Forecasting Hawaiian Coral Reef Health: Machine Learning Models in Predicting Coral Cover from Ocean Stressors

Abstract

Coral reefs, vital marine ecosystems for biodiversity and coastal protection, face increasing threats from both climate stressors and direct human impacts. Elevated sea surface temperatures, ocean acidification, and nutrient imbalances have been linked to widespread coral bleaching and declining reef resilience, alongside subsequent damage to coastal marine life which inhabit coral reefs. (Hagen 8-9) Accurate forecasting of coral reef health given these stressors is critical in informing conservation efforts. This study aims to evaluate the effectiveness of three machine learning models: linear regression, decision tree regression, and MLPRegressor (neural network) in predicting percentage coral cover based on sea surface temperature (SST), ocean pH, and nutrient levels (total dissolved phosphorous and nitrogen). Using datasets spanning 35 years (1988-2023) from the Biological & Chemical Oceanography Data Management Office (BCO-DMO) and National Oceanic and Atmospheric Administration (NOAA) hosted data from coral reef monitoring programs, these models were trained and tested on historical coral cover. Results were weak, with every model failing to effectively predict coral coverage due to limitations in data alignment and resolution. This outcome demonstrates the dependence of machine learning performance on data integrity and synchronization, emphasizing that even advanced algorithms cannot compensate for fragmented datasets. This study thus provides methodological insight for future ecological modeling, highlighting the need for temporally and spatially consistent environmental data to enable reliable coral health forecasting.