Developing and Applying Habitat Models Using Forest Inventory Data: An Example Using a Terrestrial Salamander

Abstract

We provide a framework for developing predictive species habitat models using preexisting vegetation, physical, and spatial data in association with animal sampling data. The resulting models are used to evaluate questions relevant to species conservation, in particular, comparing occurrence estimates in reserved and unreserved lands. We used an information–theoretic approach to develop and evaluate a priori models to predict the occurrence of the Del Norte salamander (Plethodon elongatus) within its geographic range on national forests in California. We then evaluated the association of P. elongatus to federal reserved lands using both an empirical and model-based assessment. For the model-based assessment, we calculated the probability of occurrence at existing Forest Inventory and Analysis (FIA) plots that we sampled for salamanders and those that were unsampled within our study area. The Del Norte salamander was more likely to be detected at plots with steeper slopes, older trees, more hardwood basal area, more canopy cover of conifers, more rock, and in areas receiving more precipitation and slightly warmer mean annual temperatures. Only the relationship of percent rock cover to probability of occupancy by P. elongatus was linear. Our best multivariate predictive model explained 66.2% of the deviance, and it correctly classified 96% of the plots at which P. elongatus was detected and 94% of the plots at which it was not. Ten-fold, cross-validation results revealed that the best model was relatively robust with correct classification rates of 87% and 89% for locations at which P. elongatus was detected and not detected, respectively. Our empirical results revealed no strong association with reserved lands. However, when we used our best model to estimate P. elongatus'; probability of occupancy at both sampled and unsampled plots, the mean probability of occupancy within reserved lands was greater than in unreserved lands, suggesting that reserved lands have higher-quality habitat relative to nonreserved lands. Overall, our results indicate that systematically collected forest inventory data can have significant value in developing wildlife habitat models when combined with samples of animal occurrence. Robust, empirically derived habitat models, such as the one we developed, may be useful tools for managers for monitoring the quantity, quality, and distribution of a species' habitat.