QUANTIFYING EFFECTS OF THERMAL CHANGES ON PREDICTED SUITABLE HABITAT OF THE DEEP-SEA CORAL LOPHELIA PERTUSA
Multiple machine learning models were trained to predict suitable habitat of Lophelia pertusa using predicted mean annual bottom temperature and 11 other environmental variables from NOAA data. A 50-tree random forest model provided the best accuracy with an out-of-bag score of 98%. This random forest model was used to calculate new predicted suitable habitat values of Lophelia pertusa by increasing the predicted mean annual bottom temperature based on sea surface temperature increases scaled to depth.
Results indicate that areas with a predicted suitable habitat greater than 0.5 have their modeled predicted suitable habitat decrease by 7% given a sea surface temperature increase of 1.0oC, and decrease by 23% given a sea surface temperate increase of 2.5oC. Areas with greater than 0.95 predicted suitable habitat have their modeled predicted suitable habitat decrease by 19% given a sea surface temperature increase of 1.0oC and decrease by 68% given a sea surface temperate increase of 2.5oC. Overall, this research provides the first quantitative support that predicted suitable habitat of Lophelia pertusa will decrease given an increase in predicted mean annual bottom temperature.
Heretofore, answering the question of how predicted mean annual bottom temperature increases correspond to decreases in predicted suitable habitat proved difficult due to sampling challenges and missing data. Machine learning models help address those challenges by incorporating all available data to reveal important patterns within deep-sea ecosystems that may not have been detectable otherwise.