Considering temporal neighborhoods in probability forecast using large ensemble simulation: case study of heavy rainfall event over Kyushu, Japan in July 2022

Abstract

This study investigates the probabilistic forecasting of a heavy rainfall event over Kyushu, Japan during July 2022 based on a series of data assimilation and forecast experiments using 1000-member ensembles. The deterministic forecast successfully reproduced the overall intensity and pattern of the heavy rainfall, but faced challenges in predicting the exact timing and location of the rainfall because of the strong nonlinearity within the atmospheric system. Based on the ensemble forecasts, the probability forecast indicated a low likelihood of heavy rainfall. Substantial uncertainties were observed, particularly with respect to the timing and spatial distribution of the rainfall, reflecting the inherent difficulties of forecasting such extreme and localized events. However, by allowing for some discrepancies in the timing and location of the rainfall, the probability of heavy rain was increased. This approach highlights the value of incorporating spatial and temporal tolerances when interpreting ensemble outputs. Such adjustments can enhance the reliability of probabilistic forecasts, providing more actionable information for those involved in disaster preparedness and risk mitigation.