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

[ATT30-P03] Diagnostics of intense precipitation from large-scale atmospheric fields for the interpretation of climate modeling in Moscow region

*Yulia Yarinich^1,2,3, Victor Stepanenko^1,4,2, Mikhail Krinitskiy^1,4,5,6, Mikhail Varentsov^1,4 (1.Research Computing Center, Lomonosov Moscow State University, Moscow, Russia, 2. Faculty of Geography, Lomonosov Moscow State University, Moscow, Russia , 3.A.M. Obukhov Institute of Atmospheric Physics of Russian Academy of Science, 4.Moscow Center for Fundamental and Applied Mathematics, Moscow, Russia, 5.Shirshov Institute of Oceanology, Russian Academy of Sciences, Moscow, Russia, 6.Moscow Institute of Physics and Technology, Dolgoprudny, Russia)

Keywords:extreme precipitation, statistical downscaling of precipitation, ERA5 reanalysis, atmospheric instability indices

Due to observing climate change an increasing frequency of extreme precipitation have impacts in various regions including Northern Eurasia and are especially destructive in big cities. Global climate change impacts are generally assessed by downscaling large-scale climatic characteristics which are better resolved in climate models into small-scale variables which cannot be explicitly resolved on a climate model grid. The previous studies have investigated machine learning approaches to downscale precipitation in several regions, but the territory of the Moscow agglomeration, the largest in Russia and Europe, remained untouched while the annual risk from floods due to extreme precipitation in this region remains very high.
In this work, statistical downscaling methods (machine learning) are used to obtain probabilistic characteristics of intense precipitation from low-resolution atmospheric hydrodynamic modeling fields. The maximum daily precipitation in the Moscow region according to long-term observations at weather stations (1988 – 2020) is presented as a predictand. The characteristic description variables are physically based large-scale predictors of intense precipitation, calculated using ERA5 reanalysis data and averaged over the Moscow region domain territory. The Ridge Regressor model was taken as a baseline. Compared to the base value of the average amount for the territory, it shows a significant improvement in the reproduction of precipitation characteristics. Rating for the feature importance of large-scale atmospheric predictors for the territory of the Moscow region was also shown.
This work was supported by Non-commercial Foundation for the Advancement of Science and Education «INTELLECT».