DIGITALNA ARHIVA ŠUMARSKOG LISTA

prilagođeno pretraživanje po punom tekstu

ŠUMARSKI LIST 3-4/2017 str. 37 <-- 37 --> PDF |

the variables with variance inflation factor > 3 (Zuur et al. 2009). After multicollinear variables were excluded, the final selection of variables was made for both types of models (Table 1). We calculated all possible models and explored the structure of all candidate models with ΔAIC scores ≤ 2 and used them for model averaging to obtain robust parameter estimates (Burnham and Anderson 2002). All statistical analyses were conducted with R ver. 3.0.2.RESULTS REZULTATI According to the Akaike information criterion, red deer presence/absence is best explained by two logistic regression models (ΔAIC≤2). The models predict that the probability of red deer presence depends on share of forest (FOREST, positive correlation), cost distance (COST_DIST, negative correlation) and size of largest forest patch (F_PATCH, positive correlation); the second best model also includes density of forest edge (F_EDGE, negative correlation; Table 2). According to the Akaike information criterion, local red deer densities are best explained by four models with ΔAIC≤2 (GLM) including following variables: distance to |