Improved storm surge prediction and extreme sea level future projections in the Indian Ocean using deep learning

Abstract Accurate storm surge and extreme sea level (ESL) prediction is crucial for effective coastal flood monitoring and hazard management. However, dynamical models for the Indian Ocean often exhibit significant biases compared to historic tide gauge (TG) based sea level records, highlighting the need for improved methods. To address this gap, we applied a long short-term memory (LSTM) deep learning network to reconstruct hourly storm surges across 29 TG locations along the Indian Ocean coastlines for the period 1979–2021. Our results demonstrate the robust performance of the LSTM model, which outperforms existing global surge reconstructions by realistically capturing storm surges associated with tropical cyclones and extreme conditions. We extended this approach to generate future projections of ESL for the Indian Ocean, which has been lacking for the densely populated coastline. The ESL projections are generated by integrating storm surges predicted via LSTM with bias-corrected Coupled Model Intercomparison Project (CMIP6) predictors, tide data from hydrodynamic models, and mean sea level (MSL) from the Intergovernmental Panel on Climate Change 6th Assessment Report (AR6). The findings reveal that future ESL changes will be predominantly driven by MSL rise, with contributions from tides and storm surges accounting for approximately 10%. Notably, the analysis indicates a rapid emergence of ESL events in the Indian Ocean. By 2030–2040, equatorial islands will experience the one-in-a-hundred-year ESL (ESL 100 ) annually, and the Arabian Sea coastline and the south subtropical regions by 2050 under a high-emission scenario. The findings underscore the urgent need to implement adaptive measures to safeguard vulnerable coastal populations.