All statistical analyses were performed using XLSTAT 2011 software (Addinsoft, Paris, France). Predictor variables that could affect detectability of larval and adult salamanders were collected during each visit. We measured the duration of survey (min), air temperature (°C), humidity (%) and classified the weather conditions (1 = rainy, 2 = cloudy, 3 = sunny). For site-occupancy modelling (MacKenzie et al., 2002), all predictor Metformin mouse variables for the occupancy and detection probability were entered and normalized in PRESENCE 4.4 (available at http://www.mbr-pwrc.usgs.gov/software/presence.html). In PRESENCE, we performed a goodness-of-fit test on the global model,
which included all occupancy predictors for S. salamandra and only terrestrial habitat predictors for S. atra (Table 1; MacKenzie & Bailey, 2004). The test (n = 5000 bootstrap samples) showed that the global model did not fit the data well because there was overdispersion (S. salamandra: χ2 = 13.57, P > 0.05, ĉ = 2.2263; S. atra: χ2 = 18.18, P < 0.01, ĉ = 7.1953). Therefore, we used quasi-likelihood Akaike information criterion (QAIC) instead of AIC for model selection and adjusted standard errors (Burnham & Anderson, 2002; MacKenzie & Bailey, 2004). We also calculated the Akaike weight of a model, which is the probability that a model has the best
explanatory power among the candidate models. The goal of site-occupancy modelling in PRESENCE was to find the set of predictors that http://www.selleckchem.com/screening/natural-product-library.html adequately describes the observed data while accounting for imperfect detection (MacKenzie et al., 2002). A intercept-only model with no covariates medchemexpress for neither occupancy nor detectability [model ψ(.) p(.)] showed that detection probabilities were high. Under this model, the estimated detection probability p for S. salamandra was 95.8 ± 2.1% (mean ± se) and 85.9 ± 4.2% for S. atra.
Given three to four visits per site, we could be highly confident that we detected both species where they were present (Kéry & Schmidt, 2008). For this reason, we did not use two-species occupancy models (e.g. MacKenzie et al., 2004). For further occupancy modelling, we only varied the predictor variables for site occupancy and always used a model with constant detection probability [i.e. model ψ(predictor variables) p(.)]. We used an a priori candidate model selection approach (Burnham & Anderson, 2002). Because sample size was small, we considered only a small set of candidate models and decided to keep candidate models simple (Anderson et al., 2001; Anderson & Burnham, 2002). We used identical sets of candidate models for both species except that we did not fit models with stream characteristics to the S. atra data. We fitted a total of 23 models to the S. salamandra data and 17 models to the S. atra data (see Supporting Information Tables S1 and S2). We defined models with a single predictor variable to assess their effect on the species occurrence [Table 1; e.g. ψ(area) p(.)].
Related posts: