An Assessment of Landscape Change in the Mid-Atlantic Region of the United States And Consequences to Streams and Terrestrial Wildlife Habitats (1973-1992)
Investigators: K. Bruce Jones1, Anne C. Neale1, Curtis Edmonds1, Maliha S. Nash1, Timothy G. Wade1, James D. Wickham2, and Thomas Loveland3
- U.S. Environmental Protection Agency, Office of Research and Development,
Las Vegas, Nevada, USA
- U.S. Environmental Protection Agency, Office of Research and Development,
Research Triangle Park, North Carolina, USA
- U.S. Geological Survey, EROS Data Center, Sioux Falls, South Dakota, USA
Landscape change is considered to be one of the greatest threats to aquatic and terrestrial resources in many areas of the United States and the world. In the Mid-Atlantic Region, urban sprawl is thought to have had a negative effect on surface waters and terrestrial habitats, especially forests. One of the primary goals of the Landscape Sciences Program is to conduct a national assessment of landscape between the early 1970s and the early 2000s, and to determine how changes over this period of time influenced aquatic and terrestrial resources. The Mid-Atlantic Region has one of the richest databases in the United States, including data on stream conditions, breeding bird populations, and landscape change. There are over 800 stream and 500 Breeding Bird Survey samples in the region, as well as several region-wide landscape databases that are necessary to calculate landscape indicators and to evaluate landscape change. Therefore, the Mid-Atlantic Region provides an opportunity to test a regional-scale, landscape change approaches, models and methods to assess the consequences of change on an aquatic and terrestrial resource streams and breeding bird habitats, respectively. The goal of this project is to develop and test assessment methods for habitat and water environmental endpoints using a spatially distributed models and GIS. The project should also result in insights as to how the landscape changed during a recent 20-year period thereby helping valid landscape change models being proposed in the ReVA program.
The study was undertaken in the mid-Atlantic region of the United States (so. New York, Pennsylvania, eastern New Jersey, Delaware, Maryland, West Virginia, Virginia, no. North Carolina). Digital land-cover maps were acquired and processed to examine the spatial concordance of temporal changes in nitrogen load and temporal changes in an index of bird habitat. Overall, the methodology was separated into three steps:
1) acquire and process land-cover data for two time periods,
2) use the land-cover data to run nitrogen load and bird habitat models for each time period, and
3) compare output of models across time and themes (nitrogen and bird habitat).
The temporal land-cover data were from the early 1970s and early 1990s. The 1970s land-cover data were created from Landsat Multispectral Scanner (MSS) data that were acquired as part of the North American Landscape Characterization (NALC). The NALC program distributed the MSS data at a resampled pixel size of 60 meters. The 1990s data were acquired from the Multi-Resolution Land Characteristics (MRLC) program.
Unlike MRLC, there was no pre-existing land-cover data from the NALC program for the 1970s. The NALC Landsat MSS data were classified into seven land-cover classes using euclidean minimum-distance–to-mean clustering and ancillary data. The primary ancillary data sets were USGS Land Use Data Analysis (LUDA) land-cover and National Wetlands Inventory (NWI) data. The seven land-cover classes were:
bare ground (bare rock, sand, mines)
Both land-cover data sets were resampled to a 120-meter pixel in order to accommodate for the differences in native spatial resolution (30-meter Landsat TM and 60-meter NALC’s Landsat MSS). Also, the two data sets were calibrated so that no urban areas in the 1970 Landsat MSS land-cover data were lost in the 1990 land-cover data because of increasing tree density in maturing residential areas. Areas classified as urban in 1970 but not in 1990 were changed to urban in the 1990 land-cover data.
The study area was tessellated into 25 kmgrid cells to accommodate the nitrogen and bird habitat models, and avoid per-pixel calculation of changes between the two land-cover maps. The nitrogen load model was taken from Jones et al. (2001), which empirically estimated the mass of nitrogen output from several watersheds in the mid-Atlantic region using flow-adjusted concentrations. The Jones et al. (2001) model estimated nitrogen load as a function of riparian cover and nitrate deposition. For this study, the model was re-calibrated with riparian cover removed because of the four-fold increase is spatial resolution (30- to 120- meter pixels). The model was re-calibrated using step-wise regression without inclusion of the riparian cover metric. The re-calibration replaced riparian cover with agriculture (with the appropriate change in signs) and retained nitrogen deposition:
Ln N = 0.02114alc + 0.00175nd - 1.58487, where
alc is proportion of watershed in agriculture, nd is nitrate deposition (Kg/yr), and N is the natural logarithm of nitrogen yield (kg/ha/yr). The R-square was 0.80, with alc and nd explaining 57% and 24% of the variance, respectively.
The bird model was taken from O’Connell et al. (2000). The model characterizes the relationship between bird community index (BCI) scores and landscape conditions. A first-level characterization of the model stratifies bird habitat into poor, moderate, and good to excellent based on the percentage of forest cover. Areas with less than 41% cover are considered in poor condition and areas greater than 80% forest are considered in good to excellent condition. Areas between 41% and 80% are considered moderate. A second level of characterization in the model considers the type of nonforest land-cover. When forest cover is less than 41% and urban is greater than 33%, the site is considered poor-urban. If forest is less than 41% and agriculture is greater than 50%, the site is considered poor-agriculture. O’Connell et al. (2000) found that different groups of birds occupied poor-urban and poor-agriculture sites.
In our implementation of the model, we found other poor areas that met neither the poor-urban nor poor-agriculture criteria. These sites were mixtures of agriculture, developed, and forest, and also included large amounts of barren (mines or large clear-cuts). We characterized these areas as poor-other. Our application of the O’Connell et al. (2000) model included five classes:
good to excellent
The models were applied to each grid cell using each land-cover data set to uncover temporal changes in each theme. The differences in each theme were also compared to determine the spatial concordance of changes in bird habitat and nitrogen yield.
The project was completed and results were reported in: Jones, K.B., A.C. Neale, T.G. Wade, J.D. Wickham, C.L. Cross, C.M. Edmonds, T.R. Loveland, M.S. Nash, K.H. Riitters, and E.R. Smith. 2001. The consequences of landscape change on ecological resources: an assessment of the United States Mid-Atlantic Region, 1973-1993. Ecosystem Health 7:229-242.
At the scale of the Mid-Atlantic region, the amount of change in bird habitat condition and nitrogen yield was relatively small. However, there was considerable spatial variation in landscape change as it relates to bird habitats and nitrogen yield. Areas that experienced declines in bird habitat and increases in nitrogen yield (worsening conditions) had common themes in landscape change. Based on the two models, these areas had agricultural and urban land-cover gains at the expense of forests. Declines in condition tended to be in spatial clusters associated with urbanization and agricultural expansion in areas with large blocks of agriculture in the early 1970s. Although not quantified in this study, urban expansion tended to occur around urban areas that existed in the early 1970s.
Differences in the grain size and patchiness of changes in bird habitat quality and nitrogen yield likely resulted from differences between the two spatially distributed models. The bird model was a rule-based model resulting in categorical scores for each of the grid cells (poor, moderate, and good; O’Connell et al. 2000), whereas the nitrogen yield model was a statistical model resulting in a continuous variable of predicted nitrogen yield (Jones et al. 2001). A bird model that creates a continuous variable of habitat quality would permit a more direct comparison of changes in the spatial pattern of bird habitat quality and nitrogen yield.
Because of the nature and quality of the 1970s Landsat MSS data, it was difficult to decipher whether transitions from forest to herbaceous and vice versa were losses or gains in agriculture, or changes in successional patterns in forests. Correctly identifying these transitions impacts the results of the bird habitat quality and nitrogen yield model. Early successional forests likely lose less nitrogen and provide better bird habitat than agricultural fields because the latter has bare soil for at least six months. Estimates of urban or developed areas in the 1990s were improved by making the assumption that urban areas could not be lost between the 1970s and 1990s. Initially, it was found that urban or developed areas were lost within the boundaries of cities and along major highways. By evaluating the transition type (e.g., urban to forest) and location (e.g., within an urban area), it was concluded that the loss of urban land-cover was due to maturing of individual trees. Therefore, it was assumed that loss of urban was very unlikely and reclassified those urban pixels that had changed back to urban. Understanding transition probabilities as well as the landscape setting may improve land-cover change estimates substantially and the spatial distributed models that depend on them.
One of the biggest issues confronting those doing regional environmental assessments is how to best represent and depict the surface of a region. Probability samples consisting of unbiased measurements of ecological indicators are one way to estimate ecological conditions over broad areas. However, decision makers and environmental managers within a region invariably want to know the condition of the area that they manage and how their area compares to other areas. Therefore, some type of spatial extrapolation is needed to estimate conditions across the surface of a region. Spatially distributed models similar to the two used in this study are one way to extrapolate conditions to a continuous surface. And when these models are quantitatively associated with landscape metrics generated from continuous spatial data, spatial concordance between multiple environmental endpoints can be evaluated, leading to a comprehensive assessment of landscape health.
It also is important to decide how to depict the spatial variability of model results given that potential users of such results need data at a range of spatial scales. Several regional scale reports have used 8-digit Hydrological Unit Code watersheds to depict and report environmental conditions, including indicator summaries based on relatively fine-scale landscape data. At the scale of a large region, this representation of spatial variation may be appropriate, but to be useful for state and county organizations, finer-scale spatial representation may be necessary. The use of 25kmgrid cells in this study demonstrated that spatial variation can be depicted at a scale useful for regional as well as finer-scale assessments. However, the models used to evaluate conditions must be related to land features and the spatial realm upon which key processes operate. In this study, both models were related to land features where a grid cell concept would apply. There are some cases, however, where a grid cell design would not apply. For example, for some water-related processes and endpoints, a watershed delineation may be more appropriate.
Finally, spatially distributed models similar to those used in this study can be used to evaluate the consequences of alternative landscape futures on environmental endpoints. In this case, the models evaluate changes between current conditions and alternative landscape conditions projected by socio-economic models, biophysical models, or those developed by land use planners to identify options with the least environmental impact.
Jones, K.B., Neale, A.C., Nash, M.S., Van Remotel, R.D., Wickham, J.D., Riitters, K.H., and O’Neill, R.V. (2001). Predicting nutrient and sediment loadings to streams from landscape metrics: a multiple watershed study from the United States Mid-Atlantic Region. Landscape Ecology 16, 301-312.
O’Connell, T.J., Jackson, L.E., and Brooks, R.P. (2000). Bird guilds as indicators of ecological condition in the central Appalachians. Ecological Applications 10, 1706-1721.
A data browser and set of analyses related to this project have been published on CD ROM and will be made available on the website in the future. The CD is entitled: Mid-Atlantic Region Land Use/Land Cover Change Data Set 1970s to 1990s (EPA/600/R-02/035) and can be obtained by contacting the Landscape Ecology Branch.
The following publication reports the complete findings of this project: Jones, K.B., A.C. Neale, T.G. Wade, J.D. Wickham, C.L. Cross, C.M. Edmonds, T.R. Loveland, M.S. Nash, K.H. Riitters, and E.R. Smith. 2001. The consequences of landscape change on ecological resources: an assessment of the United States Mid-Atlantic Region, 1973-1993. Ecosystem Health 7:229-242.
The data browser and data for this project has been published as a CD ROM and is referenced as:
Edmonds, C.M., D.T. Heggem, and K.B. Jones. 2002. Mid-Atlantic Region Land Use/Land Cover Data Set 1970s to 1990s. EPA/600/R-02/035.