This study investigates how projected changes in tropical cyclone (TC)-induced storminess, defined by extreme wind, surge, and wave events, will affect coastal flood hazards along the Sofala coastline in Mozambique under the near-future SSP5-85 climate scenario (up to year 2050). While global studies have suggested intensified storm characteristics in the Sofala region, no study to date has assessed their implications for regional coastal flood hazard in conjunction with sea-level rise (SLR). Using the synthetic STORM dataset in combination with probabilistic extreme value analysis and a storyline-based TC Idai simulation, this study quantifies the projected increase in TC-induced hazards and evaluates the relative contribution of storminess change and SLR to future flood depth and extent. Results show a consistent upward shift in the 100-year return levels for wind, surge, and wave. These changes can be mainly statistically attributed to the exceedance frequency (i.e., how often extremes occur) rather than exceedance intensity (i.e., how extreme they are when they occur). As such, the effective return period of historical 100-year events is found to be 45–60 years on average under the near-future scenario, suggesting that extreme events will become significantly more frequent. Although the upper tail dependence between surge and wave slightly weakens, the likelihood of joint surge–wave extremes still increases, lowering the joint return period from an average of 200 years to 120 years. In the TC Idai storyline, SLR accounts for 60–90% of the increase in total water levels, with a magnitude of around 0.25 m. However, storminess change contributes up to 40% in some areas, and is shown to substantially heighten flood extent and population exposure. The findings underscore the importance of integrating TC-induced storminess change in Sofala’s future coastal hazard assessments, in addition to SLR. Recommendations include extending analyses to additional synthetic TC models and emissions scenarios (e.g. SSP2-45), improving hydrodynamic model validation, and incorporating rainfall and fluvial forcing to capture the full spectrum of compound flood hazards under a changing climate.

Accurate flood prediction is essential for effective risk management, but simulating with physics-based models is computationally intensive and time-consuming, limiting their use in operational use cases. To address this challenge, this study develops a deep learning surrogate model that replicates spatial flood depth predictions of the physics-based model SFINCS, with significantly reduced computational cost.

The surrogate model is trained with a synthetic dataset of 13,000 tracks generated by TCWiSE, that have been computed with SFINCS. Input for the surrogate model are key drivers of compound coastal flooding, including cumulative precipitation, maximum wind speed, and water depth at the river outlet, which are given to the model as scalar values. The results demonstrate that the model can reproduce flood depth patterns with a RMSE of 0.054 m on the original dataset and 0.069 m on a case study application. The CSI values of 0.829 and 0.776, respectively, are comparable to values reported in recent literature. Visual inspection of spatial predictions shows that the model performs well overall but shows localized error patterns, particularly in the western part of the domain, indicating that additional input features may be needed to better capture interactions in physical processes that occur in compound coastal flooding. From a computational perspective, the deep learning model runs approximately 100 times faster than the physics-based model SFINCS on CPU, with further gains on GPU. This computational efficiency allows for running more simulations within a given timeframe, enabling the exploration of an increased number of potential flood scenarios and providing decision-makers with more time to evaluate and respond.

Overall, the study demonstrates that deep learning surrogate models offer a promising alternative to physics-based models by providing accurate and faster predictions. Although the current set-up shows accurate results, there remains room for improvement through enhanced input representation, such as incorporating spatiotemporal information of the drivers or including additional hydrodynamic features. Further gains in performance and generalizability can also be achieved by refining the model architecture and training strategies.