United States
Environmental Protection
Agency
Office of Research and
Development
Washington, D.C. 20460
EPA/600/R-99/105
June 1999
Land Cover Trends:
Rates, Causes, and
Consequences of
Late-Twentieth Century
U.S. Land Cover Change
^USGS
SEPA
United States
Erwjrorimtntil Protection Agency
-------
EPA/600/R-99/105
June 1999
Land Cover Trends:
Rates, Causes, and Consequences
of Late-Twentieth Century
U.S. Land Cover Change
Research Plan
EPA-IAG Project No. DW14938108-01-0
Thomas R. Loveland
Principal Investigator
U.S. Geological Survey, EROS Data Center
Sioux Falls, SD 57198
T. Sohl, K. Sayler, A. Gallant, J. Dwyer, J. Vogelmann, G. Zylstra
Co-Investigators
Raytheon ITSS, Inc.
USGS EROS Data Center
Sioux Falls, SD 57198
and
Tim Wade, Curt Edmonds, Deb Chaloud, and Bruce Jones
EPA Collaborators
National Exposure Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Las Vegas, NV 89193
-------
Notice
The United States Environmental Protection Agency (EPA), through its Office of Research and
Development (ORD), partially funded and collaborated in the research described here. This manuscript
has been subject to external and EPA peer review and approved for publication. Mention of trade names
or commercial products does not constitute endorsement or recommendation by the EPA for use.
-------
Table of Contents
Abstract v
Section 1 Project Rational 1
1.1 Science Issues 1
1.2 Project Goal and Objectives 2
1.3 Relevance to Other Programs 4
Section 2 Review of Past Land Cover Trends Research 5
2.1 Land Cover Change Detection 5
2.2 Land Cover Change Accuracy Assessment 7
Section 3 Overall Project Strategy 8
3.1 Framework Elements 8
32 Data Quality Objectives 10
Section 4 Methodology 11
4.1 Ecoregion Profiles and Ancillary Databases 11
4.2 Ecoregion Sampling 11
4,3 Satellite Databases 14
4.4 Land Cover Biophysical Properties 14
4.4.1 Assessment of Environmental Gradients 15
4.5 Land Cover Change Analysis 16
4.5.1 Land Cover Change Mapping 16
4.5.2 Accuracy Assessment/Validation of Results 17
4.6 Landscape Configuration Metrics 18
4.7 Ecoregion Trends Analysis 20
4.8 Assessment of Land Cover Change Drivers and Consequences 20
4.9 National Synthesis 21
Section 5 Planned Deliverables and Timelines 22
Section 6 Management Plan 24
Section 7 References 25
Appendix A Land Cover Definitions 30
Appendix B Description of Pilot Ecoregions 31
Appendix C Omernik Level III Ecoregions. Sample Information 36
in
-------
List of Figures
Figure 1. Distribution of 20 km by 20 km sample blocks using the pilot test sampling parameters ... 13
Figure 2. An example temporal comparison of patch size and frequency for a specific
land cover type across three ecoregions 18
IV
-------
Abstract
Information on the rates, driving forces, and consequences of land use and land cover change is
important in studies addressing issues ranging from the health of aquatic resources to climate change.
Land use and land cover changes occur at all scales, and changes at local scales can have dramatic,
cumulative impacts at broader scales. Consequently, land use and land cover changes are not just of
concern at local and regional levels (i.e., because of impacts on land management practices, economic
health and sustainability, and social processes), but globally as well. Unfortunately, there is a paucity of
information on land use and land cover change except at very local levels. This four-year research project
between the U.S. Geological Survey and the U.S. Environmental Protection Agency has a goal to
document the types, geographic distributions, and rates of land cover change on a region-by-region basis
over the past 30 years for the conterminous U.S., and to determine some of the key drivers and
consequences of the changes. The objectives of the study are to:
• Develop a comprehensive methodology for using sampling and change analysis techniques and
Landsat MSS and TM data for measuring regional land cover change across the U.S.
• Characterize the types, rates, and temporal variability of change for a 30-year period.
• Document regional driving forces and consequences of change.
• Prepare a national synthesis of land cover change.
The estimates of conterminous U.S. rates, driving forces, and consequences of land cover change will
be developed for 84 ecoregions defined by Omernik of the U.S. Environmental Protection Agency. Using
a 20 km by 20 km grid covering each ecoregion, a random sample of cells will be selected for each of the
84 ecoregions. The sample size will be based on the expected spatial variance of land cover change
between grid blocks, a 1.0% margin of error and a 0.85 confidence level. Land cover data and change
analyses will be developed for each ecoregion and summarized for the conterminous U.S. The analysis of
change will be based on five dates of Landsat MSS and TM data (nominally 1973, 1980, 1986, 1992, and
2000). Data for each sample block will be geocoded, calibrated, and interpreted. The land cover change
variables that will be mapped include: (1) general land cover type; (2) landscape biophysical properties;
and (3) landscape pattern. The major emphasis will be placed on mapping the general land cover types
and changes for each sample block. The biophysical variables (i.e., vegetated fraction, bare fraction, and
shadow fraction), developed using spectral unmixing methods, will be used to understand landscape
condition, such as successional or other gradual land cover transitions. Finally, landscape pattern metrics,
describing the number, size, shape, and spatial relationship of land cover patches (where patch is a
contiguous set of pixels of the same land cover type), will be generated from the general land cover data
for each block. These will permit the analysis of the spatial dimension of land cover change, and will also
contribute to the assessment of consequences of land cover change.
Our goal is to identify 1% change in general land cover within each ecoregion, at an 85%
confidence level. Initially, we will test our ability to achieve this goal in five pilot regions: (1) Montana
Valley and Foothill Prairies, (2) North Central Appalachians, (3) Northern Piedmont, (4) Southeastern
Plains, and (5) Madrean Archipelago. Based on the results of the pilot tests, we will refine and apply an
appropriate methodology to the remaining conterminous U.S. ecoregions.
-------
Section 1
Project Rational
Local land use and land cover changes are fundamental agents of global climate change and are
significant forces that impact biodiversity, water and radiation budgets, trace gas emissions, and
ultimately, climate at all scales (Riebsame et al., 1994). At local and regional scales, land cover change
can have profound impacts on aquatic systems due to new land use practices that adversely affect water
quality and sedimentation (Lowrance et al., 1985). Such changes also modify the composition of plant
communities through fragmentation, removal and introduction of species, alteration of nutrient and water
pathways, and alteration of disturbance cycles (e.g., Ojima et al., 1994).
Land use and land cover changes are driven by: (1) natural processes, such as climate and
atmospheric changes, wildfire, and pest infestation; (2) direct effects of human activity, such as
deforestation and road-building; and (3) indirect effects of human activity, such as water diversion
leading to lowering of the water table. Natural processes and human activities can both improve or
degrade the state of the land, so it is essential to distinguish beneficial from detrimental changes (Turner
and Meyer, 1991).
Land use and land cover changes occur at all scales, and changes at local scales can have dramatic,
cumulative impacts at broader scales. Consequently, land use and land cover changes are not just of
concern at local and regional levels (i.e., because of impacts on land management practices, economic
health and sustainability, and social processes), but globally as well. The challenge facing policy-makers
and scientists is that there is generally a lack of comprehensive data on the types and rates of land use and
land cover changes, and even less systematic evidence on the causes and consequences of the changes.
Lack of local and regional data of sufficient reliability and temporal and geographic detail frustrates
attempts at fine-tuned assessments of the implications of such changes.
The impacts of land use and land cover change is critical to many government programs. For
example, documenting the rates, driving forces, and consequences of change, particularly in aquatic
resources, is a central focus in the U.S. Environmental Protection Agency's Landscape Sciences Program
10-year strategic plan (Jones, et al., 1999). The strategic plan has a goal to complete a national
assessment of landscape change between the early 1970's and the early 2000's and to assess the
consequences of those changes on aquatic resources. This aggressive goal is further institutionalized in
the Agency's goals developed under the Government Performance Results Act.
1.1 Science Issues
The fundamental science questions associated with land use and land cover changes are:
• What are the types and geographic distributions of change?
• What are the overall rates of change by region and by sector (i.e., rates of conversion from
agricultural to urban land cover)?
• How do the rates vary (a) locally and regionally and (b) temporally?
-------
• What are the driving agents and consequences?
Developing our understanding in these areas will lead to improved ability to predict changes and,
consequently, to improved policies for practical, regional management of environmental resources.
Specific answers to these questions are currently not available. While federal resource inventory
programs, such as the U.S. Forest Service Forest Inventory and Analysis and the Natural Resources
Conservation Service Natural Resources Inventory, provide some answers, they fall far short of providing
the spatially explicit, thematically comprehensive data that are really required. On the other hand,
initiating a program to develop periodic, wall-to-wall mapping of land cover change for the U.S. at a
temporal interval appropriate for determining types, distributions, rates, agents, and consequences of
change is cost-prohibitive. Consider, for example, that the current 1992 land cover inventory conducted
by the USGS Land Cover Characterization Program will cost almost $10 million over 4 years. Providing
two additional periods of coverage suitable for documenting land cover changes would require a
minimum of $15 million over a four-year period. A more feasible and cost-effective strategy is to use a
sampling approach that incorporates a temporal, spatial, and information resolution appropriate for
regional and national evaluations.
As a first step in understanding the above science questions, this research project will focus on the
following four hypotheses:
1. Rates and characteristics of land cover change vary overtime and space.
2. Stratified random sampling, based on an ecoregion framework, can provide useful and
efficient measures of the spatial characteristics of land cover change.
3. Satellite data, specifically Landsat MSS and TM, can be used to provide accurate estimates of
regional land cover change which can then be aggregated to summarize change at a national
level.
4. Landscape change metrics derived from satellite remotely sensed data (e.g., land cover type,
biophysical measures, and landscape patterns) provide evidence of the driving forces and
consequences of regional land cover change.
The four hypotheses form the foundation for important basic research, with a number of measurement
components that can be addressed using remotely sensed data. Because of developments in large-area
statistical sampling techniques, availability of recent (baseline) Landsat TM-derived land cover data and
derivative products, availability and affordability of new satellite data (e.g., Landsat 7), and an established
track record for successfully handling large-area analyses, our multidisciplinary team of scientists is well-
poised to make substantial progress in addressing questions related to land use and land cover changes.
Toward this end, we will use a geographic framework for selecting and acquiring regionally
representative samples of remotely sensed and ancillary time series data in order to map the types,
distributions, rates, agents, and consequences of land cover change in the U.S. over the latter portion of
the twentieth century.
1.2 Project Goal and Objectives
Recognizing both the need and challenges for providing spatially-explicit contemporary land cover
trends data, the goal of this four-year research project will be to document the types, geographic
distributions, and rates of land cover change on a region-by-region basis over the past 30 years, and to
determine some of the key drivers and consequences of the changes. The USGS EROS Data Center and
-------
the Landscape Ecology Branch of the EPA Las Vegas Laboratory will collaborate in this effort. The
objectives for achieving this goal were defined to test the four hypotheses listed in the previous section.
The objectives are to:
• Develop a comprehensive methodology for using sampling and change analysis techniques and
Landsat MSS and TM data for measuring regional land cover change across the U.S. (hypothesis 2
and 3).
• Characterize the types, rates, and temporal variability of change for a 30-year period (hypothesis 1
and 2).
• Document regional driving forces and consequences of change (hypothesis 4).
• Prepare a national synthesis of land cover change (hypothesis 3).
A central premise of the project strategy is the use of a geographic framework for providing unbiased
estimates of regional land cover. Such an approach is sound (S. Stehman 1999, pers. commun.), and the
benefits of geographically subdividing the U.S. in order to adequately sample regions of interest has
already been demonstrated by the EPA in high-profile national aquatic resource surveys (Baker, 1990;
Kaufmann et al. 1991; Linthurst et al. 1986). Geographers have long used regional frameworks because
they capture the essence and potential of the landscape, without masking the roles of environmental,
social, and economic forces (Turner and Meyer, 1991). Peplies and Honea (1992) argue that ecoregions,
such as those defined by Omernik (1987), are the appropriate geographic framework for the study of
environmental change. In fact, the International Geosphere-Biosphere Programme's Land Use Cover
Change core science project has recommended that geographic regions be used as the strata to extrapolate
land use change observations from local to regional to global levels (Turner et al., 1995).
Relationships between the ecological regions delineated by Omernik and remotely sensed land cover
patterns have been noted, particularly in correlation with seasonal data (e.g., Ramsey et al. 1995;
Loveland et al. 1991). Ecoregions are currently being used to stratify remotely sensed data for some parts
of the U.S. in order to interpret and map land cover for the USGS/EPA Multi-Resolution Land
Characterization (MRLC) project. The ecoregions have proven useful for other types of environmental
interpretation as well, such as assessing patterns of aquatic resources (e.g., Larsen et al. 1988; Heiskary et
al. 1987).
Because Omernik's ecoregion framework was developed by synthesizing information on climate,
geology, physiography, soils, vegetation, hydrology, and human factors, the regions reflect patterns of
land cover and land use potential that correlate strongly with patterns visible in remotely sensed data. The
framework stratifies the nation into relatively homogenous units with respect to these factors. The
character of each region establishes the range of land cover changes that can potentially occur. In effect,
each ecoregion serves as a spatial model for the interplay between complex environmental and
anthropogenic factors. These ecoregions have been demonstrated to be useful strata for predicting
environmental responses, and are increasingly recognized as an important spatial framework for state-of-
the-environment reporting. By sampling land cover change for each of the ecoregions, using selected
epochs of data from the nearly 30-year record of Landsat 1-7 data, both regional and national
characteristics of change can be determined.
-------
1.3 Relevance to Other Programs
The EPA and USGS both recognize the necessity for land cover change data for use in environmental
risk and monitoring research. The need is particularly well-defined in the most recent 10-year strategic
plan for the EPA Land Sciences Program (Jones et al., 1999). This project will provide both
organizations with land cover change statistics that will aid in establishing an intellectual framework of
the temporal and spatial aspects of contemporary land use change. Additionally, the results of the project
are likely to be relevant to a number of other environmental assessment, land management, and science
programs:
• U.S. Global Change Research Program (USGCRP), including the land-use change and terrestrial
and marine ecosystems science priority and the National and Regional Climate Assessment cross-
cutting initiative
• The USGCRP Carbon Cycle Initiative which calls for the "evaluation of information from past and
current land-use changes, both from remotely sensed and historical records, to assess how human
activity has affected carbon storage on land."
• NASA Land Cover and Land Use Change Program
• Department of the Interior Inventory and Monitoring Initiative
• USGS Land Use History of North America
• Heinz Center Environmental Report Card
• Association of American Geographers (AAG), with National Science Foundation support, ongoing
project titled Global Change and Local Places, providing a model for interpretation of local driving
forces and signals of climate change
• International Geosphere-Biosphere Program Land Use Research Cover Change Core Project
• North America Free Trade Agreement Committee on Environmental Cooperation
• Intergovernmental Panel on Climate Change
The potential links between the proposed project and other programs are numerous. A key element of
the project will be liaisons with relevant programs, as well as support for key applications of our land
cover change data.
-------
Section 2
Review of Past Land Cover Trends Research
2.1 Land Cover Change Detection
There is considerable evidence demonstrating the use of Landsat data for investigating contemporary
land cover change. It is noteworthy that while a great deal has been written regarding change detection
techniques, very little guidance is available to address the special problems associated with large-area
(regional to global) image processing, and even less guidance addressing large-area change detection
(Dobson and Bright, 1994). Perhaps the most ambitious effort is the Humid Tropical Forest project,
where Landsat imagery from the 1970's to present were manually interpreted to identify patterns of
deforestation across the humid tropics (Skole and Tucker, 1993). Most land cover change studies do not
address land transformations for such a large area. The Coastal Change Analysis Project (C-CAP) is an
ongoing study of land cover change in the coastal zones of the U.S. (Dobson et al., 1995). In this project,
digital analyses of Landsat data have been used to track general land cover transformations. There are
also examples of operational programs in which sampling strategies involving both field observation and
air photo interpretation have been applied to determine the status and trends in land resources. Perhaps
the best example is the National Resources Inventory of the U.S. Department of Agriculture Natural
Resources Conservation Service. However, the use of Landsat data in a sample framework is not
commonplace.
Numerous papers discuss the various change analysis techniques commonly used (Singh, 1989).
There are two general approaches to change detection: (1) comparative analysis of independently
produced classifications and (2) simultaneous analysis of multitemporal data. Examples of the
simultaneous analysis techniques include image differencing, ratioing, principal component analysis
(PCA), and change vector analysis. Each has advantages and disadvantages.
The most straightforward technique for detecting change is the comparison of land cover
classifications from two dates. The use of independently produced classifications has the advantage of
compensating for varied atmospheric and phenological conditions between dates, or even the use of
different sensors between dates, because each classification is independently produced and mapped to a
common thematic reference. The method has been criticized, however, because it tends to compound any
errors that may have occurred in the two initial classifications (Gordon, 1980; Stow et al., 1980; Singh,
1989). The procedure has been widely used, and has successfully been employed for a variety of land
cover change investigations, including assessing deforestation (Massart et al., 1995) urbanization
(Dimyati et al., 1996), sand dune changes (Kumar et al., 1993), and conversion of semi-natural vegetation
to agricultural grassland (Wilcock and Cooper, 1992).
Simple image differencing is another technique widely used for change detection. This technique
involves taking the mathematical difference between geo-registered images from two dates. The input
data can be radiometrically calibrated raw imagery, or transformed data such as NDVI imagery. The
procedure has been used for coastal zone change detection (Weismiller et al., 1977), monitoring forest
change (Vogelmann, 1988), and detecting urban expansion (Jensen and Toll, 1982). While often
producing excellent results, it has been suggested that image differencing alone may be too simple a
procedure to adequately describe many surface changes (Weismiller et al., 1977; Jensen and Toll, 1982;
-------
Sohl, 1999). Other approaches used successfully to detect land cover change include image ratioing
(Howarth and Wickware, 1981) and PCA (Johnston and Haas, 1985; Bryne et al., 1980; Ribed and Lopez,
1995). See Singh (1989) for a discussion of the strengths and weaknesses of these approaches.
Surface change can also be described by a spectral change vector, which represents the direction and
magnitude of change from the first to the second date. An empirically derived or modeled threshold is
used to determine the minimum magnitude that represents a change occurrence. Dwyer et al. (1997) used
data from the brightness-greenness plane (Kauth and Thomas, 1976; Crist and Cicone, 1984) in the
development of a change vector analysis (CVA) toolkit. Sohl (1999) successfully used this toolkit in the
analysis of change in the United Arab Emirates. Change vector analysis has the advantage of providing a
high level of information regarding the magnitude and nature of a surface change, and is well-suited for
analyses in which continuous variables are measured, such as vegetation patterns along ecological
transitions and gradients.
An alternative strategy for detecting change involves spectral mixture modeling, whereby a
multispectral image is decomposed into spectral endmembers. There are numerous approaches to
stratifying the image into vegetated and non-vegetated components (Cochrane and Souza, 1998; Foschi
and Smith, 1997; Roberts et. al., 1993). The validation of spectral mixture modeling results can be
difficult, especially for large areas, and the determination and selection of appropriate spectral
endmembers is critical (Bateson and Curtiss, 1996; Tompkins et al., 1997). The spectral mixture
modeling approach, however, can yield meaningful results if a we 11-documented and consistent approach
is taken. Field validation of model results remains problematic, but if the modeling procedures are
prescribed and adhered to, then the comparison of fractional endmember components from one date of
imagery to another should yield consistent and interpretable results.
A number of change detection studies rely on combination or hybrid approaches that incorporate
many of the features of the techniques outlined above, or incorporate techniques that do not easily fit into
any of the above categories. Adams et al. (1995) measured changes in land cover by classifying images
based on spectral endmember fractions over a period of 4 years, with class names representing both
context and pixel history. Weismiller et al. (1977) tested the linking of decision tree classifiers for each
of two dates, thereby introducing within the tree a logic for detecting the desired changes. Sohl (1999)
used simple image differencing in combination with manually interpreted land cover information to
describe changes in agricultural and forest cover in the United Arab Emirates.
The prerequisite preprocessing step to many change analysis techniques is to calibrate the images to a
common radiometric reference. Ideally, this would involve the transformation of digital numbers to
physical values of radiance or reflectance, but the information required for this is not widely available. A
viable and widely used alternative is to perform a relative calibration between imagery from different
dates. This usually involves the use of a linear transformation in which the additive component corrects
for differences in atmospheric path radiance and the multiplicative component corrects for differences in
detector calibration, sun angle, Earth-Sun distance, atmospheric attenuation, and phase angle conditions
(Dwyer et al., 1997). Radiometric control sets representing temporally invariant features are used to
derive gains and offsets for the linear transformation. These control sets can be derived from a variety of
methods, including various methods identifying pseudo-invariant bright and dark targets (Caselles and
Garcia, 1989; Hall et al., 1991), the use of ratios of near-infrared to red radiances to identify non-
vegetated, non-water elements (Schott et al., 1988), and automated scattergram-controlled regression
(Elvidge et al., 1995).
A major attribute of the landscape is its spatial pattern and structure. Lambin and Strahler (1994)
showed that the detection of land-cover change processes by remote sensing is improved when using both
-------
spectral and spatial indicators of surface condition. They suggested that while spectral indicators are
more sensitive to fluctuations in primary productivity associated with the interannual variability in
climatic conditions, changes in landscape spatial pattern are more likely to reveal long term and long
lasting land cover changes. A wide variety of landscape metrics have been developed (Riitters et al,
1995). Turner (1990) used patch size metrics, indices of dominance, and indices of contagion to describe
changes in land use/land cover patterns in rural Georgia. Henderson and Walsh (1995) used measures of
connectivity, fractal dimension, and indices of dominance and fragmentation to describe land-cover
change on the North Carolina Piedmont.
2.2 Land Cover Change Accuracy Assessment
While a great deal has been written about change analysis techniques using remotely sensed data,
very little has been written on the subject of accuracy assessment of change products. Even fewer articles
have been written on accuracy assessment of large area change analysis databases. Standard accuracy
assessment procedures for one-point-in-time land cover products can be extremely difficult to apply to
multitemporal change analysis products. While accuracy assessment methods are well established for
small areas and single time periods, the assessment of accuracies for large areas, past time periods, and
change databases can become problematic (Dobson and Bright, 1994).
A standard accuracy assessment procedure for baseline land cover products involves the use of the
error matrix. The error matrix is an effective descriptive tool for organizing and presenting accuracy
assessment information and should be reported whenever feasible (Stehman, 1997). While the error
matrix can be modified and used for change analysis products (Macleod and Congalton, 1998), it is
difficult to apply to trend analysis or for adequately assessing more than a handful of categories of
change.
The assessment of trends requires additional tools besides the error matrix. Correlation is an
excellent tool, as it can quantify the agreement between reference and mapped data over several time
periods. For example, given mapped- and reference-derived values of percent forest cover, a correlation
approach could determine the level of agreement in the trends depicted in the data from the two sources.
This approach has the potential for other metrics than just percent cover if appropriate reference data are
available. The primary disadvantage of the approach is that errors of omission and commission can
compensate for each other, contributing to misleading overall measures of change. Accuracy assessment
of land cover classifications for individual dates could serve as a complement to the correlation values
through the incorporation of spatially dependent information.
Accuracy assessment of a large area change database is an extremely challenging task. A complete,
quantitative accuracy assessment may prove to be more expensive and labor intensive than the change
database itself. Dobson (1992) goes so far to say that it is infeasible to provide a quantitative estimate of
accuracy for a large spatial database. He states that a possible solution is to establish data quality
objectives (DQOs) designed to serve expected uses, establish and consistently implement a set of
protocols and procedures, and manage the data production process to meet the DQOs. Qualitative and
simple quantitative analyses often prove to be much more feasible than a comprehensive quantitative
analysis.
-------
Section 3
Overall Project Strategy
The project will address the types, distributions, rates, drivers, and consequences of change for the
conterminous U.S. between the early 1970's (nominally 1972-1973) and 2000. These issues will be
investigated in two phases. First, a pilot phase will focus analysis on five ecoregions:
• Montana Valley and Foothill Prairies
• North Central Appalachians
• Northern Piedmont
• Southeastern Plains
• Madrean Archipelago
These ecoregions have been selected because they: (1) offer a wide range of challenges for sampling
design, including steep environmental gradients, discontinuous and irregularly shaped strata, and a
spectrum of spatial variability in land cover types and scales of pattern and (2) relate to key interests of
the EPA Landscape Ecology Program, including assessment of hydrological dynamics of the Mid-
Atlantic region and landscape change along environmental gradients (Jones et al. 1999). Appendix B
provides a brief summary of the salient characteristics of each of the pilot ecoregions. In the pilot phase,
the remote sensing methodology will be tested and refined, sampling issues will be thoroughly evaluated,
and an approach for assessing the drivers and consequences of land cover change will be finalized.
The second phase will involve the complete analysis of all U.S. ecoregions. First priority will be the
ecoregions in the Mid-Atlantic states, with the remaining ecoregions completed according to a schedule
that will be negotiated with project collaborators.
3.1 Framework Elements
The project design consists of the following framework elements (note that each of these elements
will be explained in more detail in Section 4):
Spatial Framework: The estimates of rates, driving forces, and consequences of land cover change
will be developed for 84 ecoregions defined by Omernik (U.S. Environmental Protection Agency,
1999, a revision of Omernik, 1987). A 20 km by 20 km grid will be applied to each ecoregion to
select a random sample of 400 km2 blocks for land cover analysis. Sample size will be based on the
expected spatial variance of land cover change among grid blocks, a 1 percent margin of error, and
an 0.85 confidence level. Land cover data and change analyses will be developed for each
ecoregion and summarized for the conterminous U.S.
Temporal Framework: Five dates of Landsat MSS and TM data will be obtained, geocoded, and
interpreted for each sample block to provide land cover data on a 6-8 year cycle. The period center-
points for these dates are:
• 1973 Landsat MSS - from the North American Landscape Characterization (NALC) data set
-------
• 1980 Landsat MSS - new acquisitions
• 1986 Landsat MSS - from the NALC data set
• 1992 Landsat MSS - from the NALC data set, and Landsat TM - from the MRLC data set
• 2000 Landsat TM - from the upcoming MRLC 2000 data set
Existing, preprocessed multi-spectral data, or satellite data in which the acquisitions are already
planned, will account for 80-90% of the required imagery.
Land Cover Change Variables: Rather than develop a single set of land cover attributes, a
database of key land cover variables will be generated. Land cover change analysis will be done
for each sample block within each of the 84 ecoregions for the five periods. Three types of land
cover variables will be mapped or derived: (1) general land cover type; (2) landscape biophysical
properties; and (3) landscape pattern.
Major emphasis will be placed on mapping the following general land cover types for each sample
block (class definitions are in Appendix A):
• Urban and Built-Up
• Agriculture (Cropland and Pasture)
• Forests and Woodlands
• Rangeland/Grassland
• Wetland
• Water Bodies
• Snow and Ice
• Natural Barren
• Disturbed or Transitional
To aid in interpreting land cover from the 5 dates of satellite data, and for use in understanding
successional or other gradual land cover transitions, a set of landscape biophysical properties will be
calculated. The following variables and their planned roles are:
• Percent Tree Cover- explains forest regeneration, timber encroachment, or serai stage
• Percent Shadow- explains forest regeneration, timber encroachment, or serai stage
• Unvegetated Fraction - relates to desertification processes or urban intensification
Landscape pattern metrics, describing the number, size, shape, and spatial relationship of land cover
patches (where patch is a contiguous set of pixels assigned to the same land cover type), will be generated
from the general land cover data for each block. The landscape configuration metrics and general land
cover types strongly enhance one another. While it is possible for the relative abundance of land cover
types to remain constant through time, their spatial configuration may change. The landscape metrics will
permit the analysis of the spatial dimension of land cover change, and will also contribute to the
assessment of consequences of land cover change.
-------
3.2 Data Quality Objectives
Our goal is to identify 1% change in general land cover within each ecoregion, at an 85%
confidence level. Key steps that will be taken to reach this goal include:
• Verification of the registration accuracy of the Landsat MSS, TM, and ETM data (objective is sub-
pixel accuracy)
• Use of 1992 baseline MRLC land cover data with 85% overall land cover accuracy
• Validation, to the extent possible based on availability of source materials, of the general land cover
change maps
The methods that will be used to achieve the goal and objectives are explained throughout Section 4.
10
-------
Section 4
Methodology
The technical strategy used in this project has been designed to be extendible to other regions if
essential elements (e.g., suitable ecoregions framework, moderate resolution satellite data) are available.
For example, the scope of this project could be expanded beyond the conterminous U.S. to Alaska and
Hawaii, since compatible ecoregion frames exist. In addition, Omernik has worked with others to
develop an ecoregion map for the Western Hemisphere, so this approach could be applied throughout the
Americas. However, the scope of the current initiative is limited to the conterminous U.S.
The technical methodology will generally involve a series of steps that address sampling, mapping,
analyzing and documenting land cover trends, their causes, and the subsequent consequences. Generally,
analysis will focus on one ecoregion at a time so that the analysts can be immersed in the unique issues
and landscape patterns and conditions of that region. The following sections describe the steps of the
methodology.
4.1 Ecoregion Profiles and Ancillary Databases
At the onset of each ecoregion assessment, geographic profiles and ancillary databases will be
prepared. The geographic profiles will involve documenting population trends over the past 30 years, and
the identification of key social, economic, and environmental issues that occurred during this term. The
profiles will provide the context for the interpretation of the land cover data. Examples of the types of
issues that need to be identified include:
• Economic health of key urban centers, to provide clues about the types of urban change that may
have occurred.
• In areas dominated by public lands, a review of land management policies and practices (i.e.,
logging rates, fire protection) to aid in interpreting land cover change.
• A review of farm legislation in agricultural areas, to provide insights into the likelihood of
agricultural expansion or reduction.
• Climate data for all areas, to evaluate evidence of trends over the past 40 years and to identify
potential subtle changes in land cover condition.
The ancillary databases that will be assembled will vary according to the unique characteristics of
each ecoregion. Databases that will be needed for all regions include key population variables (e.g.,
population density, total population, population dynamics), land ownership, wetlands distributions,
existing land cover maps, and climate data.
4.2 Ecoregion Sampling
Assessment of land cover change at the population level (the entire U.S.) is impractical, both in terms
of cost and processing time. The project objective is to assess change by sampling a portion of the
conterminous U.S. and estimating change from only the sample data. Since our interests lie in assessing
11
-------
patterns, as well as relative abundance, of land cover types, the sample block size must capture the spatial
configuration within and among land cover types. Through investigation of various block sizes, it was
determined that a 20 km by 20 km block is large enough to adequately capture this information and yet is
small enough to allow for rapid analysis and processing.
The occurrence of partial blocks (blocks falling on ecoregion boundaries) poses a problem. Formulas
for estimating change when both full and partial blocks are present in an ecoregion are relatively
straightforward, but accounting for partial blocks can make sample size planning extremely difficult. For
this reason, we have used the simple case of all blocks being complete in order to simplify sample size
planning. The data from a sample block can be used in the analysis of change for more than one
ecoregion, but the sampling requires that it be identified initially with a single stratum. This was done by
initially assigning 20 km blocks to the ecoregion making up the majority of the block.
The primary problem is determining the number of these 20 km by 20 km blocks that are required to
adequately address the objective of estimating land cover change on an ecoregion by ecoregion basis. By
initially assigning each block to a single stratum, we can use a standard sample size formula for simple
random sampling. The planned sample size (Cochran, 1977) for an ecoregion is thus given by:
k = 1 / ( l/k0 + 1/K) where:
k = planned sample size
K = number of 20 km by 20 km blocks in the ecoregion
k0 = (za/m)2 with:
z = a percentile from the standard normal distribution
a = standard deviation of the number of change pixels (in each block) for the
collection of all K blocks
m = margin of error (in pixels per block)
The number of K blocks in each ecoregion is determined when all 20 km blocks covering the
conterminous U.S. are initially assigned to an ecoregion using the majority rule. The z value and the
margin of error were tested empirically to determine acceptable and feasible values. The sample size
equation was shown to be very sensitive to changes in the z value and the margin of error. Some
compromises needed to be made with regard to confidence intervals and margins of error in order to
achieve sample sizes that were feasible. For example, in many cases, over half of the 20 km blocks in an
ecoregion would need to be sampled to provide margins of error of 555 pixels per block (±0.5%) or less.
In order to achieve feasible sample sizes, it was necessary to select a margin of error of 1111 pixels per
block (±1%) and a confidence interval of 85%.
The final piece of information required to calculated sample size for an ecoregion is a a, the standard
deviation of the number of change pixels (in each block) for the collection of all K blocks. In effect, a
measure of the distribution of change within an ecoregion is required. If change is distributed uniformly
across an ecoregion, a would be low, resulting in a lower number of required sample sites. If the change
were clustered in a few locations, a would have a high value and a higher number of sample sites would
be required. Unfortunately, this information requires a prior knowledge of the distribution of change in
an ecoregion, information that this project in part intends to derive. To obtain estimates of a, NOAA
Coastal Change Analysis Program (C-CAP) data were obtained for the Chesapeake area and the San
Francisco area. The C-CAP program uses remotely sensed imagery to determine land cover change
between two dates. The dates for the two data sets were roughly 6 years apart, very comparable to what
this project proposes. The land cover information for the two C-CAP dates were aggregated to match the
land cover classes proposed by this project, and a change image was created. A 20 km grid was overlain
12
-------
on the change image, the number of pixels in each 20 km block were obtained, and a values were
calculated. The procedure to calculate a was repeated for subsets within the C-CAP data sets which
represented ecoregions or portions of ecoregions. Ideally, this information would be available for each
ecoregion prior to calculating sample size, but in lieu of such data, the estimates derived from the C-CAP
data were used as a substitute for the a value. A constant, high estimate of a was used for all ecoregion
sample site planning. This high estimate of a will ensure that ecoregions with great variability in change
pixels between blocks would be sampled adequately, but likely results in oversampling (with regard to
targeted precision marks) for ecoregions with relatively even distributions of change.
The sample size formula
given above is based on a
Gaussian distribution of the
number of change pixels per
block. With 111,111 pixels in
each 20 km X 20 km block, the
number of change pixels per
block should follow a fairly
smooth distribution. However,
the actual distribution is likely
to be skewed right. Once the
project is underway and real
data on the distribution of
change are available, planning
formulas accounting for non-
normality will be explored to
determine how strong an effect
this would have on sample size
planning computation. In
ecoregions where change is rare (e.g., near 0 - 1%), assuming a Gaussian distribution may even result in
the collection of too many samples.
The effect of classification error on sample size planning also needs to be explored. The effect of
map error would be to require a larger sample size. At this stage, it is unclear if formulae are available
which compute sample sizes in the presence of classification error, and if such formulae are available,
estimates of the magnitude of classification error would need to be calculated. This will require
investigation after the project begins, but it is likely the needed increase in the number of blocks will only
be on the order of 10% (around 1 additional block per ecoregion).
Given the effects of a non-normal distribution and of classification error, a small number of additional
sampling blocks may be required above and beyond those called for by the sampling formula outlined
above. Using the high estimate of a and rounding sample size estimates up to the nearest whole integer
will likely ensure an adequate number of sample blocks. If data from the pilot ecoregions indicate the
necessity for more samples in a given ecoregion, additional random samples will be selected. The
formula above will be used to calculate preliminary sample sizes, and a simple random sample will be
obtained for the strata of blocks associated with each ecoregion. This will result in approximately 800
sample blocks across the conterminous U.S. for each of five sample periods (i.e., approximately 4000
individual date-blocks for the entire assessment). The pilot project will allow us to evaluate the sampling
scheme, including issues related to block size, sampling density, and targeted precision marks.
Figure 1. Distribution of 20 km by 20 km sample blocks using the pilot
test sampling parameters.
13
-------
4.3 Satellite Databases
All Landsat MSS and TM scenes covering the sample blocks will be assembled initially as path/row
sets for use in the land cover characterization and change analysis. Coverage for 1973 and 1986 will
come from the NALC archive, the 1992 data will include both the NALC MSS data and the MRLC TM
coverage, and the 2000 data will be ETM scenes acquired for MRLC 2000. Landsat MSS data for the
1980 period will need to be acquired. To the extent possible, the new acquisitions must match the
approximate growing season periods of the 1973, 1986 and 1992 NALC coverage to avoid
misclassification of change due to phenological differences.
The use of cloud-free imagery is important. When scenes with cloud cover are encountered, we will,
whenever possible, replace cloud-covered portions of images with data that is cloud-free. When this is
not feasible, we will treat the cloud-covered portion of the sample as "no data".
The most critical image geometry issue is accurate scene-to-scene registration. NALC MSS and
MRLC TM data are each georeferenced to root mean square errors of 1 pixel or less, but are provided in
different projections. All NALC MSS scenes and all new MSS or ETM acquisitions will be geocoded to
a common Albers equal area map projection, the base projection of the MRLC TM data sets. MRLC TM
data, the majority of the NALC MSS data, and all new data acquisitions will be terrain corrected.
However, approximately one-third of NALC path/rows were processed prior to the implementation of
terrain-correction techniques. It is not anticipated this will cause any major problem, as these early
NALC scenes are primarily located in areas with negligible terrain variability.
The remaining task associated with this step is a relative radiometric calibration of the Landsat
satellite images. The 1992 date will serve as a baseline for both MSS and TM data sets, with 1973, 1980,
and 1986 MSS data radiometrically calibrated to the 1992 NALC MSS imagery and year 2000 ETM data
radiometrically calibrated to 1992 MRLC TM data. Staying with MSS to MSS comparisons and ETM to
TM comparisons eliminates the need for cross-sensor calibration (i.e., MSS to TM). The relative
calibration of the satellite imagery will correct for differences in atmospheric path radiance, detector
calibration, sun angle, earth-Sun distance, atmospheric attenuation, and phase angle conditions.
Radiometric control sets representing temporally invariant features will be used to derive gains and
offsets for a linear transformation. Depending on individual sample block characteristics, one of the
following methods will be used to define control sets: (1) identifying pseudo-invariant bright and dark
targets, (2) defining control sets from brightness-greenness scatterplots that represent bright and dark
pixels with low vegetation content, (3) selecting a control set corresponding to soil line pixels defined in a
brightness-greenness scatterplot, or (4) using automated scattergram-controlled regression.
4.4 Land Cover Biophysical Properties
We will independently focus on the analysis of the changes to state and condition of natural
vegetation. Spectral unmixing will complement sampling strategies for mapping land cover change, and
in particular will facilitate development of quantitative metrics that can be qualitatively assessed with
respect to the phenologic characteristics of vegetation, definitive land cover conversions, and other
potential anthropogenic or climatic influences.
Using spectral unmixing and regression techniques, percent tree cover, percent shadow, and
unvegetated fraction will be calculated. The biophysical composition of the landscape is an integral
component to the assessment of ecoregion health and functional potential. Landsat MSS and TM data for
multiple dates will be used to decompose the landscape into fractional endmembers representing relative
14
-------
proportions of percent tree cover (based on the amount of woody biomass), unvegetated fraction (i.e.,
bare soil, rock, developed structures), and percent shadow. The satellite data will be spectrally unmixed
using methods described by Adams et al. (1993) and Roberts et al. (1993) to determine changes in the
relative proportions of these three biophysical variables.
Our use of endmembers correlating to the three biophysical properties is for use in interpreting or
identifying gradual or subtle changes in the condition of the landscape that may be important in
determining the driving forces of change (i.e., climate-induced, land management policies, etc.). The
calculated variables will be used in a qualitative fashion. This is necessary because of the challenge in
quantifying the relationship between endmembers and actual landscape conditions when image quality
and phenological conditions are inconsistent. In order to use the results in a quantitative way, an
extensive and impractical amount of calibration data would be needed. In cases where we will investigate
shifts in environmental gradients (see the following discussion of the Madrean Archipelogo, Section
4.4.1), we will use large scale aerial photography to calibrate the endmembers.
The use of biophysical characteristics of land cover may provide useful information on the condition
of land cover. However, there is little operational evidence of the utility and consistency of this
information. We consider this element of the project to be experimental and will carefully evaluate the
quality and meaning of the information derived using spectral unmixing at the end of the pilot phase. If
the pilot phase results are determined to be useful, we will continue generating and interpreting these data
for the remainder of the project.
4.4.1 Assessment of Environmental Gradients
An parallel investigation of the role of the NALC MSS and MRLC TM data will be carried out in the
Madrean Archipelago ecoregion in southeastern Arizona. This ecoregion has both steep and gradual
environmental gradients. The lower elevations have transitions between desert grassland basins and arid
shrublands. Scattered mountain islands with elevations over 2500m have vegetation with distinct
zonation. Transitions range from arid scrub vegetation to chaparral, oak woodlands, and needleleaf
forests at the highest elevations. The combination of both steep and gradual gradients provide an
opportunity to investigate the ability to detect subtle environmental change along both gradients using the
1972-1992 NALC data, and 2000 Landsat ETM scenes.
The nature of vegetation changes that may be occurring will be difficult to detect using general land
cover maps since the change is most likely manifested through changes in vegetation condition. Thus, we
will test the potential for using biophysical measures (see Section 4.4) to determine shifts in vegetation
characteristics. In particular, we will study changes in the unvegetated fraction to see if there are changes
in the positions of the grassland basins from the more sparsely vegetated shrub regions, and will use both
unvegetated fraction and percent tree cover to identify subtle changes along the mountain island
topographic gradient. To improve our ability to accurately map percent tree cover and the unvegetated
fraction, we will use large scale aerial photos (e.g., BLM and USFS resource photography) as calibration
sources to estimate percent cover. The analysis will involve calculation and calibration of biophysical
characteristics for bisecting transects and sample blocks, comparison of the metrics for each date to
identify anomalies, testing of the statistical significance of the change, and an assessment of climate
trends, land management practices (to the extent possible), and other resource records to determine the
meaning of the anomalies. A scientific paper summarizing the methods, results, and potential for using
NALC data for monitoring change along environmental gradients will be prepared at the end of the
investigation.
4.5 Land Cover Change Analysis
15
-------
The primary image analysis task of this project is determining general land cover transformations
over the past 30 years. The land cover data sets are also needed for the calculation of landscape metrics
that provide measures of the changes in landscape configuration. To describe land cover change within
and between ecoregions, information regarding land cover classes and conversions between land cover
classes is required.
4.5.1 Land Cover Change Mapping
There are numerous approaches to characterizing land cover change, and each has a set of strengths
and weaknesses. Because no single approach is optimal for all types of landscapes and land cover
features, we plan to use a hybrid of available approaches; specific methods will depend on the
characteristics of specific ecoregions. We plan to use a combination of simultaneous analysis techniques
(e.g., CVA, image differencing), comparative analysis of independently produced classifications and
manual interpretations.
Processing will begin with the 1992 MRLC land cover database. This database, with classes
aggregated to general land cover types, will serve as the land cover baseline. Land cover for the 1973,
1980, and 1986 periods can be back-classified using 1992 as the template, and changes can be forward
classified for the 2000 data. One advantage of using 1992 MRLC land cover data is that it will have
known classification accuracy (preliminary results show that at the level of generalization needed for this
project, the overall accuracy is about 85-90%), and thus will provide an excellent reference data set.
Additionally, rates of change numbers derived from the trends assessments, combined with the complete
1992 national land cover classification, will provide a detailed enumeration of actual land cover that can
be used to estimate carbon stocks. The rates of land change from this period can be used to assess
progress in balancing carbon sources and sinks. Since the baseline period for the framework convention
on climate change is 1990, the statistics from this period will be relevant for a national carbon assessment.
One of the primary approaches that will be employed is an enhanced change vector approach.
Beginning with the 1992 and 1986 MSS imagery, we plan to use the change vector analysis (CVA) tools
created by Dwyer et al. (1997) to produce a "first-cut" mask depicting areas of possible change. These
tools will provide information regarding change in the brightness/greenness plane, with an empirically
derived threshold being selected to assign pixels to a "no change" or "possible change" category. Using
manual interpretation, each contiguous group of "possible change" pixels will be analyzed. Groups of
pixels not representing real land cover change will be manually coded to a null value, while groups of
pixels representing a specific type of land cover change will be manually coded to the representative class
value. A change mask is thus created which represents specific land cover changes from 1992 to 1986.
This change mask is then applied to the 1992 MRLC land cover image to create a 1986 land cover image.
The process is then repeated to back-classify for the earlier dates, and forward classify to the 2000 date.
Although the use of data from different sensors poses a serious challenge to many change analyses,
we plan to use post-classification comparisons to address this problem. The Landsat TM and MSS will
both be mapped to common thematic references, using an approach similar to the one outlined above.
Comparisons between land cover data for each of the dates will thus provide a means for equating results.
It is recognized that post-classification comparison has been criticized because it has the tendency to
compound errors found in the two individual classifications. We believe three factors will enable us to
obtain highly accurate change information from the individual classifications:
16
-------
1. We plan to rely heavily on manual interpretations to derive change information. It is
recognized that automated approaches alone can not approach the level of accuracy that can
be obtained by incorporating manual interpretation.
2. It is believed the relatively small size of the individual sample blocks will allow for accurate
land cover mapping.
3. Stratifying by ecoregion, coupled with the relatively small sample area, will result in less land
cover heterogeneity being addressed within an analysis. This should improve our ability to
identify important distinctions.
4.5.2 Accuracy Assessment/Validation of Results
While there is much value in assessing the accuracy of land cover and land cover change products, it
is extremely difficult to implement a consistent, comprehensive, quantitative accuracy assessment for
such a large area change database. One of the primary difficulties with the accuracy assessment of
change products is acquiring an adequate database of historical reference materials. Contemporaneous
(same year, same season) historical aerial photography is the preferred source of historical reference
information, but it is highly improbable that such material will consistently be available. It will likely be
necessary to incorporate a variety of historical reference materials, which will differ from ecoregion to
ecoregion. These may include historical aerial photography, satellite imagery, other local change analysis
studies, or other data sources.
Even if adequate historical reference materials can be obtained, classical accuracy assessment
procedures cannot be easily applied to the results of this project. While the error matrix can be adapted
for use in land cover change analysis studies, it becomes impractical when dealing with the numbers of
change categories that this project will include, and it is not applicable to the assessment of trends. Given
adequate reference material, a correlation approach may be used as an additional approach to analyze
trends as it can quantify the agreement between disparate references and mapped data. The use of
correlation as a validation tool has the added advantage of applicability to any of our measured variables,
given adequate reference material. However, it does not provide the necessary unbiased accuracy
information. Thus, a standard accuracy assessment of the individual dates of land cover classifications is
our first priority.
Comparisons (consistency checks) between results from small, localized studies and our efforts will
provide an additional means of assessing accuracy. While not as desirable as comprehensive, quantitative
accuracy assessment procedures, consistency checks can provide strong evidence that our procedures
produce meaningful information. Direct comparison of our product against a validated localized study
provides indirect evidence of our product accuracy.
The validation of such a large area change database has never been attempted, and more research is
required to finalize our validation approach. Our first step will be to investigate the availability of
historical aerial photography and other data sources. The choice of validation methodologies will depend
on the type of historical reference material available; thus, validation methodologies will vary among
ecoregions. Correlation procedures, accuracy assessment of individual date classifications, and
consistency will be used in combination to validate our change analyses products.
4.6 Landscape Configuration Metrics
We will calculate a suite of standard metrics that describe the number, size, shape, and spatial
relationship of patches of land cover types in each sample block and time period. Calculation of these
17
-------
metrics is completely operational through GIS techniques and landscape metrics software (e.g.,
FRAGSTATS), and is appropriate for national and regional assessments. We will include an analysis
similar to that shown in Figure 2 for assessing and interpreting changes in patch size and number across
time and across ecoregions. A number of metrics have been developed to describe and quantify elements
of patch shape complexity and spatial configuration relative to other patch types; however, it is not clear
which will prove to be the most informative and interpretable over large areas. Therefore, some
experimentation during the pilot phase of this project will help us determine the set of metrics that will be
applied for the second phase of the project.
100
A
Frequency
Time 3
Ecoregion A
Ecoregion B
Ecoregion C
I.
I,
100
>
Frequency
Time 2
(
k
)
1...
I
100
A
Frequency
Time 1
Small-
Patch Size
->- Large
Small.
Patch Size
Large
Small •
Patch Size
->- Large
Figure 2. An example temporal comparison of patch size and frequency for a land cover type
across three ecoregions. Columns represent ecoregions and rows correspond to time
(where bottom to top row = past to present). The x-axis represents patch size class
and the y-axis indicates number of patches. In Ecoregion A, the number of patches
has increased dramatically overtime, such as might happen where woodlands have
colonized prairies as a result of wildfire suppression. Ecoregion B shows a case where
large, spatially continuous patches have become highly fragmented overtime, as has
happened with some forests in the U.S. In Ecoregion C, patch size and frequency
have remained fairly constant through time.
Associated with land cover patch analyses are issues about minimum mapping unit, edge effects, and
accuracy of output:
• Minimum Mapping Unit - We propose a minimum mapping unit of two hectares for calculation of
landscape metrics. This unit is resolvable on both MSS and TM data and should be useful for many
scientific and management applications.
• Edge Effects - Landscape patches will be truncated along the edges of sampling blocks.
Underway is an experiment to quantify the effects of artificial edges imposed by the sampling
18
-------
blocks. Results from this experiment will help us develop a protocol for handling "edge-affected"
patches. For example, a given buffer distance from a block edge may provide us with sufficient
information on the full patch characteristics for, say, 70% of the patches along a block edge.
• Patch Accuracy - Patch accuracy is not the same as land cover classification accuracy and cannot
inherently be determined by the same assessment methods. Very little research has been done on
this topic (See Hess, 1994; Hess and Bay, 1997; Wickham et al. 1997). For the current study, we
do not propose to report on patch accuracy for each ecoregion; rather, we have designed a
complementary study to determine whether general rules are applicable regarding the effects of
misclassification errors on patches. This complementary study will use land cover and accuracy
assessment data obtained from another project, the MRLC land cover classification project. The
benefits of the MRLC data are that:
1. The land cover classification scheme nests within the one proposed for the Trends Project;
2. the classification was derived from TM data;
3. the accuracy assessment was designed, reviewed, and implemented in a statistically rigorous
fashion;
4. data are already available for large portions of the U.S. and will be available for the entire
conterminous U.S. within the coming year; and
5. accuracy data have been obtained in a sufficiently dense, geographically well-distributed
frequency for providing information on spatial distributions of different types of
classification error. From these MRLC data we will investigate the effects of a range of
misclassification rates (spatially distributed in a manner indicated by the accuracy assessment
data) for different patch metrics (corresponding with those selected for the Trends Project)
and across different ecoregions. We will investigate which relationships between land cover
classification errors and patch effects are consistent among ecoregions and which are specific
to within-region characteristics.
Analysis of patch accuracy will be addressed at two levels: the individual patch and the patch
"population." At the population level, we will compare patch size frequency distributions (as in the
histograms in Figure 2) for given land cover types across a range of land cover misclassification rates.
This analysis ignores the geographic location of individual patches and focuses on determining error rate
thresholds for detectable differences in patch size frequency for an ecoregion. At the individual patch
level, assessment will focus on the effects of classification errors on individual patches. Here, we are
interested in thresholds of misclassification rates that begin to decompose single patches into multiple
patches or begin to aggregate patches. Since the type and spatial distribution of classification errors will
vary from ecoregion to ecoregion, our interpretation of the effects of errors on individual patches will
vary according to within-region characteristics.
4.7 Ecoregion Trends Analysis
We are primarily interested in analyzing temporal series of our land cover variables to address
questions relating to: (1) the predominant types of land cover conversions occurring within each
ecoregion, (2) the estimated rates of change for these conversions, and (3) whether the types and rates of
change are constant or variable across time. A series of interpretive products, such as change vector maps
(Dwyer et al. 1997) and post-classification comparisons, will be generated toward these ends.
19
-------
• Land Cover Types - Global summary statistics will be calculated across blocks for each
ecoregion for each time period in order to assess regionwide changes in relative abundance. A
change matrix will be constructed for each (temporally) consecutive pair of land cover layers that
will indicate the types of conversions that have occurred. A "conversion type" map will show the
geographic distribution of conversion types. We will look for spatial correlations between
conversion types and selected environmental factors, such as terrain characteristics, proximity to
urban development, economic conditions, etc., in order to improve our understanding of potential
drivers of change. From these foundations we can explore the feasibility of developing pixel-based
transition probabilities so that predictive maps can be compiled for future scenarios of land cover.
• Biophysical Attributes - We will examine changes in each attribute by land cover type per
ecoregion. Canopy density, percent shadow, and unvegetated fraction can be represented
graphically as maps, boxplots, histograms, and/or cumulative frequency diagrams for each time
period. Additionally, the amount of change (i.e., image differencing) between time periods can also
be represented in these ways. General rate of change across the five time periods can be estimated
using the Sen slope estimator (Sen, 1968 in Gilbert, 1987). Estimate rates of change can be
graphically displayed on maps in order to depict their spatial distribution and to suggest correlations
with other environmental variables. Since data are limited to 4 time-steps, determination of
statistical significance regarding trends in biophysical attributes is problematic. However, the
degree of temporal constancy or variability in change rates can still be addressed if project analysts
define thresholds that are ecologically or economically meaningful.
• Landscape Metrics - The number and sizes of patches for individual cover types will be
compared overtime within and across ecoregions. Patch distribution can be graphically displayed
as histograms (e.g., see Figure 2). As with the analysis of changes in land cover types, we will look
for spatial correlation changes in patch characteristics and other environmental factors. Shape and
spatial configuration metrics tend to be continuous variables, and as such, can be handled similarly
to the analyses of biophysical attributes.
4.8 Assessment of Land Cover Change Drivers and Consequences
In a separate research initiative, we will work toward the identification of the most significant driving
forces of land cover change occurring in each ecoregion. The local human activities that express the
driving forces in each ecoreigon will be determined and will be assessed by measuring the rates and types
of change, landscape pattern changes, and other relevant sources of data (e.g., demographic profiles,
economic reviews, public policies, etc.). Examples of the issues that will be addressed include:
• What are the likely driving forces of local land cover transformation?
• What are the connections between land cover change and changes in (the driving forces) economic,
social, and environmental conditions?
• What are the local human activities (proximate sources of change) that are altering land cover
transformation?
• What are the likely consequences of ecoregion land cover change at the regional, national, and
global scales?
• What are the regional responses to land cover change?
20
-------
A geographic framework and methodology for assessing the causes and consequences of change will
be developed during the first year of the project. The implementation of the methodology may be more
appropriately done through a consortium of universities that each focus on a particular part of the country.
We will pursue implementations options following the completion of the assessment strategy.
4.9 National Synthesis
The ecoregion results will be summarized to identify:
a. National rates of land cover change;
b. the primary types of conversion occurring;
c. the regions having the most dynamic land cover change rates; and
d. the periods over the past 30 years in which land cover transformations have been most
dynamic. The methods used in the ecoregions trends analyses will be used to assess overall
rates of land cover change. Ecoregional comparisons will be made to understand the
connections between driving forces and impacts associated with certain landscapes. In
addition, other regional summaries (i.e., Great Plains, Pacific Northwest, Desert Southwest,
etc.) may be produced to provide different scales and perspectives of analysis.
21
-------
Section 5
Planned Deliverables and Timelines
Because this is a research project with little precedent, there is some uncertainty in the timelines for
the various project staff. In addition, the schedule for completion is dependent on overall funding and
staffing. Thus, the timelines needed to complete the planned deliverables are tentative and will be refined
annually. Our goal is to complete the analysis of the conterminous U.S. within 4 years.
Key output from the project are reports that deal with the four objectives stated in Section 1.2. The
form of most reports will be peer-reviewed scientific papers. The types of reports that will be produced
will generally provide:
• Methodological Summaries - Technical summaries of the methods used in the analysis, such as
the overall design, change detection methods, and biophysical parameterization techniques, will be
prepared for submission to remote sensing journals. Papers on the strategies for analyzing the
causes and consequences of change will be written for geographic and ecological applications
journals.
• Topical Assessments - Papers that address key issues investigated during the research will be
sent to appropriate journals, defined based on the topic. Types of topics that will be analyzed
include assessment of change along environmental gradients, impacts of key driving forces (e.g.,
Conservation Research Program), and other problems.
• Geographic Analyses - Summaries of land cover change findings for individual ecoregions will
be produced as each ecoregion is completed. At a minimum, these will be released as USGS open-
file reports. Ultimately, geographic treatments of the rates, drivers, and consequences of land cover
change will be summarized for publication in a USGS-produced atlas of contemporary land cover
change, which will include maps and summaries of ecoregion and national land cover trends
characteristics.
In addition to documentation, all project data sets, including sources and results, will be released via
an FTP site accessible through the World Wide Web. All data will be documented according to Federal
Geographic Data Committee metadata standards. Management of the data will ultimately become an
operational component of the USGS Land Cover Characterization Program. Thus, data produced through
this project will become part of the long-term land cover archive that currently includes the MRLC,
NALC, and global AVHRR and land cover data sets.
While most results will be geared to a scientific audience, efforts will also be made to communicate
results to a broader audience. A project web site will be established to provide results, summaries of
lessons learned over the course of the project, and access to all data sets produced by the project team. In
addition, the web site will provide a means for sharing information on the application of project results.
22
-------
As stated previously, the first year will involve developing, testing, and refining project methods in
five pilot study areas (see Section 3). At the end of the first year (calendar year 1999), we will deliver the
following:
• Geographic summaries of land cover trends for the five pilot ecoregions for the early 1970's
through 1992 period (note that the 2000 analysis cannot be completed until late-2000 because of the
need to acquire 2000 growing season Landsat 7 ETM data).
• Web site on project activities
In the second year (calendar year 2000), we intend to complete the analysis of an additional 30
ecoregions. The specific ecoregions that will be completed will include those within the Mid-Atlantic
states and additional ecoregions selected through discussion with project partners. We will also deliver:
• Manuscript on remote sensing aspects of the methodology
• Methodology and implementation plan for assessing the drivers and consequences of land cover
change
• Geographic summaries of land covertrends for the early 1970's through 1992 period for 30
additional ecoregions.
• Analysis of land cover change for the 1992 to 2000 period for the pilot regions and remaining
ecoregions bisecting the Mid-Atlantic regions.
• Topical assessment of subtle environmental change in the Madrean Archipelago
• Topical report on the contemporary land cover history of the Mid-Atlantic region
• Others, TBD
Thirty ecoregions will be completed during the third year (2001). Again, the specific ecoregions will
be determined in consultation with project collaborators. We plan to deliver:
• Analysis of land covertrends for the early 1970's through 2000 for the first 60 additional
ecoregions.
• Others, TBD
Finally, during the fourth year (2002), we will complete the remaining 20 ecoregions. We will also
complete an assessment of the merits and feasibility for converting the research project to an operational
activity within the USGS Land Cover Characterization Program. Deliverable s for the last year of the
project are expected to be:
• Analysis of land covertrends for the early 1970's through 2000 for the remaining ecoregions.
• Geographic assessment of land cover trends, their drivers, and the subsequent consequences for the
conterminous U.S.
• Atlas of conterminous U.S. land cover change
• Manuscript on national land cover trends
• Others, TBD
23
-------
Section 6
Management Plan
The project will involve a core team comprised of Tom Loveland (Project Leader), Terry Sohl
(geographic analysis, sampling), Kristi Sayler (geographic analysis and spectral unmixing), Jim
Vogelmann (change analysis, MRLC 2000, and Landsat 7 issues), Alisa Gallant (ecoregion
characteristics, landscape metrics, trends analysis), John Dwyer (change vector methods, spectral
unmixing), and Greg Zylstra (radiometry, analytic methods) will all contribute. In the second year, a
postdoctoral position will be added to the team. Resumes of key personnel are presented in Appendix D.
Because of the importance of statistical design, including sampling methods, accuracy assessment,
and trends analysis, Dr. Steve Stehman (State University of New York, Syracuse) will serve as a statistics
consultant throughout the project. In addition to his role in researching and advising the project team on
key statistical design problems, he will meet with the team for approximately one week twice each year.
During FY2000, Dr. Darrell Napton (South Dakota State University) will spend a sabbatical at the
EROS Data Center. He will focus on two issues: developing outlines for the ecoregion profiles and
developing and testing a strategy for assessing the drivers and consequences of land use and land cover
change.
In addition, we will establish collaborative relationships with scientists within the EPA Landscape
Ecology Branch. This relationship will have two aspects. First, it will provide a means for prioritizing
analysis, applying results, and gaining comments on project strengths and weaknesses. Second, because
of their expertise in the application and interpretation of landscape metrics, we will solicit their direct
input on the landscape metrics component of this project. Once this plan is approved, we will provide
monthly status reports to EPA project monitors and interested technical staff. These reports will be
distributed via email.
Finally, once each year, we will hold an "all-hands" project review where we will discuss all project
findings, identify problems and solutions, discuss opportunities, and prepare a work plan for the
following year. All work plans, research plans, proposals for expansion, and technical or scientific
reports will be provided to all collaborators for review before being released to a broader audience.
24
-------
Section 7
References
Adams, J.B., D.E. Sabol, V. Kapos, R.A. Filho, D.A. Roberts, M.O. Smith, and A.R. Gillespie, 1995.
Classification of multispectral images based on fractions of endmembers: Application to land-cover
change in the Brazilian Amazon, Remote Sensing of Environment 52: 137-154.
Baker, L.A., 1990. Current status of surface water acid-base chemistry. NAPAP SOS/T Report 9. In:
National Acid Precipitation Assessment Program, Acidic Deposition: State of Science and
Technology, Volume II.
Bateson, A. and B. Curtiss, 1996. A method for manual endmember selection and spectral unmixing,
Remote Sensing of Environment 55: 229-243.
Bryne, G.F., P.P. Crapper, and K.K. Mayo, 1980. Monitoring land-cover change by principal component
analysis of multitemporal Landsat data, Remote Sensing of Environment 10: 175-184.
Caselles, V., and M.J. Lopez Garcia, 1989. An alternative simple approach to estimate atmospheric
correction in multitemporal studies, International Journal of Remote Sensing 10(6): 1127-1134.
Cochran, W.G., 1977. Sampling Techniques (3rd ed.), Wiley, NY.
Cochrane, M.A. and C.M. Souza, 1998. Linear mixture model classification of burned forests in the
Eastern Amazon, International Journal Remote Sensing 19(17): 3433-3440.
Crist, E.P., and R.C. Cicone, 1984. A physically based transformation of Thematic Mapper data - The TM
Tasseled Cap, IEEE Transactions on Geoscience and Remote Sensing 22: 256-263.
Dimyati, M., K. Mizuno, S. Kobayashi, and T. Kitamura, 1996. An analysis of land use/cover change
using the combination of MSS Landsat and land use map - A case study in Yogyakarta, Indonesia,
International Journal of Remote Sensing 17(5): 931-944.
Dobson, J.E., 1992. Global challenges demand more precision, better generalization, GIS World, October
1992: 68-69.
Dobson, J., and E. Bright, 1994. Large-area change analysis: The Coastwatch Change Analysis Project
(C-CAP), Proceedings of Pecora 12, Sioux Falls, SD, August 24-26, 1993, American Society for
Photogrammetry and Remote Sensing, Bethesda, MD: pp. 73-81.
Dobson, J.E., E.A. Bright, R.L. Ferguson, D.W. Field, L.L. Wood, K.D. Haddad, H. Iredale III, J.R.
Jensen, V.V. Klemas, R.J. Orth, and J.P. Thomas, 1995. NOAA Coastal Change Analysis Program
(C-CAP): Guidance for Regional Implementation, NOAA Technical Report NMFS 123, Seattle, WA
92 p.
25
-------
Dwyer, J., K. Sayler, and G. Zylstra, 1997. Landsat pathfinder data sets for landscape change analysis,
Proceedings, International Geoscience and Remote Sensing Symposium, Lincoln, Nebraska, May 17-
31, 1996: pp. 547-550.
Elvidge, C.D., D. Yuan, R. Weerackoon, and R. Lunetta, 1995. Relative radiometric normalization of
Landsat multispectral scanner (MSS) data using an automatic scattergram-controlled regression,
Photogrammetric Engineering and Remote Sensing 61: 1255-1260.
Foschi, P.G., and K.D. Smith, 1997. Detecting subpixel woody vegetation in digital imagery using two
artificial intelligence approaches, Photogrammetric Engineering and Remote Sensing 63(5): 493-500.
Gilbert, R.O., 1987. Statistical methods for environmental pollution monitoring, Van Nostrand
Reinhold, New York. 320 p.
Gordon, S.I., 1980. Utilizing Landsat imagery to monitor land-use change: A case study in Ohio, Remote
Sensing of Environment 9: 189-196.
Hall, F.G., D.E. Strebel, J.E. Nickeson, and S.J. Goetz, 1991. Radiometric rectification: Toward a
common radiometric response among multidate, multisensor images, Remote Sensing of Environment
35:11-27.
Heiskary, S.A., C.B. Wilson, and D.P. Larsen, 1987. Analysis of regional patterns in lake water quality:
Using ecoregions for lake management in Minnesota, Lake and Reservoir Management 3: 337-344.
Hess, G. 1994. Pattern and error in landscape ecology: A commentary. Landscape Ecology 9(1):3-5.
Hess, G. R. and J.M. Bay. 1997. Generating confidence intervals for composition-based landscape
indexes. Landscape Ecology 12(5):309-320.
Henderson, B.M., and S.J. Walsh, 1995. Plowed, paved, or in succession: Land-cover change on the
North Carolina Piedmont, Southeastern Geographer 35(2): 132-149.
Howarth, P.J., and G.M. Wickware, 1981. Procedures for change detection using Landsat digital data,
International Journal of Remote Sensing 2(3): 277-291.
Jensen, J.R., and D.L. Toll, 1983. Detecting residential land-use development at the urban fringe,
Photogrammetric Engineering and Remote Sensing 48(4): 629-643.
Johnston, D., and R. Haas, 1985. Change detection in rangeland environments using Landsat MSS data -
A quantitative approach, Pecora 10 Proceedings: Remote Sensing in Forest and Range Management,
Colorado State University, Fort Collins, Colorado, August 20-22: pp. 60-188.
Jones, K.B., L.R. Williams, A.M. Pitchford, T.E. Slonecker, J.D. Wickham, RV. O'Neill, D. Garofalo,
and W.G. Kepner, 1999. A national assessment of landscape change and impacts to aquatic
resources: a 10-year strategic plan for the Landscape Sciences Program, U.S. Environmental
Protection Agency, National Exposure Research Laboratory, Las Vegas, NV 73 p.
Kaufmann, P.R., A.T. Herlihy, M.E. Mitch, J.J. Messer, and W.S. Overton, 1991. Stream chemistry in
the eastern United States. 1. Synoptic survey design, acid-base status, and regional patterns. Water
Resources Research 27: 611-627.
26
-------
Kauth, R.J., and G.S. Thomas, 1976. The tasseled cap - A graphic description of the spectral-temporal
development of agricultural crops as seen by Landsat, Proceedings of the Symposium on Machine
Processing of Remotely Sensed Data, LARS, Purdue University, West Lafayette, Indiana: pp. 4b41-
4b51.
Kumar, M., E. Goossens, and R. Goossens, 1993. Assessment of sand dune change detection in Rajasthan
(Thar) Desert, India, International Journal of Remote Sensing 14(9): 1689-1703.
Lambin, E.F., and A.H. Strahler, 1994. Indicators of land-cover change for change-vector analysis in
multitemporal space at coarse spatial scales, International Journal of Remote Sensing 15: 2099-2119.
Larsen, D.P., D.R. Dudley, and R.M. Hughes, 1988. A regional approach to assess attainable water
quality: An Ohio case study, Journal of Soil and Water Conservation 43:171-176.
Linthurst, R.A., D.H. Landers, J.M. Eilers, D.F. Brakke, W.S. Overton, E.P. Meirer, and R.E. Crowe,
1986. Characteristics of lakes in the eastern United States. Vol.1: Population descriptions and
physicochemical relationships. EPA/600/4-86-007a. USEPA, Wash., D.C. 136 p.
Loveland, T.R., J.W. Merchant, D.O. Ohlen, and J.F. Brown, 1991. Development of a land-cover
characteristics database for the conterminous U.S., Photogrammetric Engineering and Remote
Sensing 57: 1453-1463.
Lowrance, R.R., R. Leonard, and J. Sheridan, 1985. Managing riparian ecosystems to control non-point
pollution. Journal of Soil and Water Conservation 40: 87-91.
Lunetta, R.S., J.G. Lyon, B. Guindon, and C.D. Elvidge, 1998. North America landscape characterization
data set development and data fusion issues, Photogrammetric Engineering and Remote Sensing
64(8): 821-829.
Macleod, R.D., and R.G. Congalton, 1998. A quantitative comparison of change-detection algorithms for
monitoring eelgrass from remotely sensed data, Photogrammetric Engineering and Remote Sensing
64(3): 207-216.
Massart, M., M. Petillon, and E. Wolff, 1995. The impact of an agricultural development project on a
tropical forest environment: The case of Shaba (Zaire), Photogrammetric Engineering and Remote
Sensing 61(9): 1153-1158.
Ojima, D.S., K.A. Galvin, and B.L. Turner II, 1994. The global impact of land-use change, BioScience
44(5): 300-304.
Omernik, J.M., 1987. Ecoregions of the conterminous United States, Annals of the Association of
American Geographers 77: 118-125.
Peplies, R.W. and R.B. Honea, 1992. Some classic regional models in relation to global change studies,
88th Annual Meeting, Association of American Geographers (Unpublished), San Diego, California
24 p.
Ramsey, R.D., A. Falconer, and J.R. Jensen. 1995. The relationship between NOAA-AVHRRNDVI and
ecoregions in Utah, Photogrammetric Engineering and Remote Sensing 53: 188-198.
27
-------
Ribed, P.S., and A.M. Lopez, 1995. Monitoring burnt areas by principal components analysis of multi-
temporal TM data, International Journal of Remote Sensing 16(9): 1577-1587.
Riebsame, W.E., W.J. Parton, K.A. Galvin, I.C. Burke, L. Bohren, R. Young, and E. Knop, 1994.
Integrated modeling of land use and cover change, BioScience 44(5): 350-356.
Riitters, K.H., R.V. O'Neill, C.T. Hunsaker, J.D. Wickham, D.H. Yankee, S.P. Timmins, K.B. Jones, and
B.L. Jackson, 1995. A factor analysis of landscape pattern and structure metrics, Landscape Ecology
10(1): 23-39.
Roberts, D.A., J.B. Adams, and M.O. Smith, 1993. Green vegetation, nonphotosynthetic vegetation, and
soil in AVIRIS data, Remote Sensing of Environment 14: 255-269.
Schott, J.R., C. Salvaggio, and W. J. Volchok, 1988. Radiometric scene normalization using pseudo-
invariant features, Remote Sensing of Environment 26: 1-16.
Sen, P.K., 1968. Estimates of the regression coefficient based on Kendall's tau, Journal of the American
Statistical Association 63: 1379-1389.
Singh, A., 1989. Digital change detection techniques using remotely sensed data, International Journal of
Remote Sensing 10(6): 989-1003.
Skole, D. and C. Tucker, 1993. Tropical deforestation and habitat fragmentation in the Amazon: satellite
data from 1978 to 1988, Science 260: 1905-1910.
Sohl, T., 1999. Change analysis in the United Arab Emirates: An investigation of techniques,
Photogrammetric Engineering and Remote Sensing 65(4): 475-484.
Stehman, S., 1997. Selecting and interpreting measures of thematic classification accuracy, Remote
Sensing of Environment 62: 77-89.
Stow, D.A., L.R. Tinney, and J.E. Estes, 1980. Deriving land use/land cover change statistics from
Landsat: A study of prime agricultural land, in Proceedings of the 14th International Symposium on
Remote Sensing of Environment, Ann Arbor, MI, Environmental Research Institute of Michigan,
Ann Arbor, MI: pp. 1227-1327.
Tompkins, S., J.F. Mustard, C.M. Pieters, and D.W. Forsyth, 1997. Optimization of endmembers for
spectral mixture analysis, Remote Sensing of Environment 59: 472-489.
Turner, B.L. II and W.B. Meyer, 1991. Land use and land cover in global environmental change:
Considerations for study, International Social Science Journal 130: 669-677'.
Turner, B.L. II, D. Skole, S. Sanderson, G. Fischer, L. Fresco, and R. Leemans, 1995. Land-Use and
Land-Cover Change: Science/Research Plan. IGBP Report No. 35, HDP Report No. 7, International
Geosphere-Biosphere Programme, Stockholm, Sweden 132 p.
Turner, M.G., 1990. Landscape changes in nine rural counties in Georgia, Photogrammetric Engineering
and Remote Sensing 56(3): 379-386.
28
-------
U.S. Environmental Protection Agency, 1999. Level III ecoregions of the continental United States, U.S.
Environmental Protection Agency - National Health and Environmental Effects Research Laboratory,
Corvallis, Oregon.
Vogelmann, J.E., 1988. Detection of forest change in the Green Mountains of Vermont using
Multispectral Scanner data, International Journal of Remote Sensing 9(7): 1187-1200.
Vogelmann, J.E., T. Sohl, and S.M. Howard, 1998. Regional characterization of land cover using
multiple sources of data, Photogrammetric Engineering and Remote Sensing 64(1): 45-57.
Weismiller, R.A., S.J. Kristof, D.K. Scholz, P.E. Anuta, and S.A. Momin, 1977. Change detection in
coastal zone environments, Photogrammetric Engineering and Remote Sensing 43(12): 1533-1539.
Wickham, J.D., RV. 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(4)397-402.
Wilcock, D. and A. Cooper, 1993. Monitoring losses of semi-natural vegetation to agricultural grassland
from satellite imagery in the Antrim Coast and Glens AONB, Northern Ireland, Journal of
Environmental Management 38: 57-169.
29
-------
Appendix A
Land Cover Definitions
The following are the general land cover definitions that will be used in the project. To the extent
possible, the definitions are based on the original Anderson et al. (1976) level I definitions so that land
cover data developed through this project are consistent with those produced through other programs and
projects. Note that a minimum mapping unit of 4 acres will be used to determine land cover.
Urban and Built-Up - Areas of intensive use with much of the land covered with structures (e.g., high
density residential, commercial, industrial, transportation, mining, confined livestock operations), or
less intensive uses where the land cover matrix includes both vegetation and structures (e.g., low
density residential, recreational facilities, cemeteries, etc.), including any land functionally attached to
the urban or built-up activity.
Agriculture (Cropland and Pasture) - Land in either a vegetated or unvegetated state used for the
production of food and fiber. Note that forest plantations are considered as forests or woodlands
regardless of the use of the wood products.
Forests and Woodlands - Tree-covered land where the tree cover density is greater than 10%. Note
that cleared forest land (i.e., clear-cuts) will be mapped according to current cover (e.g., disturbed or
transitional, rangeland/grassland).
Rangeland/Grassland-Land predominately covered with grasses, forbs, or shrubs. The vegetated
cover must comprise at least 10% of the area.
Wetland - Lands where water saturation is the determining factor in soil characteristics, vegetation types,
and animal communities. Wetlands are comprised of water and vegetated cover.
Water Bodies - Areas persistently covered with water, such as streams, canals, lakes, reservoirs, bays, or
oceans.
Snow and Ice - Land where the accumulation of snow and ice does not completely melt during the
summer period.
Natural Barren - Land comprised of natural occurrences of soils, sand, or rocks where less than 10% of
the area is vegetated.
Disturbed or Transitional - Land in an altered unvegetated state which is in transition from one cover
type to another.
30
-------
Appendix B
Description of Pilot Ecoregions
79 Madrean Archipelago
Land use: Mostly grazing. Some irrigated cropland (cotton, corn, alfalfa, sm. grains,
other).
Concerns: Grazing practices can lead to invasion of brushy species and local gully
erosion. Irrigated agriculture associated with declining water tables and a short
supply or irrigation water.
Terrain: Elevations from 800-1400 m in basins, 1500->2500 m in mountains. SE-NW
trending mountain ranges separated by relatively smooth valleys.
Climate: 275-375 mm ppt. annually (up to 900 mm in mtns.). More than half ppt. is in
summer.
Water: All streams intermittent. Ground water used for irrigation.
Soils: Orthents, Ustolls, Argids, and Fluvents. Thermic temperature regime and aridic
moisture regime.
PNV: Forest, savanna, and desert shrub vegetation.
Actual Vegetation: Desert grasslands in basins, mountains are characterized by a multitude of life
zones, including the southernmost extension of spruce/fir forest.
Primary Natural Disturbances: Drought.
Primary Human Disturbances: Livestock grazing.
Pros & Cons for Trends Project:
Pros: 1) Distinguishing array of semiarid to arid vegetation communities over steep
environmental gradients would be challenging,
2) The fate of this ecoregion is of interest to many because land management
practices have resulted in T&E species (e.g., red squirrel) and because Native
American groups and environmental and outdoor recreation lobbies are trying to
stop the construction of a telescope on the largest (and most sacred) mountain
(Mt. Graham) in the region,
3) Ecoregion is a relatively easy size and shape for sampling.
Cons: 1) Difficult to map vegetation in semiarid and arid systems,
2) Difficult to map land cover in mountainous terrain because of terrain effects.
31
-------
65 Southeastern Plains
Land Use: 70% woodland, 15% cropland, 10% pasture, remaining is urban, rangeland,
other. All woodland privately owned (including industry ownership). Timber
production important. Cash crops include soybeans, corn, peanuts, cotton.
Vegetable crops more locally important. Pastures used mostly for beef cattle,
but some dairy cattle and hogs raised.
Concerns: Controlling soil erosion; improving drainage on low wetland areas.
Terrain: Low, rolling to irregular plains with gradual local relief. 25-200 m elevation,
increasing gradually from the coastal plain toward the interior.
Climate: 1025-1525 mm ppt. annually, increasing from north to south. Minimum ppt.
occurs in autumn.
Water: Abundant water from annual ppt., perennial streams, and groundwater.
Domestic water supplies obtained from shallow wells; water for livestock from
perennial streams and farm ponds.
Soils: Udults (Psamments and Udults in sand hills). Thermic temperature regime,
udic moisture regime.
PNV: Mixed oak-pine and oak-hickory-pine forest.
Actual Vegetation: Evergreen needleleaf trees with scattered areas of cold-deciduous and evergreen
broadleaf forest. Needleaf trees include loblolly, longleaf, slash, and shortleaf
pines; broadleaf trees include sweetgum, yellow-poplar, and red and white
oaks.
Primary Natural Disturbances: Wildfire; climate (hurricanes, occasional summer drought and
winter ice storms, infrequent tornadoes); pest infestation.
Primary Human Disturbances: Clearing of natural veg. for crops, timber harvest.
Pros and Cons for Trends Project-
Pros: 1) High turnover rate of cover types at the local scale. It would be interesting to
see whether they translate to any trends at the regional scale, or whether the local
changes are just "system noise" (there is much interest among landscape
ecologists in how processes and patterns at one scale relate to processes and
patterns at another),
2) This is a good region for developing spatial metrics because of the potentially
different local and regional pattern grains, and
3) The ecoregion is a broad, contiguous unit for sampling.
Cons: 1) The ecoregion is huge and most land is private so field reconnaissance can be
complicated.
32
-------
16 Montana Valley and Foothill Prairies
Land Use: Primarily farms and ranches. Some cropland is irrigated. Crops include
potatoes, sugar beets, peas, hay, grain, and pasture. Dryland wheat farming
occurs where annual ppt. is adequate. One third to 1/2 of the areas is in native
grasses and shrubs. Beef cattle and sheep are the main livestock; dairying is
important locally.
Concerns: Growing urbanization, rural development, shift in economic drivers.
Terrain: Nearly level to sloping valley floors, and sloping terraces and fans bordered by
the steep mountains of the Northern Rockies Ecoregion. Some valleys
modified by glacial deposits.
Climate: <250 mm annual ppt., fairly equally distributed throughout the year (but a bit
lower in summer).
Water: Perennial streamflow from snowmelt of surrounding mountains is generally
adequate for current needs.
Soils: Orthids, Borolls, Argids with a frigid temperature regime.
PNV: Grassland, with shrubs and trees in riparian areas.
Actual Vegetation: Grasses and shrubs (where not used for cropland or urban settlement).
Primary Natural Disturbances: Wildfire, flooding, large mammal grazing.
Primary Human Disturbances: Agriculture (removal of natural vegetation, addition of farm
chemicals to system, increased soil erosion, loss of wildlife
habitat), domestic livestock overgrazing, fire suppression
(allowing encroachment by woody species), human settlement.
Pros and Cons for Trends Project:
Pros: 1) A strong challenge for designing a sampling scheme,
2) Temporal analyses will illustrate land cover change patterns associated with
shifts in economic and environmental drivers.
Cons: 1) A strong challenge for designing a sampling scheme,
2) Compiling a temporal data set of cloud-free imagery will be a challenge, and
3) A short growing/field season.
33
-------
64 Northern Piedmont
Land Use: Complex mix of small farms interspersed with residential/commercial/indus-
trial development, and scattered woodland. Farms primarily in crops except in
northern New Jersey and in Virginia, where they are mostly in pasture and
woodland. Major crop types include forage crops, soybeans, and grain for
dairy cattle. Large centers of urbanization and industry.
Concerns: Spread of urban development is affecting water yields and groundwater
recharge.
Terrain: Mostly gently sloping or sloping low rounded hills, irregular plains, and open
valleys. Elevations range from sea level to > 300 m, but most are between 100
and 300 m. There are scattered serpentine barrens. Complex pattern of
bedrock and surficial geology.
Climate: Humid continental, with cold winters and hot summers. Average annual ppt.
from 900-1200 mm, primarily occurring in spring and early summer. Snow in
winter.
Water: Mature, dendritic drainage network. Natural lakes rare to nonexistent. Small
impoundments common along upper stream reaches. Some bogs, swamps, and
salt marshes in areas adjacent to the Atlantic coast and Chesapeake Bay.
Precipitation, perennial streams, springs, and groundwater provide ample water
for farm, urban, and industrial uses.
Soils: Alfisols and Ultisols with a mesic temperature regime and an udic moisture
regime, and mostly mixed mineralogy.
PNV: Deciduous hardwood forest of oaks, hickories, ash, elm, and yellow-poplar.
Some oak/hickory/pine forest occurred along the Susquehanna River.
Actual Vegetation: Appalachian oak, sugar maple-mixed hardwoods, hemlock-mixed hardwoods,
and oak-chestnut forests and woodlands. Eastern redcedar is common on many
abandoned cropland areas.
Primary Natural Disturbances: Wildfire, pest infestation.
Primary Human Disturbances: Urban expansion, point/nonpoint source pollution.
Pros and Cons for Trends Project-
Pros: 1) Interesting in that patterns in vegetation and agriculture are strongly related to
geologic formations and the intermingling effects of climate and terrain features
on soil development from parent materials.
Cons: 1) Major land cover changes, associated with urban and industrial spread, may not
particularly relate to ecoregion characteristics, so this may not provide an
interesting ecoregion analysis.
34
-------
62 North Central Appalachians
Land Use: Primarily forestry and recreation, some coal and gas extraction in the western
portion (oil wells are common). Suburban and vacation development often near
the larger lakes. Because soils are unsuited to agriculture, most of the
ecoregion remains wooded.
Concerns: Pollution from mining and oil production.
Terrain: Elevated plateau of horizontally bedded strata result in areas of plateau
surfaces, high hills, and low mountains. The eastern portion and the "fingers"
of the western portion were glaciated. Hilltop elevations up to 700 m.
Climate: Cool summers and cold winters. Ppt. from 850-1270 mm, evenly distributed
throughout year, occurring as snow in winter.
Water: Lots of surface water and wetlands.
Soils: Frigid, nutrient-poor soils derived from residuum, colluvium, and till.
PNV: Northern hardwood forest, scattered Appalachian oak forest, isolated highland
pockets of spruce/fir.
Actual Vegetation: Mixed hardwoods predominate, but hemlock and pines also occur. Also,
swamp vegetation (wooded and unwooded).
Primary Natural Disturbances: Tornadoes, wind throw, ice storms, and pest infestation.
Primary Human Disturbances: Rapid expansion of development, oil and gas extraction, timber
harvest.
Pros and Cons for Trends Project:
Pros: 1) Complex and interesting terrain;
Cons: 1) Lots of wetlands and complex surface features, often hard to interpret from
remotely sensed data,
2) The expansion of development, an important regional change, may be hard to
detect because of the forest overstory and, if settlement patterns are dispersed,
because of the limitations of our data's spatial resolution.
35
-------
Appendix C
Omernik Level III Ecoregions, Sample Information
(Based on: 20 Km Block Size
a = 2500
1% margin of error
85% confidence interval)
Omernik Level III Ecoregion
1 . Coast Range
2. Puget Lowland
3. Willamette Valley
4. Cascades
5. Sierra Nevada
6. Southern and Central California Chaparral
and Oak Woodlands
7. Central California Valley
8. Southern California Mountains
9. Eastern Cascades Slopes and Foothills
10. Columbia Plateau
11. Blue Mountains
12. Snake River Basin
13. Central Basin and Range
14. Mojave Basin and Range
15. Northern Rockies
16. Montana Valley and Foothill Prairies
17. Middle Rockies
18. Wyoming Basin
19. Wasatch and Uinta Mountains
20. Colorado Plateaus
21 . Southern Rockies
22. Arizona/New Mexico Plateau
23. Arizona/New Mexico Mountains
24. Chihuahuan Deserts
25. Western High Plains
26. Southwestern Tablelands
27. Central Great Plains
28. Flint Hills
29. Central Oklahoma/Texas Plains
30. Edwards Plateau
31 . Southern Texas Plains
32. Texas Blackland Prairies
Area in
Hectares
5400000.0
1647000.0
1485000.0
4644000.0
5283000.0
10053000.0
4599000.0
1791000.0
5616000.0
9045000.0
6480000.0
6552000.0
34164000.0
13077000.0
16902000.0
6507000.0
9288000.0
12816000.0
4464000.0
12888000.0
13770000.0
19188000.0
10908000.0
17505000.0
28611000.0
15948000.0
27378000.0
2754000.0
10260000.0
5877000.0
5427000.0
5022000.0
# Total
Blocks
135.00
41.18
37.13
116.10
132.08
251.33
114.98
44.78
140.40
226.13
162.00
163.80
854.10
326.93
422.55
162.68
232.20
320.40
1 1 1 .60
322.20
344.25
479.70
272.70
437.63
715.28
398.70
684.45
68.85
256.50
146.93
135.68
125.55
# Sample
Blocks
8.93
6.67
6.42
8.72
8.91
9.60
8.71
6.87
8.99
9.50
9.16
9.17
10.22
9.79
9.94
9.17
9.53
9.78
8.66
9.78
9.82
10.01
9.66
9.96
10.16
9.91
10.15
7.82
9.61
9.04
8.94
8.84
% area
Sampled
6.62%
16.20%
17.28%
7.51%
6.74%
3.82%
7.57%
15.35%
6.40%
4.20%
5.66%
5.60%
1 .20%
2.99%
2.35%
5.63%
4.10%
3.05%
7.76%
3.04%
2.85%
2.09%
3.54%
2.28%
1 .42%
2.49%
1 .48%
1 1 .35%
3.75%
6.16%
6.59%
7.04%
36
-------
Appendix C, Continued
Omernik Level III Ecoregion
33. East Central Texas Plains
34. Western Gulf Coastal Plain
35. South Central Plains
36. Ouachita Mountains
37. Arkansas Valley
38. Boston Mountains
39. Ozark Highlands
40. Central Irregular Plains
41. Canadian Rockies
42. Northwestern Glaciated Plains
43. Northwestern Great Plains
44. Nebraska Sand Hills
45. Piedmont
46. Northern Glaciated Plains
47. Western Corn Belt Plains
48. Lake Agassiz Plain
49. Northern Minnesota Wetlands
50. Northern Lakes and Forests
51 . North Central Hardwood Forests
52. Driftless Area
53. Southeastern Wisconsin Till Plains
54. Central Corn Belt Plains
55. Eastern Corn Belt Plains
56. S. Michigan/N. Indiana Drift Plains
57. Huron/Erie Lake Plains
58. Northeastern Highlands
59. Northeastern Coastal Zone
60. Northern Appalachian Plateau and Uplands
61 . Erie Drift Plains
62. North Central Appalachians
63. Middle Atlantic Coastal Plain
64. Northern Piedmont
65. Southeastern Plains
66. Blue Ridge Mountains
67. Ridge and Valley
68. Southwestern Appalachians
69. Central Appalachians
70. Western Allegheny Plateau
71 . Interior Plateau
72. Interior River Lowland
Area in
Hectares
4374000.0
6624000.0
15462000.0
2637000.0
2646000.0
1710000.0
10800000.0
12285000.0
1962000.0
16002000.0
34920000.0
6255000.0
16416000.0
15390000.0
20250000.0
4149000.0
2412000.0
18396000.0
8811000.0
4743000.0
3051000.0
9828000.0
8370000.0
7209000.0
2475000.0
12699000.0
3474000.0
3114000.0
3051000.0
2916000.0
8145000.0
3051000.0
33534000.0
4698000.0
11547000.0
3564000.0
5985000.0
8460000.0
12807000.0
9189000.0
# Total
Blocks
109.35
165.60
386.55
65.93
66.15
42.75
270.00
307.13
49.05
400.05
873.00
156.38
410.40
384.75
506.25
103.73
60.30
459.90
220.28
118.58
76.28
245.70
209.25
180.23
61.88
317.48
86.85
77.85
76.28
72.90
203.63
76.28
838.35
117.45
288.68
89.10
149.63
211.50
320.18
229.73
# Sample
Blocks
8.63
9.19
9.89
7.73
7.74
6.76
9.65
9.75
7.09
9.91
10.22
9.12
9.93
9.89
10.03
8.55
7.54
9.98
9.48
8.75
8.02
9.58
9.43
9.28
7.60
9.77
8.25
8.05
8.02
7.93
9.41
8.02
10.21
8.74
9.70
8.30
9.07
9.44
9.78
9.52
% area
Sampled
7.89%
5.55%
2.56%
1 1 .72%
1 1 .69%
15.82%
3.58%
3.17%
14.45%
2.48%
1.17%
5.83%
2.42%
2.57%
1 .98%
8.24%
12.51%
2.17%
4.30%
7.38%
10.51%
3.90%
4.51%
5.15%
12.28%
3.08%
9.50%
10.35%
10.51%
10.88%
4.62%
10.51%
1 .22%
7.44%
3.36%
9.31%
6.06%
4.46%
3.05%
4.14%
37
-------
Appendix C, Continued
Omernik Level III Ecoregion
73. Mississippi Alluvial Plain
74. Mississippi Valley Loess Plains
75. Southern Coastal Plain
76. Southern Florida Coastal Plain
77. North Cascades
78. Klamath Mountains
79. Madrean Archipelago
80. Northern Basin and Range
81. Sonoran Basin and Range
82. Laurentian Plains and Hills
83. Eastern Great Lakes and Hudson Lowlands
84. Atlantic Coastal Pine Barrens
Total
Area in
Hectares
13356000.0
4572000.0
13014000.0
2259000.0
3033000.0
4851000.0
4167000.0
10962000.0
11691000.0
4536000.0
5805000.0
1602000.0
# Total
Blocks
333.90
114.30
325.35
56.48
75.83
121.28
104.18
274.05
292.28
113.40
145.13
40.05
19465.20
# Sample
Blocks
9.80
8.70
9.79
7.40
8.00
8.79
8.56
9.66
9.71
8.69
9.03
6.60
755.12
% area
Sampled
2.94%
7.61%
3.01%
13.11%
10.56%
7.25%
8.21%
3.53%
3.32%
7.66%
6.22%
16.49%
6.62%
38
------- |