Open AccessIncreasing Frequency of Anomalous Precipitation Events in Japan Detected by a Deep Learning Autoencoder
Author(s) -
Murakami H.,
Delworth T. L.,
Cooke W. F.,
Kapnick S. B.,
Hsu P.C.
Publication year - 2022
Publication title -
earth's future
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.641
H-Index - 39
ISSN - 2328-4277
DOI - 10.1029/2021ef002481
Subject(s) - precipitation , climatology , environmental science , forcing (mathematics) , autoencoder , atmospheric sciences , meteorology , geology , geography , deep learning , computer science , machine learning
Abstract The frequency of large‐scale anomalous precipitation events associated with heavy precipitation has been increasing in Japan. However, it is unclear if the increase is due to anthropogenic warming or internal variability. Also, it is challenging to develop an objective methodology to identify anomalous events because of the large variety of anomalous precipitation cases. In this study, we applied a deep learning technique to objectively detect anomalous precipitation events in Japan for both observations and simulations using high‐resolution climate models. The results show that the observed increases in anomalous heavy precipitation events in Western Japan during 1977–2015 were not made only by internal variability but the increases in anthropogenic forcing played an important role. Such events will continue to increase in frequency this century. The increases are attributable to the increasing frequency of tropical cyclones and enhanced frontal rainbands near Japan. These results highlight the mitigation challenge posed by the increasing occurrence of unprecedented precipitation events in the future.