Increasing Frequency of Extreme Precipitation in Japan due to Global Warming Detected by a Machine Learning Technique

Abstract:

Frequency of extreme events of precipitation in Japan has been increasing during the summer season. A typical event of an extreme is associated with steady frontal condition that is often strengthened by moist advection through tropical cyclones. Previous studies have reported that frequency of heavy-precipitation events would likely increase almost everywhere in the world toward the end of this century due to increases in moisture in the warming. However, it is unclear if the ongoing increases in extreme events in Japan over the last decades are due to anthropogenic global warming because of luck in large-ensemble high-resolution climate model experiments to separate effect of anthropogenic forcing from intrinsic natural climate variability. Also, it is challenging how to detect extreme events objectively due to large variety of extreme precipitation events.

Here, we developed a new methodology using deep learning, autoencoder, and applied it to both observed and model simulated precipitation to objectively detect extreme precipitation event in Japan. We show that the recent increases in extreme precipitation events in Japan over 1971-2015 is largely due to anthropogenic global warming. The increase is more evident in west of Japan, centered around Kyushu Island, indicating that the extreme precipitation in July 2020 was also affected by global warming. A series of climate model simulations indicate that the extreme events will keep increasing toward the end of this century. It is shown that one of the main reasons for the increases is increasing frequency of occurrence in tropical cyclones, especially intense tropical cyclones around Japan.