| 11 Dec 2025
Investigating the predictability of Marine Heatwaves at subseasonal to seasonal timescales in New Caledonia, South Pacific
Abstract. Marine Heatwaves (MHWs) have emerged as one of the most important threat for marine ecosystems, with impacts such as coral bleaching, massive fish mortality and displacement of mobile fauna. In the context of climate change, it is urgent to develop strategies such as subseasonal to seasonal forecasting to help human societies adapt and react to the increasing frequency, duration and intensity of these events. Here we evaluate the predictability of MHWs at the scale of a South Pacific island country, New Caledonia, using ensemble forecasts from a dynamical coupled ocean-atmosphere model. We show that implementing a probabilistic approach where we extract information from the dispersion in the ensemble results in a higher skill than a deterministic approach where we simply compute the ensemble average. We find that longer, more intense, and wider MHWs, are more predictable than weaker, less intense, and shorter MHWs. We also find that the longest and widest MHWs occur in the cold season (June–October) during strong La Niña episodes, and that they can successfully be predicted up to 7 months in advance. In contrast, MHWs occurring during the warm season have poor or no predictability of more than a few weeks in advance. We discuss how this information can be efficiently transferred to marine stakeholders in terms of the usefulness and useability of the forecast. We recommend that future research should focus on identifying the drivers of different types of MHWs in order to understand their sources of predictability.
This study assessed the predictability of marine heatwaves surrounding New Caledonia, using dynamical coupled ocean-atmosphere model results. The assessment suggests that a probabilistic approach tends to have a higher skill than a deterministic approach, and stronger marine heatwaves are more predictable, as those that occurred during the La Niña events. The results suggest that the dynamical model predictions may provide useable forecast for regional stakeholders. The manuscript is generally well organised, and the following are some suggestions for the authors to improve their presentations.
My main comment is about the model prediction skills. The authors admitted that "probabilistic and deterministic skill are impossible to compare directly since they are quantified with different scores". So the conclusion that "the probabilistic approach improves the quality of the forecast" is not very intuitive. Maybe 1-2 case studies on some of the recent marine heatwave event predictions can better demonstrate this point. Such as, the statistical prediction provides a more advanced warning for stakeholders.
Is this conclusion sensitive to the selection of the 20% criteria?
The authors may want to provide more details about the AUC calculation in the supporting information.
Figure 3 a and b: it is better to use a Hofmuller diagram to show the year to year variations.