Ensemble Modelling of Sensitive Stream Fish Species Distributions in Iran: Expanding Knowledge to Aid Species Conservation
Hossein Mostafavi, Ziya Kordjazi, Roozbeh Valavi, Hossein Shafizadeh-Moghadam, Jafar Kambouzia, and Dana M. Infante
Abstract.—Species distribution models are important tools for conservation and management of aquatic ecosystems. In this study, nine fish species (Caspian Lamprey Caspiomyzon wagneri, Acanthalburnus urmianus, Alburnoides namaki, Capoeta buhsei, Mangar Luciobarbus esocinus [also known as Barbus esocinus], Luciobarbus xanthopterus, Mesopotamichthys sharpeyi, Glyptothorax silviae, and Iranocichla hormuzensis) that are sensitive to habitat changes induced by human activities were predicted by species distribution models throughout rivers in Iran. The fish data used cover several time periods (1970–2000) obtained from databases originating from field sampling, several museums, and the literature. We considered seven environmental variables, including channel slope, bank-full width, wetted width, elevation, mean air temperature, range of air temperature, and annual precipitation to model distributions of all nine species using an ensemble forecasting approach. Models used included generalized linear models, generalized additive models, classification tree analysis, artificial neural networks, surface range envelopes, boosted regression trees, random forest, multivariate adaptive regression splines, and flexible discriminant analysis. Additionally, we compared known distributions of species with modeled distributions, and we used the models to identify potential habitats for the nine species outside previously sampled areas. True skill statistic for each species was, on average, greater than 0.80 (i.e., excellent). Moreover, whereas surface range envelopes for all species had the lowest performance, random forest and generalized boosting methods had the highest performance. Among species studied, Caspian Lamprey, Luciobarbus xanthopterus, and Mesopotamichthys sharpeyi were predicted only in basins where they had been previously detected. In contrast, other species (i.e., Acanthalburnus urmianus, Alburnoides namaki, Capoeta buhsei, Mangar, Glyptothorax silviae, and Iranocichla hormuzensis) were predicted not only in basins with previous records, but also in new basins. These results deepen our understanding of distribution patterns of the studied species in Iran and can be used to guide regional conservation planning, identify critical habitats for threatened species, and inform management and conservation of inland aquatic ecosystems. For this to be effective, a mechanistic framework is needed to untie correlations in potential driving factors. Emerging data sets with fine spatial grain and broad spatial extent will support the transition from correlative models to mechanistic understanding of aquatic invasions.