Evaluating Effects of Model Type and Importance of Habitat Parameters on Prediction of Biodiversity Indices (A Case Study with Fishes of the Totkabon River in the Southern Caspian Sea)

Habitat models are applied to estimate species abundance and diversity using habitat parameters that can be used in exploitation and conservation of many aquatic species. Habitat models are effective tools in estimating species abundance and diversity of aquatic communities using habitat parameters has high importance in exploitation and conservation of aquatic resources. Identifying models with the best performance and habitat parameters with the highest importance, would affect appropriate utilisation of this tools and in making optimal management decisions. The present study evaluated performance of four models (multiple linear regression, partial least square regression, support vector machines and random forest) in prediction of biodiversity indices in fishes of a river in the southern Caspian Sea. In addition, importance of environmental parameters in prediction of those indices were calculated. Multiple linear regression and partial least square regression had weak performance. Support vector machine and random forest had the best performance. Various environmental parameters had varying importance across the examined models. In conclusion, support vector machines and random forest are suggested as suitable models for prediction of biodiversity indices of the southern Caspian fishes.