g. model ψ(area + AS) p(.) for S. salamandra]. In addition to these models, we set up candidate models with combinations of predictor variables. The first model describing the terrestrial habitat included the predictors ‘area’, ‘forest’, ‘slope’ and ‘PCA climate’ [model ψ(habitat) p(.)]. The second model, which was only used for the S. salamandra data, included the predictors ‘slope’, ‘stream bank slope’, ‘pools’ and ‘hides’ to assess VX-765 solubility dmso the effect of stream parameters on the species’ occupancy probability [model ψ(stream) p(.)]. Two more candidate models were obtained by adding the presence of the other species to the two multi-variable models.

Based on the results of the a priori models for each species, we additionally combined the predictors of the QAIC best ranked models into

four new a posteriori candidate models with combination of two or three predictor variables (see Supporting Information Tables S1 and S2). Because there was a model selection uncertainty, we used model Inhibitor Library datasheet averaging techniques for parameter estimation (Burnham & Anderson, 2002). For model averaging, models with ΔQAIC >7 were dropped from the set of candidate models for each species and Akaike weights were recalculated for the set of models with ΔQAIC ≤7. Based on the new Akaike weights, model averaging was performed for all predictor variables in models that were retained in order to assess their effect on the species’ occupancy probability. During field surveys (mean duration per visit ± standard deviation was 53.8 ± 14.5 min for Zug; 46.4 ± 14.0 min for Nidwalden), we detected Salamandra salamandra at 16/23 of

the sampling sites in the contact zone in Zug and 13/19 in Nidwalden. Salamandra atra was found at 5/23 of the sampling sites in Zug compared with 17/19 in Nidwalden. Co-occurrence of the salamanders was found at 3/23 sampling site in Zug and at 12/19 sites in Nidwalden. Table 2 shows the top-ranking models (based on QAIC) for both salamander species. For both species, top-ranking models always included ecological predictor variables and were better than the intercept-only ADP ribosylation factor models (i.e. null models). The analysis revealed that the model including ‘slope’ and ‘pools’ as predictors for the fire salamander’s occupancy probability was best supported by the data. Model averaging showed that only the 95% confidence interval of ‘slope’ did not include zero (Table 3). The positive effect of the slope of the sampling sites on the occupancy probability is shown in Fig. 2. The confidence intervals of all other predictor variables included zero. In particular, while the estimated effect of alpine salamander on fire salamander occupancy was negative, the 95% confidence interval included zero (Table 3). The observed data for S. atra were best explained by the model with the predictor variable ‘area’. In this model, we estimated a four times lower occupancy rate for S. atra in Zug (0.22, se 0.