Work

The crucial role of seasonality in improving the skill of machine learning in predicting precipitation types in China

Year: 2025

Type: article

Abstract: The occurrence and variation of precipitation types (PTs) are closely linked to hazardous extreme weather events in China, including freezing precipitation, mixed rain and snow, and hail. This study constructs multiple machine learning (ML) models to diagnose the climatological characteristics of PTs across China. This employs predictor variables from reanalysis data combined with observational PT records (as the targeted variable) from weather stations. Advanced ML-based feature selection techniques are implemented to identify critical determinants governing PT occurrences. Four distinct Random Forest Classifier (RFC) architectures are constructed, including the RFC trained by the samples of all data (RFC_China) and those trained respectively by samples over each region (RFC_Regional), each season (RFC_Seasonal) and each season over each region (RFC_Regional & Seasonal). Results reveal that key variables associated with the occurrence and variation of PTs include near-surface air temperature, dewpoint temperature, warm layer depth, and lower-to-mid tropospheric geopotential thickness. The constructed RFC models demonstrate some realism in representing PT climatology during 1990–2014, consistently performing better than the PT classifications in the ERA5 climate reanalysis dataset in terms of spatial correlations. Overall, RFC_Seasonal is the best-performing algorithm showing the best PT classification skill in winter. Lower performances are demonstrated by RFC_Regional. The results imply the importance of seasonality for interpreting the association between PTs and meteorological conditions while their regionality plays a secondary role. These findings also suggest potential applications for integrating the developed RFC models with state-of-the-art climate models to project possible changes in PT climatology over China in a warmer climate.

Source: Journal of Hydrology

DOI: 10.1016/j.jhydrol.2025.133635

Authors: Yi Wang, Jennifer L. Catto, Robin Chadwick

Institutions: University of Exeter, Met Office

Language: English

Topic: Meteorological Phenomena and Simulations

Subfield: Atmospheric Science

Field: Earth and Planetary Sciences

Domain: Physical Sciences

SDG: Climate action

Open Access status: hybrid

Estimated APC paid: $4,030