Abstract
Coastal communities face significant risk from storm-induced coastal flooding, which causes substantial societal and economic losses worldwide. Machine learning techniques have increasingly been integrated into coastal hazard modeling, particularly for storm surge prediction, due to advances in computational capacity. However, incorporating multiple projected future climate and landscape scenarios requires extensive numerical simulations of synthetic storm suites over large geospatial domains, resulting in rapidly escalating computational costs. This study proposes a cost-effective training data reduction strategy for machine learning based storm surge surrogate models that enables efficient incorporation of new future scenarios while minimizing computational burden. The proposed strategy reduces training data across three dimensions: grid points, input features, and storm suite size. Reducing the storm suite size for future scenario simulations is highly effective in guiding numerical simulations, yielding substantial reductions in simulation cost. The performance of surrogate models trained on reduced datasets was evaluated using different machine learning algorithms. Results demonstrate that the proposed reduction strategy is robust across different model types. When trained using 5,000 out of 80,000 grid points, 10 out of 12 input features, and 60 out of 90 storms, the total training dataset is reduced to approximately 5% of its original size. Despite this reduction, the trained model achieves a correlation coefficient of 0.94, comparable to models trained on the full dataset. In addition, storm selection methodologies are introduced to support efficient storm set expansion for future scenario analyses.
Introduction
To protect coastal communities from flooding, engineers rely on numerical models (like ADCIRC) to simulate storm surges. Because these models are computationally expensive, the field is shifting toward Machine Learning (ML) "surrogate" models, which can predict surges almost instantly once trained. However, training these surrogates for future scenarios (e.g., sea-level rise, land subsidence) requires running slow numerical models thousands of times to generate training data. This study introduces a three-dimensional data reduction strategy to overcome this computational bottleneck.
Methodology
The authors propose a sequential, three-pronged data reduction strategy to minimize the amount of training data required without sacrificing predictive accuracy.
- Grid Point Reduction: Instead of training on all 80,000+ spatial grid points (GPs), $k$-means clustering identifies a representative subset of 5,000 GPs based on geospatial and landscape features.
- Input Feature Reduction: By removing redundant landscape variables (e.g., dropping canopy coefficient and keeping Manning’s $n$), the model is streamlined from 12 to 10 input features.
- Storm Set Reduction (Active Learning): Using $k$-medoids clustering and adaptive sampling, the model iteratively identifies and adds storms that produce the highest prediction errors. This ensures that expensive numerical simulations are only run for the most informative scenarios.
Key Results and Insights
The proposed strategy successfully reduced the training data to approximately 5% of its original size while maintaining a correlation coefficient ($R$) of 0.94, comparable to models trained on the full dataset.
Key insights from the study include:
- Model-Guided Data Selection: Adaptive sampling allows the model to "request" training data for storms it struggles to predict, avoiding redundant simulations.
- Spatial Consistency: Inconsistent grid point allocation across different scenarios was found to cause data imbalances that degraded geographic performance.
- Algorithm Robustness: The reduction strategy proved effective across different ML architectures, including Neural Networks and XGBoost.
Conclusion
This research provides a mathematical framework for coastal engineers to decide which synthetic storms to run in future planning cycles. While challenges like "catastrophic forgetting" in incremental learning remain, the reduction framework is recommended for large-scale engineering efforts like the Louisiana Coastal Master Plan 2029 to maximize surrogate model performance while minimizing computational costs.
Summary updated