Multi-state deer occupancy
Reproductively successful deer – does with fawns – were estimated to be widespread across the study area in spring 2012 (ψb = 0.89, SE = 0.14), 2013 (ψb = 0.98; SE = 0.02) and 2014 (ψb = 0.95; SE = 0.03) when modelled without landscape covariates. The estimated probability of false absences for deer with fawns was ≤ 0.002 in all years, suggesting that we reliably detected fawns when they occurred.
Anthropogenic landscape features best explained conditional probability of fawns given occupancy by deer (R ) in 2012 (well sites and seismic lines, cumulative AICw = 0.83), 2013 (seismic lines, AICw = 0.84) and 2014 (industrial features, forest cutblocks and total footprint, cumulative AICw = 0.81) (Supplementary Information Table S2). Models in which breeding varied only with natural vegetation, or was invariant, were not supported. Hence, occupancy of breeding deer differed from that of non-breeding deer, and varied with the area of anthropogenic features across the oil sands landscape. However, multistate occupancy models contained unresolvable “border estimates” for R (those close to 0 or 1; (MacKenzie et al. 2017)) and exhibited problems with model convergence, necessitating a companion approach.