Modelling occurrence of roach “Rutilus rutilus” in streams

The selection of suitable variables is the first and important issue for fish habitat requirements and stream/river management programs. Classification trees (CTs) were used to study habitat preferences of the roach ( Rutilus rutilus ) in the Flemish stream basins (Belgium). The presence/absence of the roach was modeled based on a set of physicochemical and structural variables. The predictive performances of the CTs models were assessed based on the Cohen’s kappa statistics and percentage of correctly classified instances. An optimizer, genetic search algorithm (GAs), was combined with CTs to select the most important predictors for the fish habitat preferences. GAs improved the predictive performances of CTs and resulted in an easy interpretation of the models by making a selection of the variables that were used as input of CTs models. The models allowed to identifying the most important variables affecting habitat requirements of roach in streams. Based on the outcomes of models, the habitat preferences of roach in streams are mainly influenced by the habitat variables rather than water quality variables. It is proposed that CTs in combination with GAs can have high potential when applied for decision-making in the context of stream restoration and conservation management programs.