UC Santa Cruz

Marine heatwaves (MHWs) are periods of abnormally high sea surface temperatures (SSTs) that persist for periods of time, causing adverse impacts of marine ecosystems and coastal communities. With projections that MHWs will become more frequent and severe, it is increasingly important to be able to predict MHW events to mitigate risks for ecosystems and communities. Here, we investigate the predictability of MHWs in the northeast Pacific Ocean by employing a suite of models, including logistic regression, naive Bayes, gradient boosting, random forest, and feedforward neural network, all of which are trained on selected oceanic and atmospheric variables. We find that the random forest model performs best at predicting the presence or absence of a MHW event at the 90th percentile MHW threshold using cluster centroid balanced data. The model is able to predict with accuracy ranging from 0.98 to 0.97 for leads spanning from 1 day to 2 weeks. While all models encounter difficulties in accurately categorizing MHWs, predicting their presence or absence remains a valuable metric for informing managers and industries about impending MHW events. Short-term forecasts can be especially advantageous in alerting industries and communities to these events, empowering them to implement adaptive measures against the detrimental impacts of MHWs.