Global habitat suitability for framework-forming cold-water corals

Abstract

Predictive habitat models are increasingly being used by conservationists, researchers and governmental bodies to identify vulnerable ecosystems and species' distributions in areas that have not been sampled. However, in the deep sea, several limitations have restricted the widespread utilisation of this approach. These range from issues with the accuracy of species presences, the lack of reliable absence data and the limited spatial resolution of environmental factors known or thought to control deep-sea species' distributions. To address these problems, global habitat suitability models have been generated for five species of framework-forming scleractinian corals by taking the best available data and using a novel approach to generate high resolution maps of seafloor conditions. High-resolution global bathymetry was used to resample gridded data from sources such as World Ocean Atlas to produce continuous 30-arc second (∼1 km(2)) global grids for environmental, chemical and physical data of the world's oceans. The increased area and resolution of the environmental variables resulted in a greater number of coral presence records being incorporated into habitat models and higher accuracy of model predictions. The most important factors in determining cold-water coral habitat suitability were depth, temperature, aragonite saturation state and salinity. Model outputs indicated the majority of suitable coral habitat is likely to occur on the continental shelves and slopes of the Atlantic, South Pacific and Indian Oceans. The North Pacific has very little suitable scleractinian coral habitat. Numerous small scale features (i.e., seamounts), which have not been sampled or identified as having a high probability of supporting cold-water coral habitat were identified in all ocean basins. Field validation of newly identified areas is needed to determine the accuracy of model results, assess the utility of modelling efforts to identify vulnerable marine ecosystems for inclusion in future marine protected areas and reduce coral bycatch by commercial fisheries.

Figure 1. Global distribution of species presences used in this study.

a) All five framework forming species, b) Enallopsammia rostrata, c) Goniocorella dumosa, d) Lophelia pertusa, e) Madrepora oculata, f) Solenosmillia variabilis.

Figure 2. Geographical distribution of error between the up-scaled environmental layers and water bottle stations.

a) Temperature, b) salinity, c) nitrate, d) phosphate, e) silicate. The scales show the average difference of all water bottle stations that fall within a single five degree grid cell.

Figure 3. Validation of the environmental layer creation process for temperature.

a) Correlation (0.924) of intersected GLODAP stations with the layer, b) mean temperature relationships at depth in 50 m bins, c) mean temperature at latitude in 5° bins, d) mean temperature at longitude in 10° bins. The black lines are temperature at each GLODAP bottle station; the red lines are the value of the environmental layer at the position of each GLODAP station.

Figure 4. Bean plots of species presences intersected with the environmental variables used in the models (the small lines in the centre of each bean shows individual presence data points.

The bean itself is a density trace that is mirrored to show as a full bean [42]).

Figure 5. Habitat suitability maps for framework forming species.

a) All five framework forming species, b) E. rostrata, c) G. dumosa, d) L. pertusa, e) M. oculata, f) S. variabilis. High resolution maps are available as supplementary Figures S7, S8, S9, S10, S11, S12.

Figure 6. Local and regional area outputs for areas suitable habitat.

a) E. rostrata around New Zealand and Australia, b) G. dumosa around New Zealand and Australia, c) S. variabilis on seamounts in the Southern Ocean south of Madagascar.

Figure 7. Binary predicted presence map for scleractinian framework-forming corals based on the 10^th percentile training presence, which omits the 10% most extreme presence observations as they may represent recording errors.

White background indicates that these species are not likely to be found, red indicates probable presence.

Figure 8. Comparison between earlier predictions of suitable habitat for L. pertusa by Davies et al. and those developed in this study.

a) global L. pertusa habitat predicted by Davies et al. using ENFA, b) global L. pertusa habitat predicted by this study. Note the significant increase in spatial resolution and ability to identify suitable habitat on seamounts off Portugal.