Mapping seasonal habitat suitability for the Gunnison sage-grouse in southwestern Colorado, USA: species distribution models using maximum entropy modelling and autoregression

Abstract

Gunnison sage-grouse (Centrocercus minimus) populations have declined dramatically over the last century as a result of habitat loss and fragmentation throughout the Western sagebrush ecosystem in the US. GIS-based species distribution models (SDMs) are a powerful tool in helping conservation managers map critical habitat and understand factors affecting distribution. However, traditional modeling methods may not properly account for the inherent population structure of the species resulting from behavioural patterns or dispersal limitations. This research uses two methods which have recently been applied to ecological modelling to overcome these limitations: maximum entropy modelling (MaxEnt) and simultaneous autoregression (SAR). The study focuses on the Piñon Mesa Gunnison sage-grouse population in southwestern Colorado, using radio telemetry locations of 69 transplanted birds collected from 2010 to 2013. A two-step model was developed which input MaxEnt outputs into SAR models to identify fine-scale vegetation structure variables which affect suitability. Independent seasonal models were calculated to reflect habitat requirements at critical life stages. Elevation and distance from woodland were the most important predictors for all seasons. Breeding season and summer/autumn distributions were found to correspond strongly to elevation and associated changes in herbaceous cover, while winter models were affected strongly by terrain smoothness, mixed shrub cover, and reduced perennial grass understory. There were critical differences between SAR models and non-spatial generalized linear models, suggesting that accounting for spatial structure in the population is important for accurate modelling of sage-grouse. Model results also raise questions about the significance of elevation to sage-grouse populations, and how these affects may violate assumptions of model stationarity.