Local species distributions are often geographically restricted to a subset of environmental conditions across a species’ full range, complicating forecasting climate warming effects. However, Bayesian species distribution models (SDM) can leverage geographically restricted datasets with broader knowledge of habitat relationships across the species’ range, to forecast climate vulnerability in data-limited regions. Principles of niche tracking and niche expansion were explored using an innovative Bayesian SDM approach to refine a climate vulnerability assessment for bull trout (Salvelinus confluentus), a cold-water riverine fish. The SDM was fit to a large, spatially dense fish occurrence and stream temperature dataset to model how climatic and geomorphic factors influence the current and future distribution of bull trout near its northern range extent. To assess niche tracking, wherein modelled relationships were based on observed occurrence patterns, we fitted the SDM with uninformative priors. For niche expansion, which assumes the population can adjust to occupy a warmer niche like southerly populations, we added an informative prior for summer stream water temperature occupancy. Models projected effects of warming on the distribution of suitable habitat using Representative Concentration Pathways 4.5 and 8.5 emissions scenarios for 2061-2080. Bull trout distribution was patchy and limited to intermediate thermal and slope conditions in streams with high groundwater contributions. The latter is a key determinant of biogeographic patterns not seen elsewhere across the species’ range. Under niche tracking, suitable habitat extent is projected to decline by 36-46%, while under niche expansion, suitable habitat could increase by 25-28%. The large dichotomy between projections illustrates the importance of considering local features and adaptive capacity when forecasting potential responses of cold-water fishes to climate warming. It also highlights a need for studies to better understand the mechanisms that may prevail as species distributions shift this century.

Field surveys were conducted across approximately 400 sites with temporal replicates (i.e., sites visited on two seperate occassions within a season). Presence-absence data across sampling sites was paired with geomorphic and climatic covariates derived through a GIS platform. Climatic variables were derived from Spatial Statistical Stream network models.