Machine learning-enhanced static flood models for high-resolution peak storm surge inundation mapping in Southeast Texas, USA

Coastal floodplain surrounding Galveston, southeastern Texas Rapid, high-resolution peak storm surge predictions are essential for emergency response and infrastructure protection in coastal communities. However, existing models face inherent trade-offs: physics-based models are accurate but computationally expensive; machine learning (ML) models are efficient but often lack spatial resolution; and static flood models are fast but limited by sparse input data. To address these limitations, we propose a hybrid framework—C1PK-Flood—that employs an ML approach to enhance the accuracy and applicability of a static flood model, particularly in data-scarce regions. We demonstrate the hybrid framework's effectiveness in Galveston, Texas, over 19 million grid cells, where it successfully generated high-resolution predictions within minutes and improved the accuracy of an existing static flood model by incorporating multiple synthetic observation points. We found that an optimized input configuration reduced the normalized RMSE by up to 10 %, and that time series inputs generally outperformed landfall-only inputs, particularly for extreme surge. Moreover, the model maintained comparable accuracy with less than half of the original training data, demonstrating the benefits of efficient training data design. Its low memory requirements and fast prediction speed make it appropriate for rapid emergency response and real-time flood forecasting, enabling digital twins and strengthening coastal resilience. • A hybrid storm surge model, combining ML and a static flood model, is presented. • It generates high-resolution peak surge maps from hurricane data within minutes. • C1PK-Flood broadens the applicability of static flood models by integrating ML. • Optimized TC time series inputs improved predictions over prior common approaches. • We showed reliable prediction is possible with fewer training inputs.