Species-environment relationships and potential for distribution modelling in coastal waters

- Review of the species-environment relationship in coastal waters
- Hydrography constitutes an important and heterogeneous group of predictors.
- Depth and exposure have high potential for species distribution modelling.
- Acknowledging interaction effects and biotic features enhances modelling efforts.

Due to increasing pressure on the marine environment there is a growing need to understand species-environment relationships. To provide background for prioritising among variables (predictors) for use in distribution models, the relevance of predictors for benthic species was reviewed using the coastal Baltic Sea as a case-study area. Significant relationships for three response groups (fish, macroinvertebrates, macrovegetation) and six predictor categories (bottom topography, biotic features, hydrography, wave exposure, substrate and spatiotemporal variability) were extracted from 145 queried peer-reviewed field-studies covering three decades and six subregions. In addition, the occurrence of interaction among predictors was analysed. Hydrography was most often found in significant relationships, had low level of interaction with other predictors, but also had the most non-significant relationships. Depth and wave exposure were important in all subregions and are readily available, increasing their applicability for cross-regional modelling efforts. Otherwise, effort to model species distributions may prove challenging at larger scale as the relevance of predictors differed among both response groups and regions. Fish and hard bottom macrovegetation have the largest modelling potential, as they are structured by a set of predictors that at the same time are accurately mapped. A general importance of biotic features implies that these need to be accounted for in distribution modelling, but the mapping of most biotic features is challenging, which currently lowers the applicability. The presence of interactions suggests that predictive methods allowing for interactive effects are preferable. Detailing these complexities is important for future distribution modelling.