Spatial prediction of demersal fish diversity in the Baltic Sea: comparison of machine learning and regression-based techniques

摘要: Marine spatial planning (MSP) is considered a valuable tool in the ecosystem-based management of marine areas. Predictive modelling may be applied MSP framework to obtain spatially explicit information about biodiversity patterns. The growing number statistical approaches used for this purpose implies urgent need comparisons between different predictive techniques. In study, we evaluated performance selected machine learning and regression-based methods that were fish community indices. We hypothesized habitat features can influence assemblage investigated effect environmental gradients on demersal diversity (species richness Shannon–Weaver Index). data from Baltic International Trawl Surveys (2001–2014) maps six potential predictors: bottom salinity, depth, seabed slope, growth season temperature, sediments annual mean current velocity. compared alternative approaches: generalized linear models, additive multivariate adaptive regression splines, support vector machines, boosted trees random forests. repeated 10-fold cross-validation, using accuracy as measure model quality. Finally, forest best performing algorithm implemented it prediction Proper Kattegat. To reliability confidence developed which are essential MSP, estimated uncertainty predictions with standard deviation obtained all ensemble method. showed how state-of-the-art techniques, based easily available simple Geographic Information System tools, reliable diversity. Our comparative work highlighted method reduce error MSP.