Optimizing habitat preference models of Azolla filiculoides (Lam.) [Azollaceae] for reducing ecological modelling complexity

An ecological study was done in Selkeh Wildlife Refuge (Anzali wetland, northern Iran) with the main objective of applying ecological modelling techniques for predicting habitat requirements of an invasive exotic aquatic fern species, i.e. Azolla filiculoides (Lam.). The applied database consisted of measurements from 7 different sampling sites in the wetland during the 2007-2008 period. The measured variables were a combination of water quality and structural habitat variables with cover percentage of A. filiculoides as output variable. In the present paper, we attempt to explore and compare the use of greedy stepwise (GS) and genetic algorithms (GA) as two optimizer techniques that can be combined with two data-driven models, i.e. classification trees (CTs) and support vector machines (SVMs) to select the most relevant input variables for predicting habitat preferences of A. filiculoides in the wetland. The predictive power of both data-driven models was assessed by two performance criteria, i.e. the percentage of Correctly Classified Instances (CCI %) and Cohen’s kappa statistics (k). Results show that after variable selection by both optimizers, the predictive performances of CTs were improved. In contrast, the predictive performances of SVMs weren't improved after optimizing the SVMs with GA and GS. However, GA outperformed GS in both CT and SVM, leading to reliable prediction. Results also show that both structural habitat and physico-chemical variables can affect habitat requirements of A. filiculoides in the wetland. However, the dependence of this aquatic fern on structural habitat parameters was very well confirmed by the CTs and SVMs. The outcomes of our study can help wetland managers and decision makers in order to set up successful wetland restoration/conservation and management programs