ForestERA Habitat modeling tools

home > tools > wildlife modeling

ForestERA Habitat Modeling Tools

Many different techniques have been used to create predictive models of taxonomic distributions across landscapes. Examples of commonly used techniques include Bayesian analysis, Environmental Envelopes (EEs), Classification and Regression Trees (CART), Generalized Linear Models (GLMs), Generalized Additive Models (GAMs), ordinations such as Canonical Correspondence Analysis (CCA), and Neural Networks (ANNs). For a comprehensive review and examples of these techniques, see Guisan and Zimmerman (2000). All of these techniques are statistically valid, and produce results that are as accurate as simple habitat models constructed from overlays or other non-statistical techniques. Major differences between modeling techniques exist, including the amount of data required, the types of predictive variables that can be used, outputs (e.g., presence/absence, probability of occurrence or abundance of the organism), statistical rigor, ease of interpretation of the models, and the amount of statistical and biological knowledge necessary to understand and apply the models. We considered all of these factors in selecting our modeling procedures, which are species specific.

Using a multi-criteria selection process and considerable input from stakeholders, we created a short list of important species or taxonomic groups for which we were to construct models. The list includes a variety of species that are representative of various functional or taxonomic guilds, and that respond to landscape level habitat patterns at different scales. We included species that are considered “management indicators” or “sensitive” because these species are most likely to be taken into account by managers when making treatment decisions. Finally, we considered the amount of data available for a species or group in selecting both the taxa to be modeled and the modeling procedure to be used. The following is a list of the taxa that were chosen for the Western Mogollon Rim (WMPALA) study area, and the procedures used to create models.

Northern Goshawk (Accipiter gentilis) and Mexican Spotted Owl (Strix occidentalis lucida)

We created models of potential nesting habitat for owls and goshawks using a set of simple rules and Mahalanobis distance. First, we determined a range of values for habitat conditions (basal area, tree density, canopy cover, slope, and aspect) under which owls and goshawks would nest, relying on the literature, management guidelines (Reynolds et al. 1992; USFWS 1995) and expert opinion. Drawing on georeferenced nest sites, coupled with information from our spatial data layers, we then determined thresholds, within the range of habitat conditions, that could be used to separate areas with and without nests. As a second step we employed the Mahalanobis distance (M-distance) statistic to develop a measure of the likelihood that a given portion of the landscape would be used for nesting. M-distance is a measure of dissimilarity between two multivariate datasets (Farmer & Kadmon 2003). In the case of wildlife habitat modeling, one dataset is the mean vector of habitat characteristics for a set of locations used by a species (in this case nest sites), while the other dataset is the range of conditions across the entire landscape (Clark et al. 1993; Farmer & Kadmon 2003). The first vector is usually assumed to represent “preferred” habitat conditions for a species (Farmer & Kadmon 2003). Thus, the “distance” value can be used as an index of habitat preference. The following scientists collaborated on this modeling effort: Bill Block, Joseph Ganey, and Jeff Jenness of the USDA Forest Service Rocky Mountain Research Station; Paul Beier of Northern Arizona University, and Micheal Ingraldi of the Arizona Game and Fish Department.

Predicted northern goshawk nesting habitat

Predicted mexican spotted-owl nesting and roosting habitat

Tassel-eared Squirrel (Sciurus aberti)

We worked in collaboration with Norris Dodd of the Arizona Game and Fish Department to create the models for the Tassel-eared squirrel. Based on his data and expertise (Dodd et al., 1998; Dodd et al., 2003), we applied GLM regression techniques (Hilbe 1994) to establish relationships between our forest structural data layers and squirrel density and recruitment. This resulted in several possible models for each. We then used an information-theoretic approach to select the best models to use for each attribute. A model that included only local basal area as an input variable was best for squirrel density. For squirrel recruitment, a model that included local basal area, canopy cover over a 165 ha spatial extent, and an interaction effect worked best. The relationships between forest structure and population attribute in both of these models were highly significant (density n = 18, P < 0.001, r² = 0.85; recruitment n = 18, P < 0.001, r² = 0.72).

Passerine Birds

We used Classification and Regression Tree (CART; Breiman et al. 1984) models to predict passerine avian species richness and to create individual probability of presence models for 10 species of birds across the western Mogollon Plateau. CART procedures have proven to be very useful in ecological contexts because 1) both continuous and discrete predictive variables can be used in the models, 2) the models are statistically rigorous, and 3) the outputs are easily understood (De’ath and Fabricius 2000). Because they are nonparametric and divide datasets into independent groups, CART models have several additional advantages over other techniques: input data does not need to be normally distributed; it is not necessary for predictor variables to be independent; and relationships between predictor variables and observational data are modeled well when relationships between them are not linear. In published studies, CART models have performed well when compared to models created by regression (Anderson et al., 2000; De’ath and Fabricius 2000, Dettmers et al. 2002) and EE (Skidmore et al. 1996) techniques. We are collaborating with a number of investigators, including Bill Block of the Forest Service, Brett Dickson of Colorado State University, Carol Chambers of NAU, Micheal Ingraldi and Steve Rosenstock of the Arizona Game and Fish Department, and Kerry Griffis-Kyle of Syracuse University. They have provided data and expertise to guide our modeling efforts. We are now in the process of checking the accuracy of those models. One way to do that is to compare our results to the existing literature on avian habitat requirements (e.g., Finch and Block, 1997).

Pronghorn (Antilocapra americana)

Because few studies have overlapped, both spatially and temporally, with our study, we built some models using sets of rules created through literature review and consultation with experts. We created a predicted habitat suitability layer, using published literature on general Pronghorn habitat requirements in Arizona (Lee et al. 1998, Ockenfels 1994). The layer was built using simple rules with slope and canopy cover as predictor variables. Values for the habitat suitability index range from 0 to 1, and are unitless. It is important to note that habitat suitability does not represent habitat quality, which would require including some measure of survivability or reproductive success. We are using spatial data from pronghorn radiotracking studies, conducted by Rick Ockenfels (Arizona Game and Fish Department) on the Kaibab National Forest, to test the model.

In response to recommendations from the wildlife breakout group at the May 2004 WMPALA workshop, we created a layer that proposes possible corridors to facilitate Pronghorn movement between large patches of existing habitat. Because many areas that were formerly used as corridors by pronghorn are now overgrown with trees, high quality patches of Pronghorn habitat are becoming isolated. We set out to identify areas where forest thinning could result in increased connectivity between patches of high quality Pronghorn habitat. This was a three-step process. First, we identified large patches (> 40ha) of existing Pronghorn habitat, including all areas that exhibit a habitat suitability index of 0.7 or greater, and were larger than 40 ha. Second, we identified areas that would be unsuitable as corridors for pronghorn. These areas included Mexican Spotted Owl Protected Activity Centers, Northern Goshawk nesting stands, areas with slopes greater than 20 degrees, private property, and areas within 1/4 mile of major highways. In the third step, we digitized potential corridors between habitat patches that were less than 5km apart. We felt that more isolated patches were unlikely to be connected. We did not place corridors across any areas identified as unsuitable for pronghorn, but did place corridors between habitat patches along paths that would provide the highest quality habitat available between those patches. All corridors have a width of 450m (~ 1/4 mile).

Merriam’s Wild Turkey (Melagris gallipavo merriami)

A model of turkey habitat suitability was constructed using published literature and expert opinion. Habitat suitability was based on the suitability of the habitat for roosting, because Wild Turkeys are typically found in close proximity to roost sites. Inputs for this model are slope, basal area, and canopy cover. As each of these factors increases, the habitat suitability index increases. As with the pronghorn habitat model, the values for the suitability index range from 0 to 1 and are unitless. We obtained information on Wild Turkey habitat requirements in Arizona from the following literature sources: Mollohan et al. (1995), Wakeling (1991), and Wakeling and Rodgers (1995). We also obtained spatial data from turkey surveys conducted by Brian Wakeling (Arizona Game and Fish Department) on the Coconino National Forest. These data are being used to test the accuracy of our model.

References

Anderson, M. C., J. M. Watts, J. E. Freilich, S. R. Yool, G. I. Wakefield, J. F. McCauley, and P. B. Fahnestock. 2000. Regression-tree analysis of desert tortoise habitat in the central Mojave Desert. Ecological Applications 10: 890-900.

Block, W. M. and D. M. Finch. 1997. Songbird ecology in Southwestern Ponderosa Pine forests: a literature review. USDA Forest Service General Technical Report RM-GTR-292.

Block, W. M., M. L. Morrison, and M. H. Reiser. 1994. The Northern Goshawk: ecology and management. Studies in Avian Biology 16. Cooper Ornithological Society.

Breiman, L., J. J. Friedman, R. A. Olshen, and C. J. Stone. 1984. Classification and Regression Trees. Wadsworth and Brooks Publishing, Monterey, California.

Clark, J. D., J. E. Dunn, and K. G. Smith. 1983. A multivariate model of female black bear habitat use for a Geographic Information System. Journal of Wildlife Management 57: 519-526.

De’ath, G. and K. E. Fabricius. 2000. Classification and Regression Trees: a powerful, yet simple, technique for ecological data analysis. Ecology 81: 3178-3192.

Dettmers, R., D. A. Buehler, and J. B. Bartlett. 2002. A test and comparison of wildlife-habitat modeling techniques for predicting bird occurrence at a regional scale. Pp. 607-616 in Predicting Species Occurrences: Issues of Accuracy and Scale. J. M. Scott, P. J. Heglund, M. L. Morrison, J. B. Haufler, M. G. Rapheal, W. A. Wall, and F. B. Sampson, eds. Island Press, Washington D. C.

Dodd, N. L., S. S. Rosenstock, C. R. Miller, and R. W. Schweinsburg. 1998. Tassel-eared Squirrel population dynamics in Arizona: index techniques and relationships to habitat conditions. Arizona Game and Fish Department Research Technical Report #27.

Dodd, N. L., J. S. States, and S. S. Rosenstock. 2003. Tassel-eared squirrel population, habitat condition, and dietary relationships in north-central Arizona. Journal of Wildlife Management 67:622-633.

Farmer, O. and R. Kadmon. 2003. Assessment of alternative approaches for bioclimatic modeling with special emphasis on the Mahalanobis distance. Ecological Modelling 160: 115-130.

Guisan, A. and N. E. Zimmermann. 2000. Predictive habitat distribution models in ecology. Ecological Modelling 135: 147-186.

Hilbe, J. M. 1994. Generalized linear models. The American Statistician 48: 255-265.

Kaufman, N. M. 1995. Recovery plan for the Mexican Spotted Owl (Strix occidentalis lucida). USDA Forest Service and USDI Fish and Wildlife Service.

Lee, R. M., J. D. Yoakum, B. W. O’Gara, T. M. Pojar, and R. A. Ockenfels. 1998. Pronghorn Management Guides. Eighteenth Biennial Pronghorn Antelope Workshop. Arizona Antelope Foundation.

Mollohan, C. M., D. R. Patton, and B. F. Wakeling. 1995. Habitat selection and use by Merriam’s Turkey in northcentral Arizona. Arizona Game and Fish Department Research Technical Report #9.

Ockenfels, R. A., A. Alexander, C. L. D. Ticer, and W. K. Carrel. 1994. Home ranges, movement patterns, and habitat selection of Pronghorn in central Arizona. Arizona Game and Fish Department Research Technical Report #13.

Reynolds, R. T., R. T. Graham, M. H. Reiser, L. Bassett, P. L. Kennedy, D. A. Boyce Jr., G. Goodwin, R. Smith, and E. L. Fisher. 1992. Management Recommendations for the Northern Goshawk in the Southwestern United States. USDA Forest Service General Technical Report RM-GTR-217.

Skidmore, A. K., A. Gauld, and P. Walker. 1996. Classification of kangaroo habitat distribution using three GIS models. International Journal of GIS 10: 441-454.

Wakeling, B. F. 1991. Population and nesting characteristics of Merriam’s Turkey along the Mogollon Rim, Arizona. Arizona Game and Fish Department Research Technical Report #7.

Wakeling, B. F., and T. D. Rodgers. 1995. Winter habitat relationships of Merriam’s Turkeys along the Mogollon Rim, Arizona. Arizona Game and Fish Department Research Technical Report #16.