Each of the 425 grids is named according to a modification of the species' element code, which is a 10-character unique identifier assigned by The Nature Conservancy. Codes beginning with AA indicate amphibians; AR, reptiles; AB, birds; and AM, mammals. To conform with 8.3 conventions when naming grids, element codes have been truncated by dropping the first and last characters; now, grid names beginning with A indicate amphibians; R, reptiles; B, birds; and M, mammals. These truncated names remained unique in all but one case, the Columbian Sharp-tailed Grouse subspecies. Here, the grid name has been adjusted to BNLC1304 to distinguish it from the grid for the Sharp-tailed Grouse species, BNLC1303. Please see Entity_and_Attribute_Overview, as well as MTVERT.DBF (dBase file) and MTVERT.DAT (INFO database), for a list of species names, element codes, and grid names.
Grids were constructed using an iterative five-step process: 1) determine species to be included in the modeling process; 2) delineate known ranges for each species using latilongs for birds and EPA hexagons for amphibians, reptiles, and mammals; 3) develop a Wildlife-Habitat Relationships (WHR) database to document associations between species and habitat features such as land cover, elevation, and distance to water; 4) prepare necessary GIS layers, then use a raster-based modeling approach to combine known distributions and WHR databases into predicted distributions for each species (see below); and 5) subject modeling rules and distribution maps to review by more than 50 biologists from around the state. After review, changes were made and the process was repeated. Once all predicted distributions were complete, species checklists for wildlife refuges and other management units were used to evaluate the accuracy of these maps. For 14 validation areas, mean accuracy was calculated as 59.0% for amphibians, 64.1% for reptiles, 67.2% for birds, and 55.0% for mammals. For more information, please refer to the the Montana Gap Analysis Final Report (Redmond et al. 1998; see below).
Each species was assigned to one MT-GAP staff member. Amphibians and reptiles were coordinated by Melissa Hart, but were modeled by a group of biologists, including Paul Hendricks (Montana Natural Heritage Program, MTNHP), Bryce Maxell (University of Montana), Chuck Peterson (Idaho State University), and the late Jim Reichel (MTNHP). Wendy Williams, Poody McLaughlin, Claudine Tobalske, and Melissa Hart were responsible for birds, and mammals were handled by Polly Thornton.
Primary inputs to the modeling process include land cover, topography, and hydrography, in addition to hexagon and latilong distribution maps. Of the 425 models developed, 136 were based solely on land cover, 73 on land cover and forest canopy, 28 on land cover and elevation, and 9 on land cover, elevation, and canopy. Many of the rest were based on some combination of land cover, canopy, elevation, and buffers on hydrographic features, although additional layers such as slope were used to a limited degree. Models were created using 5 AMLs (Arc Macro Language programs). These AMLs contain 19 different loops (some of which are repeated among AMLs) used to create output maps from a specified set of inputs (e.g., "land cover only" vs. "land cover plus elevation"). In addition to loops, 123 species had unique queries written for them. Models for all species, however, follow the same logic:
1. Create a grid delineating the known distribution based on either hexagons or polygons.
2. Generalize the edges of the known distribution using land cover polygons merged to 100 ha MMU. Assign each land cover polygon the highest probability of occurrence within its borders so that distributions are effectively extrapolated outward by one polygon, always favoring the highest code when land cover polygons span hexagons or latilongs.
3. Query layers for predicted habitat based on land cover types and other features. Code all areas selected as habitat using values from the extrapolated hexagon or latilong boundaries. Assign a special code to predicted habitat outside the known distribution to distinguish it from areas not predicted to be habitat.
4. Recode the output of Step 3 to reduce it from a range of values (probabilities of occurrence) to 1/0 values indicating a species' presence or absence. Eliminate areas with lower probability of species occurrence. For amphibians, reptiles, and mammals, keep hexagon codes 1-3 (confirmed-possible). For breeding birds, keep latilong codes 1-4 (breeding and possible breeding); for wintering birds, codes 5 and 7, and for migrating birds, codes 5 and 6.
For more detailed methods, results, and discussion, please see: Redmond, R.L., M.M. Hart, J.C. Winne, W.A. Williams, P.C. Thornton, Z. Ma, C.M. Tobalske, M.M. Thornton, K.P. McLaughlin, T.P. Tady, F.B. Fisher, S.W. Running. 1998. The Montana Gap Analysis Project: final report. Unpublished report. Montana Cooperative Wildlife Research Unit, The University of Montana, Missoula. xiii + 136 pp. + appendices. (Available digitally in Adobe Acrobat PDF format.)
For writeups of species-habitat relationships, key references, modeling rules, input data layers, and predicted distribution maps, please see: Hart, M.M., W.A. Williams, P.C. Thornton, K.P. McLaughlin, C.M. Tobalske, B.A. Maxell, D.P. Hendricks, C.R. Peterson, and R.L. Redmond. 1998. Montana atlas of terrestrial vertebrates. Unpublished report. Montana Cooperative Wildlife Research Unit, The University of Montana, Missoula. vii + 1302 pp. (Also available digitally in Adobe Acrobat PDF format.)
The following metadata elements are required by GAP, but do not parse using the FGDC ms parser (although similar elements can be found later in this document for several of these). For the convenience of GAP users, these elements are listed here. Data Set Identity: MT-GAP Predicted Vertebrate Distributions; Raster File Format: ARC/INFO GRIDs; Raster File Sensor: NA; Vector File Format: NA; Nonspatial File Format: NA; Source Distance Resolution: 90 meters; Raster File Number of Bytes per Pixel: 4; Native Data Structure: Raster.
The datasets comprising MT-GAP were created with the ARC/INFO Grid module running on IBM RS/6000 workstation computers (under AIX 4.1) with at least 128 megabytes of RAM and 4 gigabytes of local disk. The vertebrate dataset is large (1.125 gigabytes) and complex, made up of 425 separate grids with fairly high spatial resolution (90 m cell size). Because of this, software constraints prevented construction of a "hypercoverage" intersecting all predicted species distributions and habitat inputs on a cell-by-cell basis, which might have simplified use, especially queries. Dataset size and complexity also translates to potentially time-consuming queries and analyses on some computers, although powerful computers should not be necessary to process the data. For users without access to ARC/INFO, display and query should be feasible using ARC/VIEW Spatial Analyst software.
MT-GAP data were produced with the intent that they be analyzed and applied at the ecoregional level, that is, across geographic areas extending from several hundred thousand to millions of hectares in size. Because every occurrence of every species or habitat could not be mapped in the state, the data are best suited for coarse-filter applications (1:100,000+ scale), or for providing context for finer-level applications.
Appropriate uses include: 1. statewide biodiversity planning; 2. regional and large area resource planning; 3. coarse-filter evaluation of potential impacts or benefits of major projects/initiatives on biodiversity, such as utility or transportation corridors, wilderness proposals, open space or recreation proposals; 4. environmental impact assessment for large projects; and 5. education at all levels for both students and citizens.
It is far easier to identify appropriate uses than inappropriate ones. A "fuzzy line" eventually is crossed, however, when differences in resolution of these data, size of geographic area being analyzed, and precision of the answer required for the question are no longer compatible. Examples include: 1. use of the data to map small areas (less than thousands of hectares) that typically require mapping resolution at 1:24,000 scale and using aerial photographs or ground surveys; 2. combining these data with any data finer than 1:100,000 scale for analysis; 3. generating specific measurements from the data finer than the nearest thousand hectares; 4. establishing exact boundaries for regulation or acquisition; 5. establishing definite presence or absence of any element (e.g. vertebrate species); 6. determining abundance, health, or condition of any element; 7. establishing a measure of accuracy for any other data by comparison with these data; 8. using the data withouth acquring and reviewing not only this document, but also associated reports as cited in the Abstract above; and 9. altering the data in any way, then redistributing it as a GAP product.
For more information on accuracy assessment, please refer to the Montana Gap Analysis Final Report (Redmond et al. 1998; see Abstract for full citation). Of course, the accuracy of the overall dataset depends on the accuracy of its inputs, including species' known range extents (populated by latilong or hexagon), land cover, topography, and hydrography; the whole cannot be better than the sum of its parts.
The 425 terrestrial vertebrate species included in this dataset are considered to be complete for Montana. They were selected from roughly 565 terrestrial vertebrates known to occur in Montana. Many of those 565 species are rare or accidental migratory birds, found in the state only a handful of times, or other species with similarly uncertain occurrence in the state. For Montana Gap Analysis, we selected species known to breed in the state, and those that are regularly occurring non-accidentals. Decisions were made primarily based on state ranks assigned to each species by Montana Natural Heritage Program, although other sources were also consulted. For more information, please see the Montana Gap Analysis Final Report (Redmond et al. 1998; full citation provided in the Abstract).
Note that throughout the vertebrate modeling process, which was iterative, the inputs and outputs are intertwined. It can be hard to distinguish between the two in some cases: the steps below for the most part were ongoing tasks that occurred simultaneously. They only approached sequential order for the final modeling run, which took place in May 1998 after the review period (step 5) had ended.
For birds, distributional limits were mapped using an existing latilong grid system, P.D. Skaar's Montana Bird Distribution database (fifth edition, Montana Bird Distribution Committee, 1996). A copy of the database was acquired from Montana Natural Heritage Program and converted to GIS format. Our GIS layer differs from the published version in four ways: 1. all observations are mapped at the latilong level, collapsing quarter-latilong observations to this coarser scale; 2. observation types were combined into a new coding scheme devised for MT-GAP; 3. new observations were added by MT-GAP reviewers; 4. after the review period, an in-house update was conducted to fill in holes in species' distributions using professional judgment.
For each species, latilongs were coded as: 1. breeding and winter observations (B + W/w from the Skaar database); 2. breeding only (B); 3. possible breeding only (b); 4. possible breeding and wintering observations (b + W/w); 5. transient and wintering observations (t + W/w); 6. transient only (t); and 7. wintering only (W/w).
For amphibians, reptiles, and mammals, distributional limits were mapped using the EPA's EMAP hexagon grid system. MTNHP was contracted to compile species point locations and populate hexagons. Point observations were compiled from museum specimens, published literature, unpublished reports and theses, existing MTNHP databases, wildlife observations recorded by agency and independent biologists, and a U.S. Forest Service database on carnivores. Approximately 16,000 records were compiled. Next, an overlay analysis was used for initial population of hexagons; observations were buffered by their data precision, then overlayed with the hexagon coverage. Draft maps were then subjected to in-house (MTNHP) review, and later versions were reviewed by biologists around the state during the overall review period for vertebrate modeling.
The following confidence levels were assigned to hexagons for each species: 1. confirmed: >95% of an observation's precision-buffered area within the hexagon, or professional estimate of >95% probability that the species occurs within the hexagon; 2. probable: 80-95% of an observation's buffered area within the hexagon, or corresponding professional estimate; 3. possible: 10-80% of an observation's buffered area within the hexagon, or corresponding professional estimate; 4. historical: reported prior to 1950; 5. excluded: eliminated from species' range based on professional judgment (mostly observations deemed accidental; occasionally, detected species misidentifications).
The information found above -- grid names, TNC element codes, common names, and scientific names -- also can be found in database format in MTVERT.DBF (dBase file) and MTVERT.DAT (INFO file).