Return to previous page
Idaho Vertebrate Modeling Methods

Modeling of vertebrate distributions for ID-GAP followed a 7-step process. First, we compiled a list of species to be modeled in Idaho. Second, we collected occurrence and habitat association data for each species. Third, we used the occurrence data to approximate the range boundaries of each species in Idaho. Next we assembled the habitat association information into a format acceptable by our modeling programs. Fifth, we combined the range approximation with the coded habitat associations to produce a GIS model of the predicted distribution of each species. Sixth, biologists familiar with the distribution of Idaho’s wildlife reviewed the models. Finally, each model was subject to an accuracy assessment with independent occurrence data.

Step 1 – Idaho Species List

Over 500 species have been recorded within Idaho (G. Stephens, personal communication). Many of these are migratory, wintering or accidental birds that do not regularly breed in the state. At this time, GAP only considers species that are “known to breed in the project area and that are regularly occurring non-accidentals” (Csuti and Crist 1998). Csuti and Crist (1998) suggest, as a general definition, that “regular breeders” are those species breeding in the state at least 5 of the past 10 years. However, this is often difficult to document.

The species list for the ID-GAP project was compiled in cooperation with the Idaho Conservation Data Center (IDCDC) and was based on species listed as regular breeders in the Idaho Biological Conservation Database (BCD). However, this list did not include many non-native species. Because many exotic species influence the distribution and/or densities of native species, or are valued as game species, we also included all regularly occurring non-native species.

We verified the birds included on the Idaho species list with Stephens and Sturts (1997) breeding records. We also removed the Whooping Crane (Grus americana) from the species list because it is an experimental population and has not successfully bred in Idaho (S. Bouffard, personal communication).

Step 2 – Collecting Species Data

Once the species list was finalized, collection of data began for each species. Two types of information were needed to model the distribution of terrestrial vertebrates for GAP: occurrence data and habitat association information. We created a database in Microsoft Access to store and organize the habitat information collected for each species. 

Information on the occurrences of species came from many sources. Most of the data sets did not record the same information for each occurrence. In light of this, we required a minimum subset of information for inclusion of occurrence data in Idaho Gap Analysis: taxonomic name of species, location coordinates (e.g., Lat/Long, UTM, PLSS), accuracy of point coordinates, and source of information. Date of collection was desirable, but not required. The IDCDC provided us with the Point Occurrence Database (POD), a digital record set of all bird and mammal museum specimens from Idaho. In addition, IDCDC provided us with the element occurrence database, a portion of the BCD that tracks the locations of rare or sensitive species (referred to as “elements”). For birds, additional information was collected from Idaho Breeding Bird Survey routes, the US Forest Service Region 1 Landbird Monitoring Project, and Stephens and Sturts (1997). For amphibians and reptiles, occurrence data was obtained from field surveys and incidental observations collected by C. Peterson at Idaho State University. Data for big game species came from Idaho Department of Fish and Game (IDFG) annual reports. In addition, model reviewers provided locations for species that they reviewed. Checklists from special management areas in Idaho were withheld from the model building process for accuracy assessment.

Habitat association information for each species was obtained from published references and expert opinion for each species. Specifically, we collected information regarding use of cover types and special habitat features (e.g., riparian areas, distance to water). We recorded limits of use for canopy cover, elevation, slope, aspect, and temperature (estimated by mean Julian date of lilac blooming). We also integrated information from model reviewers.

Step 3 – Approximating Range Extent

General Methods

The primary unit for approximating the range of a species for ID-GAP was the Environmental Protection Agency Ecological Mapping and Assessment Program (EMAP) hexagon (White et al. 1992). Because hexagons have a constant shape and size, and are easily aggregated or tessellated, they overcome many problems associated with delineating species ranges using county boundaries (Boone 1998). Any other standardized unit could have been used to approximate species ranges; however, we chose hexagons to be consistent with methods of adjacent state Gap Analysis projects.

An Arc Macro Language (AML) program was written by B. Butterfield (IDFG), modified by J. Karl (Landscape Dynamics Lab), to attribute each species observation record to one or more hexagons. This was accomplished by constructing a buffer around each point based on the placement accuracy of that point. The percent of that point buffer that fell within each hexagon was used to code the hexagon as confident, probable, or possible (Table 1, Map 3.1). Points with placement accuracy greater than 5km were deemed too inaccurate for Idaho Gap Analysis since they would result in such a large number of hexagons being attributed for a single record. Any record coded as confident or probable that had a date older than 1950 were coded as historic. All records prior to 1950 with a possible rating were discarded.

For example, a museum record for American marten (Martes americana) may read, “Cougar Creek, Kootenai County.” Since no coordinate was recorded for this point, the most likely location would be established from a 1:100,000 map of Kootenai County. The accuracy rating for this record would be determined by how large Cougar Creek is or any other information recorded with the observation. The AML program would buffer the point by its accuracy, intersect it with the hexagon coverage, and calculate the area of the buffer in each hexagon that it intersected. The hexagon would be attributed for this record according to the rules in Table 1. Thus, the more accurate a point is, the smaller its buffer, and the more likely that buffer is to be contained within a single hexagon. Conversely, an inaccurate point, receiving a large buffer, will overlap several hexagons. Since the location of the observation cannot be attributed to a single hexagon with confidence, we code the hexagons based on the likelihood (based on simple area) that the point occurred in each hexagon.

As a result of several regional studies and expert opinion, we developed rules that allowed us to appropriately expand or contract the range for each species. These methods were originally developed for the reptile and amphibian models, but were subsequently applied to mammal and bird models as well. Rule 1 said that any hexagon adjacent to another hexagon rated as confirmed or probable would be rated possible (Map 3.1). Rule 2 said that any hexagon bordered by 3 or more hexagons rated confirmed, probable, possible, historical, or Rule 1 would also be rated possible. The states that border Idaho (Montana, Wyoming, Oregon and Washington) had all completed GAP models for reptiles and amphibians. Therefore, any hexagon rated by one of these states for any species was given that same rating for Idaho. This allowed for continuity of species range maps across state boundaries.

Methods Specific to Birds

Lack of occurrence records for many bird species prompted us to consider alternative sources of occurrence information other than those listed above. Stephens and Sturts (1997) recorded the occurrence of birds in Idaho by latilong (1-degree latitude, 1-degree longitude blocks). Idaho contains only 17 latilongs, a resolution much coarser than the hexagons being used for other occurrence data. However, Stephens and Sturts (1997) is regarded as the most complete source of bird distribution information in Idaho (almost all bird reviewers relied heavily on it). For this reason, we digitized the latilongs for Idaho and attributed them according to Stephens and Sturts (1997). We then intersected this coverage with the hexagons for each species. All hexagons attributed in this manner were coded as latilongs.

Methods Specific to Reptiles and Amphibians

ID-GAP mapped 15 amphibian species and 22 reptile species. Dr. Charles Peterson of Idaho State University has conducted research on the range and habitat for these species for 13 years (Peterson 1993, 1994, 1995a, 1995b). Two layers of information were used to map reptiles and amphibians that were not used for mammals and birds. Climate data from PRISM was used to refine the WHR models for 21 species known to have temperature parameters in habitat selection. In addition, inclusive buffers were created around wetlands and riparian for species with known affinity for water. Draft WHR models for reptiles and amphibians were reviewed and revised by C. Peterson and rerun as necessary. As part of this review process, we hand-edited each model so that habitat predictions would not extend beyond the defined hexagon range for each species. 

Step 4 – Coding Wildlife Habitat Relationships

Species habitat associations determined through literature searches and expert opinions were converted to codes consistent with the available GIS layers for ID-GAP. Four major thematic classes of GIS layers were available for use in approximating species habitat associations within their range (Table 2). Land cover was used to predict habitat for every species using the codes from Appendix C. Other layers were used as information was available that suggested their importance in determining suitable habitat. 

Use of Hydrologic Buffers for Modeling Species Habitats

Some species will use a variety of habitat types as long as they are within a certain distance of water. We used buffers of lakes, streams, major rivers, and wetlands to improve the accuracy of our habitat predictions for these species.

Often times, the exact distance that a species will inhabit away from the special feature is unknown. To accommodate this, we used standardized-distance buffers of 90m, 250m, 500m, 1km, and 5km. These buffers allowed us to make use of relative information (e.g., close or far) on the distance away from special features that a species would inhabit.

In the case of most amphibians, the transitional zone between aquatic and terrestrial ecosystems is most important. Thus they will use both water and upland cover types, but will not venture far away from the aquatic-terrestrial ecotone in either direction. To accommodate this, we created a series of buffers that included only a specified distance outside and inside of the perimeter of the water polygon. For example, the model for the Columbia spotted frog (Rana luteiventris) used a buffer of lakes that included terrestrial habitats up to 90m outside of the lake and portions of the lake up to 30m in from the lake perimeter. In doing so, we eliminated a large amount of potential commission error from the model.

Step 5 – Creation of Wildlife Habitat Relationship Models

The wildlife habitat relationship models for Idaho Gap Analysis were constructed using a series of step-wise algorithms that allowed for different input layers to be included for each species. Using an AML script, modeling parameters were retrieved from a database for each species. The first model layer for each species consisted of the land cover types identified by the model. The land cover model layer was then used as a template (or mask) for the next model layer (e.g., elevation), if appropriate. For example, elevations within the limits for a species would be selected from the appropriate land cover types. The newest model layer would then be set as a template for the next model parameter if necessary. This process continued until all parameters were included in the model.

Once a preliminary model was created, we restricted the prediction to only the area (in hexagons) estimated to be its range. Intermediate model steps for a species were not restricted to range limits approximated by hexagon maps. The process of clipping the model to the range limits involved several steps. First, hexagon records for the species were selected, made into a coverage and converted to a grid. This hexagon-range grid was used as a template (mask) and all predicted areas within it were selected. While this process is valid, it creates sharp, artificial edges of predicted habitat at the range boundaries. The commonly accepted practice in GAP is to extend predicted habitat beyond the range boundary to include continuous habitat patches (Csuti and Crist 1998). To do this, we first aggregated the intermediate, un-restricted model to a 1-km resolution. Next, we ran the Arc/Info Grid REGIONGROUP function to assign a unique number to each habitat patch. From here we selected all habitat patches that had at least 1 cell within the estimated range (Map 3.2).

Completed models were run through several steps to reduce file size and increase compatibility with programs not having raster data capability (e.g., older versions of ArcView). Models were converted to binary coding (1 for predicted, 0 for not predicted) and then made into geo-referenced TIFF images with compression. The models were also aggregated to 90m resolution for product delivery to the National GAP Office.

Step 6 – Model Review

Once completed, each model was reviewed by an expert on that species. Forty experts provided reviews of species’ habitat associations, coded models, references, and distributions by hexagon. Reviewers were asked to evaluate the completeness of all materials, suggest additions, and delete information not relevant to the distribution of that species in Idaho. The Idaho Gap Analysis project staff reviewed all final models and minor corrections were made.

We had difficulty finding experts willing to review some models. For 59 mammals and 39 birds either no adequate reviewer was identified, or reviewers were not willing to cooperate with the Idaho Gap Analysis project. These species were reviewed by the Idaho Gap Analysis project staff and compared to previous modeling efforts in Idaho and adjacent states. All reptile and amphibian models were reviewed.

Step 7 – Accuracy Assessment

Assessing the accuracy of the predicted vertebrate distributions is subject to many of the same problems as assessing land cover maps, as well as a host of other challenges related to both the behavioral aspects of species and the logistics of detecting them. These are described further in the Background section of the GAP Handbook. It is, however, necessary to provide some measure of confidence in the results of the gap analysis for each species (comparison to stewardship and management status), and to allow users to judge the suitability of the distribution maps for their own uses. Therefore we provided users with a statement about the accuracy of GAP predicted vertebrate distributions within the limitations of available resources and practicalities of such an endeavor. We acknowledge that distribution maps are never finished products, but are continually updated as new information is gathered. However, we feel that assessing the accuracy of their current iteration provides useful information about their reliability to potential users. We especially encourage wildlife biologists and amateur naturalists to treat the predicted distributions as testable hypotheses and engage the process of validation and iterative modeling. Our goal was to produce models that predict distribution of terrestrial vertebrates with an accuracy of 80% or higher. Failure to achieve this accuracy indicates the need to refine the data sets and models used for predicting distribution.

To get an indication of the omission errors of the WHR models, it was necessary to gather information on the actual (not potential) occurrences of the 379 species in Idaho. Since actual field inventory data is severely lacking for the majority of species, we used wildlife species lists from special management areas across the state (Csuti and Crist 1998). While this method is not without its shortcomings (e.g., many of the “independent” observations used to create the models came from the same managed areas, many of the experts assessing the models are also very familiar with the managed areas and may have even recorded the species observations there), it is the only feasible method of assessing accuracy for all (or at least most) of the WHR models across the entire state.

We contacted managers and biologists of special management areas in Idaho and requested lists of species observations on those areas. Of the 72 useable species lists we gathered, 62 areas were actually defined (or easy to define) in the stewardship layer. These areas included 6 state parks, 24 IDFG wildlife management areas, 6 USFWS National Wildlife Refuges, 3 National Parks or Monuments, 5 reserves owned by The Nature Conservancy, 11 USFS Research Natural Areas, and 7 other areas with extensive field surveys (e.g., Idaho National Engineering and Environment Laboratory, Snake River Birds Of Prey Area) (Map 3.3).

Each list was given a quality rating of low, medium, or high depending on completeness. Lists based upon potential occurrence were excluded and only areas with high and medium quality lists were used in the analysis. A high quality list had a complete (or near complete) list of species and included breeding status and/or abundance. Medium quality lists were only partial listings (e.g., just birds) and lacked either breeding status for birds or abundance for mammals, reptiles, and amphibians.

Species were included in the accuracy database if they were listed as, or it could be inferred that, the species was a regular breeder in the area. To be a “breeder” they had to be listed as 1) breeder or 2) summer, year-round, or permanent resident. When abundance was indicated, only abundant, common, or uncommon (not occasional or rare) species were included. All mammals, reptiles, and amphibians listed were included as regular breeders unless the list specifically mentioned “wintering”.

Because the species could have been detected anywhere within the management area, we used the entire management area as the analysis window to assess accuracy. If predicted habitat occurred within the analysis window, then a value of 1 was written to an output table; otherwise it remained 0. Since we were only using areas of species occurrences (value = 1), there were two possible outcomes of the computer programs, either the species was detected and predicted (correct present), or the species was detected but not predicted (omission error). We calculated the percent correct present for each model as the number of correct-present areas divided by the sum of the number of correct-present areas and the number of omissions.

IDVMD is a product of the 
Landscape Dynamics Lab
Idaho Cooperative Fish & Wildlife Research Unit
P.O. Box 444412
University of Idaho
Moscow, ID 83844-4412
Phone: 208-885-3774    Fax: 208-885-3021
email:jason@artemisia.wildlife.uidaho.edu

The information presented on this page was collected/created as part of the Idaho Gap Analysis project. While we have made every attempt to ensure its quality, we make no claims as to its accuracy and are not responsible for its use. If you have additional questions regarding this information, please fill out and submit a Comment Form. This page is best viewed with Microsoft Internet Explorer.