Telling the North American beaver tale: modelling Castor canadensis distribution in Mi'kma'ki (Nova Scotia, Canada)

Abstract

The American beaver (Castor canadensis) is a keystone species of both significant ecological and biocultural importance in Mi'kma'ki (Nova Scotia). In North America, several environmental covariates are known to influence C. canadensis habitat selection, including distance to watercourse, stream gradient, and distance to preferred hardwoods; however, specific distances and species vary greatly throughout their continental range, prompting the need for local study. In Nova Scotia, occurrence data have never been systematically collected as C. canadensis is a common and unthreatened species, resulting in geographic knowledge gaps. This thesis remedied this knowledge gap by using Maximum Entropy modelling (Maxent) to create a species distribution model using environmental conditions present in areas of known occurrences to predict areas of C. canadensis distribution across the province. Input layers consisted of predominantly citizen-science occurrence data, and environmental covariates which characterized geomorphology and forest composition. Correlation analysis and reverse stepwise elimination were used to generate two models: an ecological model, and a human footprint model, where the latter investigated the influence of anthropogenic disturbance. Each model was an average of 10 replicates, using 500 iterations, 10,000 background pseudo-absence points, and a jackknife test to measure variable importance. The ecological model produced a high averaged area under the receiver operating characteristic curve (AUC) for the replicated runs (0.80 +/- 0.02), where the strongest contributors to distribution based on the permutation importance were 'Watercourse' (26.6%), 'Elevation' (15.1%), 'Red Oak' (11.8%), 'Aspen' (11.4%), and 'Gray Birch' (9.9%). Response curves indicated proximity to watercourses, low elevation, proximity to aspen, and distance from red oak and gray birch were important habitat associations. The human footprint model showed a positive relationship between occurrence points and human footprint, likely due to citizen-science collected occurrence data. Areas of known historic usage overlap at a landscape scale with distribution. The habitat association findings are consistent with previous studies suggesting watercourses, elevation and specific hardwood tree species are the main drivers of distribution, highlighting important areas of the predicted distribution of C. canadensis in the Wabanaki-Acadian Forest. Modelling this distribution is a practical conservation and management tool, which can contribute to future efforts to map biocultural connectivity in Mi'kma'ki (Nova Scotia), inform land use management, and provide insight into future species' sampling needs. Keywords: American beaver, Castor canadensis, SDM, Maxent, Nova Scotia