Modelling habitat preference of an alien aquatic fern, Azolla filiculoides (Lam.), in Anzali wetland (Iran) using data-driven methods

A crucial element in modelling habitat requirements of any alien plant species is selection of the most important predictive variables. A database consisting of measurements collected at 7 different sampling sites at Selkeh Wildlife Refuge (Anzali wetland, Iran) was applied to predict the habitat preferences of an exotic species, Azolla filiculoides (Lam.). The measured variables were a combination of physico-chemical, structural-habitat and cover percentage data of A. filiculoides collected during the 2007-2008 period. We used support vector machines (SVMs) combined with two search algorithms, i.e. genetic algorithm (GA) and greedy stepwise (GS) in order to select the most important explanatory variables for the target species. The models with the best performing exponent were run five times after randomization to check the models' robustness and reproducibility. The results of paired Student's t-test showed that the two optimizers (GA and GS) were unable to improve the predictive performances of the SVMs. Yet, GA outperformed GS resulting in a better prediction. All applied methods showed that both structural-habitat and physico-chemical variables might play key roles for meeting the habitat preferences of the exotic fern in the wetland. However, structural-habitat parameters (particularly wetland depth and air temperature) were the most predictive ones. Among the water quality variables, orthophosphate and sulphate were also recognized as important predictors. The physico-chemical variables selected by the models revealed that reduction of industrial pollution loads and also decreasing nutrient and organic pollution inputs into the wetland could be effective way to reduce the growth of Azolla's population.