The marginal effects of the predictors on habitat suitability ofA. erythropus (i.e. occurrence probability responds to changes in a specific explanatory variable while other covariates are assumed to be held constant as mean) were presented in Figure 5. The response curves showed that the effects of environmental factors on the occurrence ofA. erythropus were strongly nonlinear.
For topographic variables, the probability of A. erythropuspresence declines with increasing elevation up to 500 m, with locations higher than 500 m elevation were virtually devoid of A. erythropus (Fig 5A). Also, the response curve of LDFG indicated that the geese prefer relatively flat sites (Fig 5I). In terms of bioclimatic variables, the probability of A. erythropus presence increases with precipitation of the warmest quarter to around 55 mm and mean temperature of the warmest quarter to around 14oC, after which there is a sharp decrease (Fig. 5B, Fig. 5D). Human disturbance also influences summering habitat, with suitability increasing the further the site is from human settlement (Fig. 5F). The response curve of habitat occurrence probability to distance from rivers (Fig. 5C) suggests that the geese were highly dependent on wetlands and riparian areas (Fig. 5C). Within the riparian zone, the summering habitat suitability decreases sharply with increasing distance from water courses, and after about 4.5 km virtually no birds are found.A. erythropus generally prefers land cover types waters (code 210) and shrubland (120; Fig. 5G). The modeling results suggest that the probability of occurrence increases with land productivity range (Fig. 5E) and homogeneity (Fig. 5H).