[A-TT30] Machine Learning Techniques in Weather, Climate, Ocean, Hydrology and Disease Predictions

[ATT30-08] Machine learning regression for building climate and weather-driven scenarios

*Alessandro Damiani¹, Noriko N Ishizaki¹, Hidetaka Sasaki¹, Hitoshi Irie², Dmitry Belikov², Sarah Feron³, Raul R. Cordero⁴ (1.National Institute for Environmental Studies, 2.CEReS, Chiba University, 3.University of Groningen, The Netherlands, 4.Universidad de Santiago de Chile, Chile)

Keywords:machine learning, downscaling, climate, scenario

To fully exploit the big data created within Earth science and climate research, machine learning (ML) techniques are frequently used. Commonly, ML regression is devoted to predicting a continuous-valued attribute associated with an object, and within this framework, its goal is often building a given scenario. For example, creating locally accurate climate information from global-scale data allows building high-resolution future climate scenarios essential for adaptation plans at a regional scale. On the other hand, weather-driven business-as-usual scenarios produced by ML regression allow distinguishing the human-induced impact on a given climate variable or pollutant from that of the changing meteorology. In this presentation, we will discuss our recent achievements obtained with different ML algorithms trained by observation and show the advantages and limitations of our approaches. We will face two case studies, i.e., downscaling of future CMIP6 climate simulation and assessing the pandemic signatures on the environment, both focused on central Japan.