Towards monitoring subsurface marine heatwaves based on sea surface properties in the Eastern Pacific

As marine heatwaves (MHWs) become a growing concern for marine ecosystems, an effective ecosystem management necessitates precise monitoring of such periods with exceptionally high water temperatures. As satellite-based temperature measurements do not reach beyond the sea surface, identifying subsurface MHWs has so far relied on lower-resolution data obtained from (autonomous) in-situ measurements. In this study, we assess to which extent subsurface MHWs, defined statically by a seasonally varying 90th percentile, can be deduced from surface properties that can be remotely-sensed at a high spatio-temporal resolution. To this end, we build a Random Forest (RF) classification model with daily data from a high-resolution numerical hindcast simulation focused on the Eastern Pacific (1979-2019). The RF is trained to distinguish between extreme and non-extreme temperatures at the depth of the climatologically maximum mixed layer depth (MLD), i.e. a depth that is decoupled from the sea surface throughout most parts of the year. We train the RF on the first 80% of the hindcast simulation data (i.e., 1979-2011) and use a range of predictor variables, such as anomalies of sea surface temperature (SST), height (SSH) and salinity (SSS) as well as derivatives of these physical variables. Testing the model on the last 20% of the hindcast simulation (2012-2019), the RF correctly identifies more than two thirds of all subsurface extreme states, leaving only about 30% of subsurface extremes unidentified. Yet, of all RF-based subsurface extreme classifications, about 40% of subsurface temperatures are false positives. Nevertheless, the RF model outperforms a simple SST based extrapolation of extreme states into the ocean interior. The RF-based classification is mostly guided by SSH and SST anomalies (together reduce impurity by about 50%), followed by climate indices like the Oceanic Nio Index (ONI) and the Pacific Decadal Oscillation (combined impurity reduction by 20%). This simulation-based study emphasizes the potential of exploring remote sensing data, particularly SST and SSH, to extend the monitoring of MHWs beneath the sea surface. Integrating this high-resolution statistical estimate with lower-resolution in-situ hydrographic information has the potential to make subsurface MHW monitoring a feasible and valuable tool for marine ecosystem management.