Vertebrate Modeling

Overview

One of the major objectives of the National GAP Project is to document the representation of native vertebrate species in regional assesments. In order to meet this objective, we developed a spatial database of predicted species distributions throughout the region. While we don't have the resources to conduct a complete survey for every species, we do have access to a tremendous knowledge base that allows us to model distrubutions based on known range and habitat relationships.

The composition and structure of the dominant vegetation is an important and easily described measure of habitat for animals (Scott et al. 1993) and has long been used as an indirect indicator of animal distributions (Austin 1991). Other biotic and abiotic factors (i.e. elevation, wetland type, and distance from standing water) can also play a major role in defining a particular species' habitat. Many studies of vertebrate species have been conducted over the years documenting this type of information. In addition, data on known ranges for vertebrate species has also been collected. This includes not only survey data records, but the cumulative field experiences of biologists who work with these species on a daily basis. If we take a habitat description, break it down into vegetative communities and other environmental factors (e.g. elevation, wetland type, etc.), and then identified those areas within a known range, we would be creating a predicted distribution map. That is essentially what we did for each terrestrial vertebrate species modeled in the Southeast. The final step was to overlay those maps together to get a representation of biodiversity.

Why did we model only terrestrial vertebrates? Again, its a matter of resources. Realistically, we can only model a subset of all species. Terrestrial vertebrates are a logical choice because we generally know more about their habitat relationships and range extents than those of other taxa. Even so, there are over 600 terrestrial vertebrates modeled. The key to filling in the 'gaps' in the documented data is to have a thorough review by experts at every stage.

Return to Top

---

Species List Development

Selection Criteria

Gap Analysis predicted distribution maps are being created for all vertebrate species that breed in the Southeastern U.S. or use habitat there for an important part of their life history. Decisions were made on which species' habitats will be mapped based on standard Gap Analysis guidelines. Species lists have been created for each of the southeastern states by state-level Gap projects; these lists were compiled and pared down to remove some subspecies and domesticated species. Subspecies were only included when supported as distinct and non-overlapping with either the full species or other subspecies. The final list includes amphibians, mammals, reptiles, breeding birds, and wintering waterfowl.

Species Lists

After review, 602 species remained for habitat modeling:

	131 Amphibian species:	View List
	253 Avian species:	View List
	95 Mammalian species:	View List
	123 Reptilian species:	View List

Models of presence/absence for a species' predicted distribution may include a number of spatially explicit data sources. GAP models typically involve land use/land cover data as the primary input. However, other environmental features that make up the landscape constituting species' habitats can be valuable inputs to modeling. The SE-GAP project has attempted to use a number of data layers (ancillary data) other than land cover to develop species models. Many of these data layers act as surrogates for one or more aspects of a species' habitat that may only be inferred from available, remotely sensed information. The following document is meant to describe these data layers and their development, their range of parameters, and usage in modeling species' predicted distributions.

NOTE: The term "map unit" will be used throughout the document to mean the individual class or category mapped to represent land use and land cover, i.e. ecological systems.

All 602 terrestrial vertebrate species' geographic range extents were delineated as single or multiple polygons. Migratory species were primarily represented by breeding season ranges, however, wintering ranges for waterfowl and migratory bats were also delineated (33 species) Processes used to create range polygons were unique because information on the current geographic range of a species varies widely. However, a generalized approach used a variety of sources to develop species' ranges including sources of information in two broad categories: 1) species location records and range maps available digitally or in print, and 2) digital spatial data of environmental parameters described in later sections of this document.

Usage in Modeling

Species ranges were used as model delimiters in predicted distribution models.

Development

Ecological systems and land use mapped for the SE-GAP project are the primary input to species models. The development of these data is described in detail in subsequent documents. However, derivatives of these data were developed separately to further refine species models.

Primary and secondary (auxiliary) map units, and patch size (contiguous and non-contiguous patches) are parameters utilized as a model proceeds. That is, they are not stand-alone, independently created data layers. As such, they would not be considered "ancillary" data.

Usage in Modeling

Primary map units are those land cover types critical for nesting, rearing young, and/or optimal foraging.

Secondary or auxiliary map units are those land cover types generally not critical for breeding, but are typically used in conjunction with primary map units for foraging, roosting, and/or sub-optimal nesting locations. These map units are selected only when located within a specified distance from primary map units.

The type and size of clusters of habitat can be assessed with spatial modeling. As a final step to the distribution modeling process, we used these parameters for species shown to require minimum amounts of habitat. This includes not only directly adjacent habitats, but those that are contextually adjacent. Both these parameters are utilized as the final step in a predicted distribution and therefore incorporate all other modeling inputs.

The edge or ecotone between forested and non-forested environments can be a critical aspect of the habitat landscape. We grouped map units into forested, non-forested, and shrubland/woodland land cover types to create unique data layers. These data layers can then be buffered at specified distances to identify species' habitats.

All map units categorized as forest, non-forest, and shrubland/woodland were aggregated into separate data layers. Aggregated map units can be compared and contrasted to identify areas of transition between these broad categories. They can also be used to identify "core" areas or contiguous blocks of similar type (i.e. interior) through buffering procedures.

Forested map units were defined as those ecological systems and land uses that correspond to the National Land Cover Dataset (NLCD) forest/woodland classes (Homer et al. 2004) (Appendix A). These included NLCD classes "deciduous forest", "evergreen forest", "mixed forest", "palustrine forested wetland", and "estuarine forested wetland" (class values 41, 42, 43, 91 and 93 respectively). Non-forested map units were defined as those ecological systems and land uses not categorized as forest using the criteria stated above. Woodland/shrubland map units were defined as those ecological systems and land uses containing a majority of short, scrubby, woody vegetation or sparsely canopied treed vegetation (Appendix B). Unlike the forested map units, the woodland/shrubland layer was not explicitly derived from NLCD classes but was selected individually based on attributes defined by the systems and land uses.

Once these data layers were created, buffering procedures created unique datasets containing unique categories of distance: > 4000, 4000, 2000, 1000, 500, 250, 120, 60, and 30 meters into and away from the specified feature.

Figure 1. Example of an ecotone/edge buffer on forest/non-forested environments.

Development

The HUC is part of a hierarchical classification system for surface water drainage in the US (Seaber et al. 1987). The numerical code represents a "cataloging unit" of delineation approximately larger than 1800 square kilometers. Within the SE-GAP project boundary there are 577 8-digit HUCs (Figure 2). Vector format (i.e. polygon) data were compiled from United States Geological Survey (USGS) sources. No ancillary data was produced directly from HUCs, but they were utilized in numerous other ancillary data layer processes. Most notably, HUCs allow for partitioned processing when scripting complex spatial analysis tasks.

Figure 2. The 577 8-digit HUCs in the SE-GAP project region.

Usage in Modeling

HUCs were not used directly in species modeling. However, some species' geographic ranges were developed using the 8-digit hydrologic unit code polygons. This was only appropriate for species whose ranges would conceivably be limited to certain watershed drainages and their associated topographic distinctions. In most cases, these species were partially or entirely aquatic during some part of their life history (e.g. salamanders, turtles).

Development

Ecoregions developed by Omernik (1987, 1995) and available from the U.S. Environmental Protection Agency (US EPA 2007) were produced through an analysis of biotic and abiotic patterns relating to ecosystem differentiation. Although no ancillary date were created directly from ecoregions, they were utilized in numerous other primary and ancillary data layer processes including land cover/ecological systems, species' ranges, and landforms.

Usage in Modeling

Ecoregions were not used explicitly in predicted distribution models. However, Omernik Level III and Level IV ecoregions were used to help define polygon boundaries for some species' geographic ranges when appropriate. Some species are clearly restricted to the physiographic and geographic patterns and boundaries that ecoregions define. For these species, ecoregions were used as all or a portion of the polygon(s) delineation. If there was a doubt as to the limitations imposed by such boundaries, ecoregions were not used.

Development

State geographic soils database (STATSGO) soil units polygons were developed by the U.S. Department of Agriculture's Natural Resource Conservation Service (NRCS) at a scale of 1:250,000 for the conterminous U.S. (USDA 1995). Although no ancillary date were created directly from soils, they were utilized in numerous other primary and ancillary data layer processes including land cover/ecological systems, species' ranges, and landforms.

Usage in Modeling

Soils were not used explicitly in predicted distribution models. However, soils were used to help define polygon boundaries for some species' geographic ranges when appropriate. Although ecoregions consider soils as an abiotic factor in their delineation, for certain species it was necessary to focus more narrowly on soil type to help define a range polygon. For species whose habitat requirements are restricted to certain soil types (particularly some reptiles), soil polygons were used to help delineate all or a portion of their range.

Development

Roads data were compiled from the U.S. Census Bureau's TIGER/Line files (USCB 2007). As a preliminary and stand alone layer, road density grids were created so that they could be used for production of imperious surface data layers and, by proxy, land cover/land use map units, as well as in production of an avoidance mask for vertebrate modeling. Part of the procedure to develop road density required the summarization of the attribute code 'CFCC' within the TIGER database (see technical documentation for TIGER/Line files at the URL listed in the Literature Cited section for more information). In general, seven codes were combined into three road category types (Table 1). A line density algorithm was used to calculate the linear length of roads within a circular neighborhood. The algorithm utilized a kernel smoothing function on a 30 meter cell size within a 564.19 meter radius (1 km2).

Usage in Modeling

Roads were not used explicitly in predicted distribution models. However, the road density data layer derived from roads was used to develop an avoidance mask as an index to human disturbance (see Urban).

Development

Elevation data are often represented in digital elevation models (DEMs). SE-GAP utilized the USGS compiled National Elevation Dataset (NED) as a primary source (USGS 2003). However, because of the varying quality of the data, it was necessary to incorporate other datasets to create an improved, region wide product. These included data from NASA's Shuttle Radar Topography Mission (SRTM) at 30 meter resolution (NASA 2007), Light Detection and Ranging (LIDAR) data from the North Carolina Floodplain Mapping Program (NCFMP 2007), and hypsography data from the USGS's Digital Line Graphs (DLGs) at 1:24,000 and 1:100,000 scales (USGS 2007a). Areas of inconsistent, erroneous, or systematically flawed data were identified visually and "tagged" for fixing. A number of algorithms were then used to reassign elevation values using interpolations based on the higher quality data. This essentially promoted the best available information for a given area using a number of sources, as opposed to re-interpolating data from the same flawed source.

Usage in Modeling

Some species respond to environments directly related to altitudinal variation. Elevation is easily implemented in spatial modeling by limiting the model to the minimum and maximum values explicitly stated in the literature. DEMs are utilized directly and are measured in meters above mean sea level.

Development

Landforms are derived from DEMs, ecoregions and hydrography using a complex model of slope, aspect, location, elevation, flow direction and accumulation, and a variety of other interpolated data. These are meant to incorporate a mixture of environmental inputs important for species.

1. Extract data inputs for a given region to process (mapping zones for example).
2. Subdivide into smaller processing units (HUCs for example) that can be overlapped during iteration loops to facilitate merging during production of the final output.
3. Calculate Topographic Relative Moisture Index (TRMI) based on a DEM and hydrography. TRMI is an estimate of potential moisture at a specific location created by combining relative slope position, aspect, slope and curvature from the DEM, and defining valley bottoms that correspond to high resolution hydrography data.
4. Process the moisture index grid to reclassify it into binary categories of moist and dry; smoothing the output with a majority filter to remove single cells and linear artifacts.
5. Calculate land position based on the inverse distance weighted elevation of each cell in relation to its neighbors.
6. Calculate the landform for each processing unit by combining the moisture, land position, slope and aspect.
7. Create the final landform for the region by combining landforms of overlapping processing units into a single output.

Usage in Modeling

Species distributions are restricted to the selected landforms. Landforms include: Cliffs, Steep Slopes, Slope Crests, Side Slopes, Upper Slopes, Flat Summits, Coves, Slope Bottoms, Moist Flats, Dry Flats, and Wet Flats.

Development

Water and its location on the landscape is a very important aspect of species' habitats. SE-GAP used a number of water related data layers to refine species models. These include water type (i.e. flowing or open/standing), distance to and from water, and stream flow and underlying gradient.

The source for hydrographic data was the USGS National Hydrography Dataset (NHD)(USGS 2007b). The NHD is composed of the most up-to-date nationwide data of surface waters compiled at two scales: 1:24,000 and 1:100,000. SE-GAP began acquiring NHD data in 2003. At the time, NHD was available only in ArcInfo coverage format organized by 8-digit Hydrologic Unit Code (HUC). NHD 8-digit HUCs are a complex mixture of polygon and arc (and point) features with numerous related tables of attributes. These coverage data are vector data types, i.e. arcs, polygons, nodes etc. with their incumbent data tables.

Distance INTO and FROM Water Features
We used 18 categories of distance from the water features flowing water, standing water and wet vegetation. A process was used to buffer hydrography at various distances and code grid cells based on a defined set of distances. Table 3 summarizes the 18 code categories for buffer grids. Each water type has a region wide grid coded for buffer distance from that type of hydrography.

Velocity (Stream Gradient)
Stream velocity (or more appropriately stream gradient) was derived from a combination of streams and slopes calculated from a digital elevation model (DEM). A process was developed that determined the cell-to-cell drop in elevation along a stream grid based on flow direction. The process created three categories of slope calculations for use as a surrogate for stream gradient/velocity (Table 4). A double pass focal majority function was used to "smooth" the output.

Usage in Modeling

Water Type

Distance INTO and FROM Water Features

Interior (INTO) and Exterior (FROM) buffers on water features measured in meters and represented as discrete, cumulative distances. For example, a 250 meter exterior buffer on flowing water includes all areas up to 250 meters from the flowing water feature.

Velocity (Stream Gradient)

For some aquatic species, this can represent an important aspect of their habitat such as oxygenation levels, presence of invertebrate prey, and amount of sediment within the water column and on streambed substrates.

Development

Water salinity is a major factor when considering habitat conditions for many species. However, the dynamic and complex nature of water systems makes the development of a highly refined and reliable datalayer impractical, if not impossible. Therefore, we developed three general categories to include in species habitat models.

Zones of water salinity were hand digitized using NLCD map units as a guide. All palustrine and upland map units were identified, as were all estuarine map units. A line was digitized between non-estuarine and estuarine map units and served as the boundary between fresh and brackish water. An additional line was digitized across all ocean inlets, essentially the shortest distance between two bodies of land. This second line served as the boundary between brackish and salt water.

Usage in Modeling

Development

Environments dominated by human disturbance such as roads, cities, and the constructed materials that support human habitation have profound effects on species' distributions. Map units derived for SE-GAP include three categories of development intensity - high, medium and low. These classes rely on spectral signatures from satellite data in association with measurements of impervious surface. SE-GAP created an urban/avoidance data layer using a combination of road density (see Roads section above) and the three map units of development intensity. In effect, the derived data layer acts as an index for a species' intolerance to human environments.

Usage in Modeling

For most species, this data layer was used to "mask out" or exclude species from a portion of the landscape (see Figure 3). However, some species respond favorably to human habitats. In this later case, the data layer was used in an inclusionary manner see (Figure 4).

Level of Tolerance/Intolerance
This parameter is ranked at three levels: high, medium, and low. No selection of this parameter indicates the species' model is not contingent on an index of human disturbance.

Figure 3. The index of human disturbance at three levels: light gray = low; gray = medium; black = high. Using an exclusionary mask, the predicted distribution of a species is excluded from the black areas and limited to the white areas at intolerance levels of high, medium and low (top to bottom).

Figure 4. The index of human disturbance at three levels: light gray = low; gray = medium; black = high. Using an inclusionary mask, the predicted distribution of a species is excluded from the black areas and limited to the white areas at tolerance levels of high, medium and low (top to bottom).

Last updated: June. 24, 2011