[M-GI26] Data-driven approaches for weather and hydrological predictions

[MGI26-04] Advancing Multi-Hazard Worst-Case Scenario Analysis: Maximizing the Potential of Large Ensemble Tropical Cyclone Forecasts

*Md. Rezuanul Islam¹、Le Duc^1,2、Yohei Sawada^1,2 (1.The Univ. of Tokyo、2.Meteorological Research Institute, JMA)

キーワード：Tropical cyclone, Multihazard, Worst-case scenario, Ensemble forecasting, Pareto optimality, Hazard assessment

Tropical cyclone (TC)-induced multi-hazard corresponds to events with multiple concurrent or consecutive drivers (e.g., storm surge, rainfall, wind-induced disasters) leading to substantial impacts on society. In many applications of TC disaster prediction, however, multi scenarios of multi-hazards are ignored. While ensemble forecasts for TCs are commonly used in numerical weather prediction systems, their potential for multi-disaster prediction has not been fully harnessed. This study presents a novel, efficient, and practical framework of multi-hazard scenarios analysis, which leverages large ensemble TC forecasts consisting of 1000 members to analyze storm surge, rainfall, and wind hazard scenarios. To demonstrate the application of our approach, we conducted the simulation of TC Hagibis (2019) using the Japan Meteorological Agency's (JMA) non-hydrostatic model. The resulting atmospheric predictions served as inputs for modeling storm surge, rainfall, and wind hazards. We demonstrate that Pareto-optimized solutions from multi-hazard forecasts can describe potential worst (maximum) and optimum (minimum) multi-hazards while exemplifying a complex diversity of tradeoffs between different types of hazards among different places. For instance, ensemble members that produce worst storm surges do not necessarily produce heavy rainfall in the river basins. Ensemble members that produce heavy rainfall in one river basin do not necessarily generate heavy rainfall to the neighboring river basin. We illustrate the important role of cluster analysis in effectively classifying Pareto optimal solutions when the Pareto-based scenario selection is not straightforward. We employed affinity propagation to classify over 300 multi-hazard worst scenarios. It allows us to objectively identify and select the cluster that represents the ideal worst scenario among many. Furthermore, our study underscores the importance of conducting a detail evaluation of Pareto optimal solutions to elucidate the influence of meteorological variables, such as TC track, intensity, and size, on both worst and best-case multi scenarios of multi hazard. This proposed assessment is expected to empower disaster communities and authorities to make informed decisions and strengthen their resilience in the face of TC impacts.