Unveiling the Drivers of Tropical Indian Ocean Warming through Machine Learning–Assisted Surface Wind

Abstract The tropical Indian Ocean (TIO) has experienced pronounced warming trends in recent decades, with dynamical processes recognized as key drivers. However, the role of thermal processes remains uncertain due to discrepancies in surface-wind-induced heat flux across existing datasets. The present study introduces a random forest machine learning algorithm that synergistically integrates the advantages of in situ observations and satellite data, yielding a monthly surface wind [machine learning–assisted wind (MLAWind)] dataset and corresponding air–sea heat flux from 1950 to 2022 with a horizontal resolution of 1° × 1°. The MLAWind exhibits high accuracy and robust generalization capability based on evaluations using both satellite and buoy observations. Besides, it is capable of effectively representing spatial and temporal characteristics of surface wind. In contrast to the majority of existing reanalysis datasets, MLAWind reveals a decline in surface wind over the TIO since 1950, which is further supported by the west–east asymmetrical variations in sea surface height and thermocline depth. The attenuation of surface wind is more significant in the eastern TIO as compared to the western TIO, leading to a remarkable reduction in evaporative cooling within the eastern TIO. The thermal processes associated with surface-wind-induced heat flux serve as the essential drivers of the warming in the eastern TIO, with a contribution accounting for approximately 45% of that of dynamical processes. The findings of our study challenge existing reanalysis results but are aligned with state-of-the-art models, highlighting that the significance of thermal processes is substantially underestimated in most existing reanalysis datasets. Significance Statement This study aims to identify the drivers of the significant warming of the tropical Indian Ocean in recent decades. This is important because the underlying mechanisms remain controversial due to uncertainties in existing datasets. We employ a machine learning algorithm to generate a reliable surface wind dataset that combines the advantages of in situ observations and satellite data. Our findings obtained from this dataset reveal that the role of turbulent heat exchange associated with surface winds has been overlooked in most reanalysis datasets.