Search for Models
Class / /
Class / /
FREE TEXT SEARCH
Predicted Suitable Habitat Models
This page provides access to predicted suitable habitat model outputs and associated metadata for animal species resident in Montana for at least a portion of each year and for select plant species, focusing on Species Of Concern in the state.
As more models become available for plant and animal species,
they will appear in the search boxes on the left.
If you have need of a model for a particular species, please contact us.
Model Goals and Inputs
Two approaches are used for modeling predicted suitable habitat for most species; deductive and inductive.
Deductive models are simple, rule-based, associations with streams or ecological systems. The goal of the deductive models is to spatially represent the ecological systems commonly and occasionally associated with individual species during their primary/breeding season of occupancy (e.g., breeding for resident and summer migrants, winter for winter migrants, and migratory for solely migratory species) across each species' known breeding range in Montana. The assignments of common and occasionally associated ecological systems, and how those assignments were made, can be seen in a tabular form under the "Ecological Systems Associated with this Species" section of species accounts in the Montana Field Guide.
Inductive models are constructed using Maximum Entropy software (Phillips et al. 2006, Ecological Modeling 190:231-259) in conjunction with a variety of statewide biotic and abiotic layers standardized to 90 x 90-meter raster pixels and presence only data for individual species contributed to Montana Natural Heritage Program databases and filtered to ensure spatial and temporal accuracy and reduce spatial auto-correlation. The goal of inductive model outputs is to predict the distribution and relative suitability of habitat during the primary season of interest (usually breeding habitat, but overwintering habitat for winter migrants) at large spatial scales.
Model outputs are in the form of a logistic value that ranges from 0-1 with lower values representing areas predicted to be less suitable habitat and higher values representing areas predicted to be more suitable habitat. If enough observations were available to train and evaluate the models, the continuous output is reclassified into suitability classes - unsuitable, low suitability, moderate suitability, and high suitability for each 90 x 90-meter pixel. We then aggregated the classified model output into hexagons at a scale of 259 hectares per hexagon. We evaluated the output of the Maxent model with two metrics, an absolute validation index (AVI) (Hirzel et al. 2006, Ecological Modelling 199:142-152) and deviance (Phillips and Dudik 2008, Ecography 31: 161-175). Detailed descriptions of the environmental layers and observations used for modeling, the modeling process, and cutoffs used to designate habitat suitability classes are included with the write-ups for models of individual species.
For older plant models, outputs are provided as jpg images and as geo-referenced png image files available for download and viewable in ArcGIS. Predicted suitable habitat for those species whose habitat occurs across a large geographic area, but in small scattered or linear patches, may not always be readily visible on the jpg images and it is highly recommended that the PNG files be used to view areas of predicted suitable habitat for these species.
For animal models and more recently developed plant models, fully evaluated model outputs for both inductive maximum entropy and simple deductive habitat associations are provided in a pdf document. Inductive maximum entropy model outputs are available in GIS format from Braden Burkholder (see contact information to the left). Deductive model outputs can easily be constructed using the model write-up in association with the statewide Land Cover layer that can be downloaded via the Montana GIS Data List.
Models are based on statewide biotic and abiotic variables originally mapped at various spatial scales and standardized to 90 x 90-meter raster pixels. As a result, model outputs are not appropriate for use at smaller spatial scales. We recommend using model outputs at the scale of local landscapes using the classified model output aggregated into the 259 hectare hexagons; this is the finest spatial scale suggested for management decisions and survey planning. Evaluations of predictive accuracy and specific limitations are included with the metadata for models of individual species. Model outputs should not be used in place of on-the-ground surveys for species. Instead model outputs should be used in conjunction with habitat evaluations to determine the need for on-the-ground surveys for species.