LASSE: Learning Active Sampling for Storm Tide Extremes in Non-Stationary Climate Regimes

TL;DR

The paper tackles the high computational cost of identifying tropical cyclones that generate extreme storm tides by leveraging surrogate models trained with informative online learning. It demonstrates that a batch XGBoost surrogate generalizes across present and future climate scenarios, achieving high precision and recall with only a fraction of hydrodynamic simulations. An informative, iterative approach (including Ens-CGP) can reach either perfect precision or total recall after evaluating as little as 20% of the TC catalog, greatly accelerating risk assessment. The methods are scalable to large catalogs, generalizable to different climate scenarios, and adaptable to prioritize precision or recall, offering a practical framework for rapid cyclone hazard analysis and resilience planning.

Abstract

Identifying tropical cyclones that generate destructive storm tides for risk assessment, such as from large downscaled storm catalogs for climate studies, is often intractable because it entails many expensive Monte Carlo hydrodynamic simulations. Here, we show that surrogate models are promising from accuracy, recall, and precision perspectives, and they "generalize" to novel climate scenarios. We then present an informative online learning approach to rapidly search for extreme storm tide-producing cyclones using only a few hydrodynamic simulations. Starting from a minimal subset of TCs with detailed storm tide hydrodynamic simulations, a surrogate model selects informative data to retrain online and iteratively improves its predictions of damaging TCs. Results on an extensive catalog of downscaled TCs indicate 100% precision in retrieving rare destructive storms using less than 20% of the simulations as training. The informative sampling approach is efficient, scalable to large storm catalogs, and generalizable to climate scenarios.

Figures (8)

• Figure 1: The standard workflow to quantify TC-induced storm tide hazards includes downscaling a climate model followed by expensive hydrodynamic simulations that make hazard assessments of extremes computationally intractable.
• Figure 2: ML-based surrogate allows for faster risk assessment and search for damaging TCs. Few detailed simulations train the surrogate to locate the rare destructive storms, that is, those causing extreme storm tides, from a large number of downscaled TCs.
• Figure 3: Average confusion matrix from repeated trials of training model on a random split of 75% of ERA5 hersbach2020era5 data and testing on the remaining 25%.
• Figure 4: Confusion matrix of results from training model on all ERA5 hersbach2020era5 data and testing on all SSP5-8.5 data gmd-15-2973-2022o2016scenario.
• Figure 5: Averaged confusion matrix from repeated trials of training model on a random split of 20% of ERA5 hersbach2020era5 data and testing on all SSP5-8.5 data gmd-15-2973-2022o2016scenario.