dumping
of toxic material or undetected leakage of raw sewage may be occurring. Such factors
could make the watershed highly vulnerable to ecological damage, but may not be
incorporated into the ReVA analysis.
Then again, it is unlikely that the ReVA methodology would produce many false
positives. A watershed is identified as vulnerable in ReVA because it is known to
contain important ecological resources and is known to be subject to multiple factors that
stress the resources. While it is conceivable that the ecological system is uniquely
resistant and resilient at this location, this possibility is unlikely. Therefore, when the
methodology identifies a watershed as vulnerable, it is reasonable to assume that
managerial action is called for - or, more conservatively, that the responsible officials
need to examine these watersheds more closely.
40 Final Draft
-------
Section 7
Results Communication
Audience and Assessment Needs
How the results of any assessment are communicated depends largely on the intended
audience and their specific assessment needs. ReVA's analyses are designed to be of
value primarily for decision-makers, rather than stakeholders in general or for the general
public, but the assessment information can be extended for environmental outreach.
However, any kind of environmental outreach would require substantial work to interpret
results in lay terms, and this goes well beyond the scope of the guidance offered here.
Providing ReVA results as outreach information requires both a thorough understanding
of the ecological data and results, and good communication skills to translate the
scientific results into information that is readily understandable by nonscientists.
Within the broad category of decision-makers, different levels of detail are needed.
These can vary depending upon the type of assessment question that is being asked and
the expertise of the decision-maker that needs the information. Compare, for example,
the needs of a U.S. EPA Deputy Regional Administrator, who must determine which
division within the region should have the largest share of discretionary funds based on
critical issues, versus a Water Division Director, who must determine if funding should
go toward restoration efforts in one watershed or toward establishing partnerships with a
local community to promote smart growth practices in another watershed. The higher-
level decision-maker (in the current example, the Deputy Regional Administrator) may
not need detailed information on individual endpoints such as water quality or future
vulnerability of aquatic biota. Rather, his/her needs could include a review of all
endpoints, using an index that represents the current conditions across the region.
Similarly, the Water Division Director may have little interest in endpoints other than
those specific to water. A specific example of how ReVA has addressed these
differences in needs is provided in the Regional Growth Decision Tool (RGDT) that was
created for the Sustainable Environment for Quality of Life (SEQL) project (Figure 9).
This toolkit provides options for three levels of users, each of which have different
assessment needs. The level of detail of information is reflected in the types of indices
used to "roll up" information (e.g., across multiple endpoints for decision-makers at
higher policy levels, versus individual endpoints for analysts who need information in its
most detailed format).
Final Draft 41
-------
EXECUTIVE INDICES MANAGEMENT INDICES
INDIVIDUAL VARIABLES
Landscape Quality
(Riparian Habitat + Overall Habitat - Human Use)
Water Quality
(Riparian Habitat - Nutrients - Sediment)
Human Use
Riparian Habitat
Nutrients
Sediment
Quality of Life
Quality of Life
Riparian Cropping, Percent
Agriculture, Percent Urban, Percent
Agriculture on Slope, Road Density,
Streams Crossing Roads. Percent
Forest Edge, etc.
Riparian Forest, Riparian Shrub,
Riparian Grass, Stream Density,
Percent Forest, Percent Wetland,
Percent Shrub, Percent Native
Grass, Percent Connectivity,
Percent Transitional Forest, etc.
Dissolved Phosphorus, Nitrate and
Nitrite Nitrogen, Riparian Cropping,
Percent Agriculture, Road Density,
Streams Crossing Roads, etc
Percent Barren, Riparian
Agriculture, Riparian Cropping,
Agriculture on Slope,
Imperviousness, Streams Crossing
Roads, etc.
Percent Employed in Professional
Occupation, Employment Diversity,
Travel Time to Work, Change in
Travel Time to Work, Violent
Crimes, Housing Affordability,
Seasonal Housing Per Capita,
Protected Areas Per Capita, etc.
Figure 9. Graphic depicting an example of variables and indices produced for different levels of
users within the Sustainable Environment for Quality of Life (SEQL) project.
Visualization
Visualization of results is an effective way to communicate information. Since the
approaches presented here are designed for spatially explicit analyses, mapped results are
an assumed product. Careful attention to the details of how results are communicated is
important even for an analytically-minded audience because differences in mapping can
lead the user to very different conclusions. Examples of aspects that must be considered
in mapping results include: (1) how relative differences in metrics or indices are "binned"
across the region, (2) the choice of color codes for representing high/low or good/bad
conditions, (3) the most appropriate representation of data distributions (normal versus
skewed), and (4) how best to visualize metadata (i.e., the distribution of sample points or
error/uncertainty maps).
The ReVA methodology generally encourages the use of overviews of individual metrics
and indices for regional perspectives. However, ReVA users inevitably have the urge to
drill down to individual reporting units and examine conditions at fmer-than-regional
spatial scales. Relationships between variables are best represented using standard
statistical graphics such as scatter plots, bar charts, and box diagrams (Figures 10-12).
42
Final Draft
-------
Quick visualization of integrated results for individual reporting units also can be
accomplished by using graphics such as the "radar plot" (Figure 13).
o
I
^
8
I
'g.
g
12 -
10 -
8 -
6 -
4 -
2 -
0 -
20 40 60
Crop agriculture land cover along streams - 60 meters (%)
80
Figure 10. Graphic depicting a scatter plot comparing two variables, nonpoint source nitrogen
loadings as estimated by the model LTHIA and percent crop agriculture along streams
within a 60-m buffer.
250 -i
200 -
* 100 -
50 -
0 J
0 20 40 60
Crop agriculture land cover along streams - 60 meters (%)
r250
-200
-150
-100
-50
KM
Figure 11. Graphic depicting a histogram of number of watersheds and percent crop agriculture
within a 60-m buffer along streams.
Final Draft
43
-------
E
I
I—
A
N.kgha>3.3
1.8 < N.kgha < 3.3
:2 0.4 < N.kgha < 1.8
N.kgha<0.4
0 20 40 60
Crop agriculture land cover along streams - 60 meters (%)
Figure 12. Graphic depicting a box plot of nonpoint source nitrogen with percent crop agriculture
within a 60-m buffer along streams.
e . • a <* •»
Figure 13. Graphic depicting a screenshot from a ReVA Environmental Decision Toolkit (EOT) with a
radar plot for a displayed 8-digit HUC. In a radar plot, each spoke of the wheel represents
an individual variable and the amount of green represents the relative rank of that variable
in relation to the same variable in all other 8-digit HUCs across the region (green
represents good conditions, not green represents poor conditions).
44
Final Draft
-------
From the perspective of ReVA and many types of ecological analysis, it is ideal to have
all data available as a surface map or data in a form that could be reformulated into a
surface using some type of model. However, because data are not always available in
this form, data are aggregated into reporting units and relative values for these reporting
units are mapped across the region. Various options are available for dividing data into
categories for comparison. These options include: (1) equal numbers of reporting units
(e.g., watersheds or counties) in each bin or category (Figure 14), (2) equalized value
ranges within each category (that is, if the range of values is 1-10 and the user wants five
classes or bins, each bin would have a value range of 2) (Figure 15), (3) natural breaks in
the data, where classes are based on natural groupings of data values (Figure 16), and (4)
customized binning, in which classes are designed to highlight specific points in the data
distribution, such as all reporting units exceeding a threshold and the spread of reporting
units not exceeding this threshold.
The choice of color codes is important for several reasons. The first, and probably most
important, reason is that choice of color can impart a subtle (or not so subtle) value
judgment, such as occurs in the use of red-to-green colors selected to represent poor-to-
good conditions across the map. This color-code choice may be the message a ReVA
user wants to communicate. However, for some metrics, such as socioeconomic data, use
of these colors may convey an unintended message. A good resource for selecting colors
to represent relative differences in metric or index values is the ColorBrewer Web site,
located at: http://www.personal.psu.edu/cab38/ColorBrewer/ColorBrewer intro.html.
MLCD 2001 (% Forest)
0.458 - 8.08
8.fl8 -19.337
19-337 - 34.479
34.479 - 52.298
52.298-91.962
Figure 14. Graphic showing the percent of forest cover for every 8-digit HUC across EPA Region 5 as
displayed using quintiles as the binning method.
Final Draft 45
-------
NLCD 2001 (% Forest)
_] 0.458 - 18.759
^\ 18.759 -37.059
>H 37.059 - 55.36
BH] 55.36 - 73.661
IB 73.661 -91.962
Figure 15. Graphic showing the percent of forest cover for every 8-digit HUC across EPA Region 5 as
displayed using equal intervals as the binning method.
NLCD 2001 (% Forest)
M 0.458 -13.174
H 13.174-27.855
BB 27.855-45.122
^45.122-63.721
BB 63.721 -91.962
Figure 16. Graphic showing the percent of forest cover for every 8-digit HUC across EPA Region 5 as
displayed using natural breaks as the binning method.
46
Final Draft
-------
A clear understanding of data distributions is important when reviewing results, because
differences in data distributions can inform the user as to how to interpret the mapped
output. This understanding also is important because some of the data integration
methods assume normal distributions. Bar charts, such as those that are provided by most
statistical packages, are a good way to inspect data distributions; these charts allow a user
to rapidly judge whether the different datasets are normally distributed, multimodal, or
highly skewed. Evaluating data distributions also is useful in that it allows the user to
better determine binning for reporting units when mapping results.
When feasible, metadata should be visualized, as this can be extremely valuable for users
of individual data coverages. An example of this is the case in which a surface coverage
has been developed using monitoring data. A map showing the number and distribution
of monitoring points can be invaluable in communicating sampling density and areas of
coverage. Similar benefits are evident for models that have error or uncertainty estimates
that can be mapped; such maps can help communicate the validity of the model.
Other options for displaying results of analysis while maintaining the spatial context
include using techniques such as linked micromaps. Key characteristics of the micromap
template, which enhances graphical perception, are the ability to: (1) use position along a
scale to represent estimates, (2) include multiple variables, (3) display confidence
intervals for estimates, and (4) group large amounts of information into meaningful and
manageable units for human interpretation. The micromap template consists of four
elements: (1) parallel sequences of panels, (2) sorted study units, (3) partitioned study
units, and (4) linked study units across corresponding panels (Carr et al., 1998; Carr et al.,
2000; Carr et al., 2003). Figure 17 shows an example of linked micromaps used within a
ReVA Environmental Decision Toolkit.
Final Draft 47
-------
Figure 17. Graphic depicting linked micromaps.
Another consideration for visualizations of mapped results is that of adding locational
information to help orient users of the information. This type of "orientational"
information may include things such as state boundaries, major cities, county lines, etc.
Finally, especially when comparing alternative scenarios, difference maps are particularly
effective. Difference maps highlight the differences between the current state and each of
the alternatives. These maps allow users to see the trade-offs for the entire region and
trade-offs among individual reporting units (Figure 18). If individual variables/metrics
are also mapped with difference maps, the trade-offs can also be tracked among various
endpoints. One watershed, for example, might gain in economic development under one
scenario, but suffer declines in water quality.
48
Final Draft
-------
Landscape Quality Index for Watersheds
Medium Density Compact Centers
n
Scenario Change
Medium Density better
Same
Compact Centers better
Less stressed
D
n
n
More stressed
5 20
10 50
Difference Map
Figure 18. Graphic depicting the comparison between two future alternative scenarios (upper maps)
with a difference map highlighting both individual watershed differences as well as overall
regional differences.
Final Draft
49
-------
50
Final Draft
-------
Glossary
Area-weighting (areal interpolation): A method of apportioning data from one
geographic boundary to another when the boundaries do not match. For example, if
20% of a county is located in HUC 1 and 80% is located in HUC 2, then 20% of the
population for the county would be assigned to HUC 1 and 80% would be assigned to
HUC 2. An area-weighting method involves the assumption that values (the number
of people, in this example) are evenly distributed across space (in this case the
county).
Bayesian statistical methods: Statistical methods characterized by the updating of
prior knowledge and estimation of conditional probabilities using Bayes' theorem and
by the treatment of probabilities as subjective degrees of belief.
Block groups: As defined by the U.S. Census Bureau, a block group is a cluster of
census blocks having the same first digit of their four-digit identifying numbers
within a census tract. Block groups generally contain between 600 and 3,000 people,
with an optimum size of 1,500 people.
Continuous variables: A quantitative variable that can take on any value over its
range, including fractional values. Examples are measures of time, temperature, and
chemical concentrations.
Criticality analysis: An integration method similar to the State Space method in that
it measures distance from a reference state. But in this case, the reference state is a
postulated prehuman or totally non-disturbed state.
Difference map: A GIS analytical technique where map algebra is used to subtract
the values from one map from another.
Directionalization: In order to combine multiple variables into aggregate indices,
some variables may have their values reversed (directionalized) in order to maintain a
consistent definition for improvement or deterioration with a change in a variable
score, such as "higher is better." Scores that formerly ranged from 0-100 might be
reversed so that 0 becomes 100 and 100 becomes 0, with all other values inverted
proportionally.
Discrete (integer/categorical) variables: Variables for which the values are not
observed on a continuous scale because of the existence of gaps between possible
values. Examples include integer values (number of people) or qualitative values
(fair, good, moderate) that may be represented as numeric values.
Ecoregion: A large area whose boundaries are fixed by geography, topography,
climate, vegetation, and other easily recognized natural features of landscape.
Ecoregions contain many landscapes with different spatial patterns of ecosystems.
Final Draft 51
-------
Ecosystem: The sum of the biotic and abiotic environment within which most or all
nutrients are recycled.
Ecosystem services: The goods and services that people value that have natural
functions or features as inputs. These goods and services cover a broad range, from
food products to spiritual and cultural benefits. Ecosystem services can be divided
into use and nonuse services, where use services are distinguished primarily by their
requirement that users have direct access or proximity to sites generating goods and
services, whereas nonuse services can accrue to those who are not close to the site
and may never intend to visit the site.
Ecotoxicological ECx values: A concentration above which an associated adverse
effect occurs, for "X" percent of the individuals in a population.
Empirical model: A mathematical model that is derived by fitting a function to data
using statistical techniques or judgment.
Endpoints: A technical term used to describe the environmental value that is to be
protected. An environmental value is an ecological unit and its characteristics. For
example, salmon are valued ecological units; reproduction and age class structure are
some of their important characteristics. Together "salmon reproduction and age class
structure" form an endpoint.
Euclidean distance: The straight-line distance between two points on a plane.
Euclidean distance, or distance "as the crow flies," can be calculated using the
Pythagorean Theorem.
Extrapolation: The use of related data to estimate an unobserved or unmeasured
value.
Fuzzy logic: A form of logic in which variables can have degrees of truth or
falsehood.
Geographic Information System (GIS): A GIS is a system of hardware and
software used for storing, retrieving, mapping, and analyzing geographic data. It is a
computer technology that brings together all types of information based on
geographic location for the purpose of query, analysis, and generation of maps and
reports. GIS is both a database designed to handle geographic data and a set of
computer operations ("tools") that can be used to analyze the data. In a sense, GIS
can be thought of as a higher-order map.
Geospatial data: Information about the locations and shapes of geographic features
and the relationships between them, usually stored as coordinates and topology; any
data that can be mapped.
52 Final Draft
-------
Hydrologic Unit Code (HUC): A hierarchical, numeric code that uniquely identifies
hydrologic units. The first two digits identify the region, the first four digits identify
subregions, the first six digits identify accounting units, and the full eight digits
identify subbasins. From the above example (definition of a hydrologic unit), the
hydrologic unit codes are:
02 - the region (Mid-Atlantic)
0206 - the subregion (Upper Chesapeake)
020600 - the accounting unit (Upper Chesapeake. Delaware, Maryland, Virginia,
and Pennsylvania)
02060002 - the subbasin (Chester-Sassafras. Delaware, Maryland, Pennsylvania)
Zeroes in the two-digit accounting unit field indicate that the accounting unit and the
subregion are the same. Zeroes in the two-digit subbasin field indicate that the
subbasin and the accounting unit are the same.
Index: A combination of multiple indicators.
Indicator: A concise measure of cumulative effects and ecosystem vulnerability.
Interpolation: A method of constructing new data points within the range of a
discrete set of known data points. Interpolation can be performed on spatial or
nonspatial datasets.
Inverse distance weighting (IDW): An interpolation technique that estimates values
in a raster from a set of sample points that have been weighted so that the farther a
sampled point is from the cell being evaluated, the less weight it has in the calculation
of the cell's value.
Kriging: An interpolation technique in which the surrounding measured values are
weighted to derive a predicted value for an unmeasured location. Weights are based
on the distance between the measured points, the prediction locations, and the overall
spatial arrangement (or autocorrelation) among measured points. The resultant
interpolated points do not necessarily have to pass exactly through the input points.
Land cover: Anything that is visible from above the Earth's surface. Examples
include vegetation, exposed or barren land, water, snow, and ice.
Land use: The way land is developed and used with respect to the kinds of
anthropogenic (human-induced) activities that occur (e.g., agriculture, residential
uses, industrial uses).
Final Draft 53
-------
Mahalanobis distance: A multivariate distance measure that is based on correlations
between several variables. It is a useful way of determining similarity of an unknown
sample set to a known one. It differs from Euclidean distance in that it takes into
account the internal correlations of the dataset and is scale-invariant, i.e., not
dependent on the scale of measurements.
Metadata: Data that describe the content, lineage, quality, condition, and other
characteristics of data. They are "data about data."
Model: A mathematical, physical, or conceptual representation of a system.
Monte Carlo uncertainty analysis: A computational method that involves repeated
random sampling from the original (i.e., full) dataset in order to calculate results.
Typically Monte Carlo simulations are performed when the underlying parameter
cannot be estimated using deterministic methods.
Non-Indigenous Species (NIS): Nonnative plant, animal, or microbe species
introduced into a region. Often, NIS can have significant impacts such as:
overwhelming, crowding out, or disrupting relationships among native species,
degrading habitats, and contaminating the gene pools of indigenous species.
Examples include the wooly adelgid (an insect damaging hemlock trees in the Smoky
Mountains), kudzu, and fire ants in southern U.S. and more than 160 known aquatic
species in the Great Lakes.
Nonpoint Source Pollution: Pollution with a nonspecific location (i.e., those that
are not discharged from a pipe outfall). The sources of the pollutant(s) are dispersed,
not well defined, and typically not constant. Rainstorms and snowmelt often
transport pollutants, increasing impacts. Examples include sediments from
construction sites and chemical-bearing runoff from road surfaces and agricultural
fields.
Normalization: A statistical technique that divides multiple sets of data by a
common variable in order to negate that variable's effect on the data, thus allowing
underlying characteristics of the different variables in a dataset to be compared. One
common normalization technique subtracts the mean from each value and divides it
by the standard deviation. This particular normalization will result in all the variables
having a mean of 0 and a standard deviation of 1.
Ordination: A general class of multivariate statistical procedures used to create
categories (or groups) of similar values. PCA is one of several different ordination
techniques.
Overlay method: As applied by ReVA, an integration method that attempts to
identify watersheds where important resources still exist but the remaining resources
are under significant stress. Such watersheds are vulnerable in the sense that further
stress, e.g., from additional development, could result in the loss of valued resources.
54 Final Draft
-------
The stressors and the sensitive receptors in a location are summed separately so they
can be compared. The two sets of values are overlaid to create a 2-dimensional
scoring system that includes the potential for four end-members: 1) low-stress, low-
resource; 2) high-stress, low-resource; 3) low-stress, high-resource; and 4) high-
stress, high-resource. The latter is the most vulnerable situation.
Principle Components Analysis (PCA): A widely used multivariate statistical
technique which can be used to reduce the number variables analyzed. PCA
orthogonally transforms the original variables into a new set of uncorrelated variables
based on the covariance (or correlation) matrix.
Projective modeling: A model used to predict future conditions based on an
extension of past trends.
Prospective modeling: A model used to predict future conditions based on a change
in existing trends (e.g., change in management practices or land use).
Quantile: Points taken at regular intervals from the distribution of a variable,
dividing ordered data into equal-sized data subsets. Quantile classification is well-
suited to linearly distributed data and histograms are a common graphic
representation of quantiles. When quantiles are used to display spatial data, results
must be interpreted carefully because similar features may be separated into adjacent
classes, or features with widely different values can be lumped into the same class.
This distortion can be minimized by increasing the number of classes.
Quintile: A special name when data are split into 5-quantiles.
Raster: A spatial data model that defines space as an array of equally-sized cells
arranged in rows and columns, and composed of single or multiple bands. Each cell
contains an attribute value and location coordinates. Unlike a vector structure, which
stores coordinates explicitly, raster coordinates are contained in the ordering of the
matrix. Groups of cells that share the same value represent the same type of
geographic feature.
Reporting unit: Any defined area (e.g., an 8-digit USGS hydrologic unit code
"HUC," county) for which a landscape metric (e.g., percent urban) is calculated.
Resource: Any feature, good, or quality that can serve as an input into production of
a desired outcome. A resource can be a natural endowment such as fresh water, a
built product such as a road, or a social institution such as the people associated with
a particular school.
Revised Universal Soil Loss Equation (RUSLE): A soil erosion model developed
by the USD A Agricultural Research Service in 1993. It contains the same general
formula as Universal Soil Loss Equation (USLE), but has several improvements in
determining factors. These include some new and revised isoerodent maps, a time-
Final Draft 55
-------
varying approach for a soil erodibility factor, a subfactor approach for evaluating the
cover-management factor, a new equation to reflect slope length and steepness, and
new conservation-practice values.
Scale: The spatial or temporal dimension over which an object or process exists, as
in, for example, a landscape, or a forest ecosystem or community.
Shapefile: A vector data storage format specific to ESRI (Environmental Software
Research Institute) for storing the location, shape, and attributes of geographic
features.
Spatially explicit: An indication that geo-referenced data are used or created and
that a relatively fine scale of spatial disaggregation is used in evaluation.
Splining: An interpolation method in which values are estimated using a
mathematical function that minimizes overall surface curvature, resulting in a smooth
surface that passes exactly through the input points. Splines can be mathematically
adjusted by increasing or decreasing the tension in between points
State space analysis: An integration method that measures the distance between two
points (i.e., two watersheds) in multivariate space.
Stressor: A physical, chemical, or biological factor that can disrupt, change, or
otherwise alter ecosystem health and/or human health in a negative way. For
example, pesticides used in agriculture can be stressors to both ecosystem health and
human health.
Trend surface analysis: A surface interpolation method that fits a polynomial
surface by least-squares regression through the sample data points. This method
results in a surface that minimizes the variance of the surface in relation to the input
values. The resulting surface rarely goes through the sample data points. This is the
simplest method for describing large variations, but the trend surface is susceptible to
outliers in the data. Trend surface analysis is used to find general tendencies of the
sample data, rather than to model a surface precisely.
Universal Soil Loss Equation (USLE): A widely used soil erosion model first
developed by the U.S. Department of Agriculture in the early 1960s. USLE predicts
the long-term average annual rate of erosion on a field slope based on rainfall pattern,
soil type, topography, crop system, and management practices.
Variable: A quantity that can take on discrete or continuous values to represent
condition (e.g., of an ecosystem). Often used interchangeably with an indicator.
Vector (vector element): A coordinate-based data model that represents geographic
features as points, lines, and polygons. Each point feature is represented as a single
coordinate pair, while line and polygon features are represented as ordered lists of
56 Final Draft
-------
vertices. Attributes are associated with each vector feature, as opposed to a raster data
model, which associates attributes with grid cells.
Watershed: A watershed is an area of land that is drained by a single stream, river,
lake, or other body of water. Ridges form the dividing lines between watersheds.
Water on one side of the ridge flows into one stream and water on the other side of
the ridge flows into a different stream. Thus, a watershed is a natural unit defined by
the landscape.
Final Draft 57
-------
58
Final Draft
-------
References
Andreasen, J.K., R.V. O'Neill, R. Noss, and N.C. Slosser. 2001. Considerations for the
development of a terrestrial index of ecological integrity. Ecological Indicators
1:21-35.
Banerjee, S., B.P. Carlin, and A.E. Gelfand. 2004. Hierarchical Modeling and Analysis
for Spatial Data. Chapman and Hall/CRC, New York, New York, USA. 472pp.
Carr, D.B., A.R. Olsen, J.P. Courbois, S.M. Pierson, andD.A. Carr. 1998. Linked
micromap plots: named and described. Statistical Computing & Graphics
Newsletter 9(l):24-32.
Carr, D.B., A.R. Olsen, S.M. Pierson, and J.P. Courbois. 2000. Using linked micromap
plots to characterize Omernik ecoregions. Data Mining and Know ledge
Discovery 4:43-67.
Carr, D.B., S. Bell, L. Pickle, Y. Zhang, and Y. Li. 2003. The state cancer profiles web
site and extensions of linked micromap plots and conditioned choropleth map
plots. Proceedings of the Fourth National Conference on Digital Government
Research.
Chambers, J.Q., G.P. Asner, D.C. Morton, L.O. Anderson, S.S. Saatchi, F.D.B. Espirito-
Santo, M. Palace, and C. Souza Jr. 2007. Regional ecosystem structure and
function: ecological insights from remote sensing of tropical forests. Trends in
Ecology and Evolution 22(8): 414-423.
Clarke, K.C., L. Gaydos, and S. Hoppen. 1997. A self-modifying cellular automaton
model of historical urbanization in the San Francisco Bay area. Environment and
Planning B: Planning and Design 24:247-261.
Cressie, N.A.C. 1993. Statistics for Spatial Data. Wiley, New York, Revised edition.
Wiley, New York. 900pp.
De Maesschalck, R., D. Jouan-Rimbaud, and D.L. Massart. 2000. The Mahalanobis
di stance. Chemome tries and Intelligent Laboratory Systems 50:1-18.
Fagerstram, T. 1987. On theory, data, and mathematics in ecology. Oikos 50:258-261.
Ferenc, S.A., and J.A. Foran (eds.) 2000. Multiple stressors in ecological risk and impact
assessment: approaches to risk estimation. SETAC Press, Pensacola, FL.
Foran, J.A., and S.A. Ferenc (eds.) 1999. Multiple stressors in ecological risk and impact
assessment. SETAC Press, Pensacola, FL.
Final Draft 59
-------
Greene, E.A., A.E. LaMotte, and K.A. Cullinan. 2005. Ground-Water Vulnerability to
Nitrate Contamination at Multiple Thresholds in the Mid-Atlantic Region Using
Spatial Probability Models. USGS Scientific Investigations Report 2004-5118.
Grimm, J.W., and J.A. Lynch. 2000. Enhanced wet deposition estimates for the
Chesapeake Bay watershed using modeled precipitation inputs. Rep. CBWP-
MANTA-AD-99-2. Maryland Department of Natural Resources, Annapolis.
Harbor, J., andM. Grove. 1997. L-THIA: Long-Term Hydrological Impact Assessment -
A Practical Approach. Ohio Environmental Education Fund/Purdue University,
Manual.
Hardie, I.W., and PJ. Parks. 1997. Land use in a region with heterogeneous land
quality: An application of an area base model. American Journal of Agricultural
Economics 79:299-310.
Hardy, R.L. 1971. Multiquadric equations of topography and other irregular surfaces.
Journal of Geophysical Research 76:1905-1915.
Hunsaker, C.T.D., A. Levine, S.P. Timmins, B.L. Jackson, and R.V. O'Neill. 1992.
Landscape characterization for assessing regional water quality, pp. 997-1006. In
D.HMcKenzie, D.E. Hyatt, and VJ. McDonald (eds.), Ecological Indicators.
Elsevier Applied Science, New York, NY.
Jackson, L.E., S.L. Bird, R.W. Matheny, R.V. O'Neill, D. White, K.C. Boesch, and J.L.
Koviach. 2004. A Regional approach to projecting land-use change and resulting
ecological vulnerability. Environmental Monitoring and Assessment 94:231-248.
Jones, K.B., A.C. Neale, M.S. Nash, R.D. Van Remortel, J.D. Wickham, K.H. Riitters,
and R.V. O'Neill. 2001. Predicting nutrient and sediment loadings to streams
from landscape metrics: a multiple watershed study from the United States Mid-
Atlantic Region. Landscape Ecology 16(4):301-312.
Klir, G.J., and Y. Bo. 1995. Fuzzy Sets and Fuzzy Logic: Theory and Applications.
Prentice Hall, Upper Saddle River, NJ. 592 pp.
Liotta, P.H. 2005. Through the looking glass: creeping vulnerabilities and the reordering
of security. Security Dialogue 36(1):49-70.
Liotta, P.H., and J.F. Miskel. 2004. Redrawing the map of the future. World Policy
Journal 2l(l):\5-2l.
Liu, Y., H. Guo, Z. Zhang, L. Wang, Y. Dai, and Y. Fan. 2007. An optimization method
based on scenario analysis for watershed management under uncertainty.
Environmental Management 3 9:678-690.
60 Final Draft
-------
Pandey, S., J. Harbor, and B. Engel. 2000. Internet based geographic information
systems and decision support tools. Urban and Regional Information Systems.
Park Ridge, IL. 36pp.
Parks, J.P., I.W. Hardie, C.A. Tedder, and D.N. Wear. 2000. Using resource economics
to anticipate forest land use change in the U.S. Mid-Atlantic region.
Environmental Monitoring and Assessment 63:175-185.
Peterson, A.T., M. Papes, and D.A. Kluza. 2003. Predicting the potential invasive
distributions of four alien plant species in North America. Weed Science 51:863-
868.
Prestemon, J.P., and R.C. Abt. 2002. The Southern timber market to 2040. Journal of
Forestry 100(7): 16-22.
Reckhow, K.H., M.N. Beaulac, and J.T. Simpson. 1980. Modeling Phosphorus Loading
and Lake Response Under Uncertainty: A Manual and Compilation of Export
Coefficients. USEPA 440/5-80-011.
Schaberg, R.H., and R.C. Abt. 2004. Vulnerability of Mid-Atlantic forested watersheds
to timber harvest disturbance. Environmental Monitoring and Assessment 94:
101-113.
Smith, E.R., L.T. Tran, and R.V. O'Neill. 2003. Regional Vulnerability Assessment for
the Mid-Atlantic Region: Evaluation of Integration Methods and Assessments
Results. EPA/600/R-03/082.
Smith, E.R., L.T. Tran, R.V. O'Neill, and N.W. Locantore. 2004. Regional Vulnerability
Assessment of the Mid-Atlantic Region: Evaluation of Integration Methods and
Assessments Results. EPA/600/R-03/082 (NTIS PB2004-104952). U.S.
Environmental Protection Agency, Washington, DC.
Theobald, D.M. 2005. Landscape patterns of exurban growth in the USA from 1980 to
2020. Ecology and Society 10( 1): 3 2
Tran, L.T., and L. Duckstein. 2002. Comparison of fuzzy numbers using a fuzzy
distance measure. Fuzzy Sets and Systems 130(3):331-341.
Tran, L.T., R.V. O'Neill, and E.R. Smith. 2006. A generalized distance measure for
environmental integrated assessment. Landscape Ecology 21:469-476.
Tran, L.T., R.V. O'Neill, E.R. Smith, and C.G. Knight. 2007. Sensitivity analysis of
aggregated environmental indices with a case-study of the Mid-Atlantic Region.
Environmental Management 39:506-514.
Final Draft 61
-------
Tran, L.T., R.V. O'Neill, and E.R. Smith. 2007. Determining the most dominant
stressors and most impacted resources for integrated environmental assessment.
(In review).
U.S. EPA. 2000. Projecting Land-Use Change: A Summary of Models for Assessing the
Effects of Community Growth and Change on Land-Use Patterns. EPA/600/R-
00/098. U.S. Environmental Protection Agency, Washington, DC. 260 pp.
U.S. EPA. 2003. Generic Ecological Risk Assessment Endpoints (GEAEs) for
Ecological Risk Assessment. EPA/630/P-02/004F. U.S. Environmental
Protection Agency, Washington, DC. 67 pp.
U. S. EPA. 2004. Analytical Tools Interface for Landscape Assessment (ATtlLA).
EPA/600/R-04/083. U.S. Environmental Protection Agency, Las Vegas, NV. 39
pp.
U.S. EPA SAB (Science Advisory Board). 2002. A Framework for Assessing and
Reporting on Ecological Condition: An SAB Report. EPA-SAB-EPEC-02-009.
U.S. EPA SAB (Science Advisory Board). 2005. Advisory on ORD 'sRegional
Vulnerability Assessment Program. EPA-SAB-ADV-06-001. 41 pp.
Wagner, P.P., R.V. O'Neill, L.T. Tran, M. Mehaffey, T. Wade, and E.R. Smith. 2006.
Regional Vulnerability Assessment for the Mid-Atlantic Region: Forecasts to
2020 and Changes in Relative Condition and Vulnerability. EPA/600/R-06/088.
37 pp. U.S. Environmental Protection Agency, Las Vegas, NV.
Weingand, D.E. 1995. Preparing for the new Millennium: the case for using marketing
strategies. Library Trends 43(3):295-317.
Wheeler, J.O., P.O. Muller, G.I. Thrall, and TJ. Fik. 1998. Economic Geography. John
Wiley and Sons, New York, NY. 416 pp.
Wickham, J.D., R.V. O'Neill, K.H. Riitters, T.G. Wade, and K.B. Jones. 1997.
Sensitivity of selected landscape pattern metrics to land-cover misclassification
and differences in land-cover composition. Photogrammetric Engineering and
Remote Sensing 63:397-402.
62 Final Draft
-------
-------
&EPA
United States
Environmental Protection
Agency
Office of Research
and Development (8101R)
Washington, DC 20460
Official Business
Penalty for Private Use
$300
EPA/600/R-08/XXX
April 2008
www.epa.gov
Please make all necessary changes on the below label,
detach or copy, and return to the address in the upper
left-hand corner.
If you do not wish to receive these reports CHECK HERE D;
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
V
Recycled/Recyclable
Printed with vegetable-based ink on
paper that contains a minimum of
50% post-consumer fiber content
processed chlorine free
------- |