Estimation of extreme weather event using AI and numerical simulation

Abstract

Flood damage tends to increase recent years. The estimation of frequency of occurrence is very important to reduce the flood risk. However, it is quite difficult to estimate the risks because of a lack of observation of extreme weather event. To cope with the issue, we show the estimated precipitation by machine learning using a part of d4PDF data, which is the simulation products called Database for Policy Decision-Making for Future Climate Change (d4PDF). We found that the feature of estimated heavy precipitation (99 percentile) is well corresponded to that of observation (AMeDAS) and the combination of machine learning and the cumulative distribution function method is capable of bias correction and downscaling effectively. It is expected that the flood risks can be evaluated by the estimated heavy precipitation using the combination of machine learning (AI) and numerical simulation.