Species distribution modelling of benthic invertebrates in the south-eastern Baltic Sea

The distribution of benthic invertebrates is one of the key parameters for the marine spatial planning and management, however traditionally the data on benthic invertebrates are based on point sampling. Recently statistical methods of predictive modelling are used to create maps of species distribution, nevertheless, no comparative analysis of different modelling methods has been yet performed in the Baltic Sea region. In this study the occurrence and biomass distribution of 23 benthic species in the southeastern Baltic Sea were modelled. A comparison of the following predictive modelling methods was performed: random forests (RF), generalized additive models (GAM), multivariate adaptive regression splines (MARS) and maximum entropy (MaxEnt). In order to assess the consistency of the methods, 100 iterations with different train/test datasets were made for each of them. Random forests achieved the highest predictive performance for both species occurrence and biomass distribution models; also it was the most consistent for different iterations. Predictive performance of GAMs and MARS followed RF, whereas MaxEnt accurately predicted occurrence only for the species with a relatively low distribution range.