Regional habitat suitability modelling

Reliable maps of marine species distributions across the Atlantic, both benthic and pelagic, are required in order to implement international area- and ecosystem based- management. To address this need, iAtlantic undertook habitat suitability modelling (HSM) of vulnerable marine ecosystem (VME) indicator taxa and commercially important species at multiple spatial scales, to feed into conservation and management strategies at different levels. The broad, Atlantic-wide HSM work is presented in Deliverable 2.2. However, most marine spatial planning and management actions take place at national and regional scale. Hence, more specific and detailed HSMs were developed at regional scales, for VME indicator taxa and commercial species that reflected regional requirements in terms of conservation and/or management needs. The chosen taxa mainly included cold-water corals and commercial fish, sharks and crustaceans, and varied between regions depending on regional priorities and data availability. The work was carried out across eight of the iAtlantic regions as part of iAtlantic work package 2 ‘Mapping deep Atlantic Ecosystems’.

Habitat Suitability Models combine known species presence (and absence, where confirmed) records with information on the local environment, to establish, either statistically or through Machine Learning techniques, the characteristics of the species’ environmental niche, expressed in the species-environment relationships. Those relationships in turn are used to predict where the species or taxa of interest may occur in the wider region, in locations that have not been sampled or surveyed. For the regional HSMs, species presence records were obtained from open-source data depositories, museum records, image annotation and fisheries and scientific trawl data. Environmental data included bathymetry maps and their derivatives (e.g. slope, aspect, terrain ruggedness index etc., describing the terrain characteristics), water column characteristics (e.g. temperature, salinity, …), parameters describing the hydrodynamic regime (current speeds & direction, kinetic energy dissipation) and proxies for food supply (e.g. primary productivity). Modelling approaches included Random Forest (RF), Maximum Entropy (MaxEnt), Generalized Boosted Models (GBM/BRT) and Generalised Additive Models (GAMs) algorithms. Where multiple algorithms were used, ensemble models of outputs were produced to improve robustness of the HSM predictions.

The resulting regional HSMs indicate that Shelf and Slope environments of increased terrain complexity and increased regional or local productivity are most likely to support the VME and commercially important taxa modelled. The HSMs point to water mass properties (particularly temperature, but also chemistry), proxies for food supply, and terrain as the most important environmental predictors for species occurrence. However, the relationship with terrain is more variable, being also dependent on species substrate requirements, and is weaker for pelagic species. Model parameters such as resolution of environmental data, sampling intensity across the modelled extent, resultant sample size and class imbalance of species/(pseudo) absence records are all factors that influenced model performance. This is common among HSMs and can be particularly challenging in areas with limited data. Hence, as part of the regional HSM work in iAtlantic, new methods were developed in some of the regions to account for small sample sizes and imbalance in the sample input datasets (imbalance between presence/absence records, or imbalance in spatial distribution of the presence data). The outcomes appear promising for overcoming these challenges. Low spatial resolution of available physical oceanography and chemistry data is another important limitation determining predictor importance, as it determines the degree to which ecologically relevant environmental heterogeneity can be resolved.  

Consequently, collecting and integrating higher resolution datasets, including additional species records is strongly recommended to improve future HSMs across the Atlantic. Still, the integrated approach of iAtlantic has proven powerful in identifying key environmental predictors of species habitat distribution at regional scales. New parameters describing the environmental conditions and variability were developed during the project, and tested in several of the Study Regions (e.g. kinetic energy dissipation in the bottom waters, consecutive disparity index (D) of seasonal variables including temperature, primary production reaching the seabed, or a new water chemistry index). Many of these were found to be significant environmental descriptors, and therefore open opportunities for HSM in future studies/locations.

Comparative studies of common species also enable stronger hypotheses of underlying mechanisms driving that species’ distribution to be made. Collectively, the provision of HSMs and continuous distribution maps of key VME taxa and commercial species across the eight iAtlantic regions has increased our understanding of regional environmental controls on species habitat distributions and provides important information on their spatial extent for future marine spatial planning and area- and ecosystem-based management.