Over the past two centuries, persecution and habitat loss caused grizzly bears (Ursus arctos) to decline from a population of approximately 50,000 individuals to only 4 fragmented populations within the continental United States. In recent decades, these populations have increased and expanded in size and range due to collaborative conservation efforts and protections under the Endangered Species Act. Today, population estimates exceed 1000 animals each in the Northern Continental Divide Ecosystem (NCDE) and Greater Yellowstone Ecosystem (GYE). The Selkirk Ecosystem (SE) has approximately 50 grizzly bears, and augmentations into the Cabinet-Yaak Ecosystem (CYE) helped boost the population to an estimated 50 – 60 animals. To date, the Bitterroot (BE) and North Cascades Ecosystems (NCE) lack any known permanent residents.
Eventual connectivity between populations is a conservation goal, as is establishment of populations in currently unoccupied recovery areas. An understanding of habitat selection by grizzly bears within existing populations is crucial for predicting potential linkage zones and suitable habitat. A second urgent conservation challenge is identification of areas where grizzly bears are likely to disperse among recovery ecosystems, and proactive efforts to reduce human-grizzly bear conflicts. Our overall objective in this study was therefore to model grizzly bear movement, habitat use, and population connectivity to identify specific areas that are important for habitat use and natural connectivity among recovery ecosystems. We developed a multi-phase approach to accomplish these goals.
In phase 1, we aimed to increase understanding of how males and females use habitat within the NCDE. We employed multiple stages to test hypotheses of behavior, use newly gained knowledge to mechanistically simulate individual movements, translate results to predictive habitat maps, and test their predictive power across a large scale. Mechanistically modeling grizzly bear movements using integrated step selection functions (iSSFs) for GPS-collared grizzly bears (F = 46, M = 19) demonstrated that grizzly bears have highly individualistic spatial behaviors. Some individuals avoided whereas others preferred areas of vegetation green-up, terrain ruggedness, forest edge, riparian areas, building densities, and secure habitat. Such individualism supported the need for an individual-based modeling approach to understand and predict grizzly bear behavior. External validation demonstrated high predictive accuracy with mean Spearman rank scores of >0.90 across seasons and years, and overall scores of 1.0. The top 5 classes of our predictive habitat maps contained 73.5% of female fixes and 83.6% of male fixes, and the top class (comprising 10% of the mapped area) contained 25.6% and 41.7% of female and male fixes, respectively. Results of this phase of our research provide tools for conservation planning and served as the basis for sequential phases of our research.
In phase 2, we tested whether our iSSFs developed for NCDE bears in phase 1 were transferable to the SE, CYE, and GYE. We simulated 100 replicates of 5,000 steps for each iSSF in each ecosystem, summarized relative use into 10 equal-area classes for each sex, and overlaid GPS locations from bears in the SE, CYE, and GYE on resulting maps. Spearman rank correlations between numbers of locations and class rank were ≥0.96 within each study area, indicating models were highly predictive of grizzly bear space use in these nearby populations. Assessment of models using smaller subsets of data in space and time demonstrated generally high predictive accuracy for females. Although mostly high across space and time, predictive accuracy for males was low within some watersheds and in summer within the SE and CYE, potentially due to seasonal effects, vegetation, and food assemblage differences. Altogether, these phase 2 results demonstrated high transferability of our models to landscapes in the Northern Rocky Mountains, suggesting they may be used to evaluate habitat suitability and connectivity throughout the region to benefit conservation planning.
In phase 3, we simulated connectivity pathways for grizzly bears between recovery ecosystems in the Northern Rockies. Building on phases 1 and 2, we modeled movements to identify potential pathways for dispersal, using the iSSFs developed in phase 1. We applied the models in a >300,000 km2 area including the NCDE, CYE, GYE, and BE. First, we simulated directed movements (randomized shortest paths with 3 levels of exploration) between start and end nodes for routes between populations. Second, we simulated undirected movements from start nodes in the NCDE, CYE, or GYE (no predetermined end nodes). We summarized and binned results as classes 1 (lowest relative predicted use) - 10 (highest relative predicted use) and evaluated predictions using 127 outlier grizzly bear locations. Values at outlier locations indicated good model fit and mean classes at outlier locations (≥7.4) and Spearman rank correlations (≥0.87) were highest for undirected simulations and directed simulations with the highest level of exploration. Our resulting predictive maps will facilitate on-the-ground application of this research for prioritizing habitat conservation, human-bear conflict mitigation, and transportation planning. Additionally, our overall modeling approach has direct utility for myriad species and conservation applications.
Additional phases of this work are ongoing.