Application of XGBoost in Disentangling the Fingerprints of Global Warming and Decadal Climate Modes on Seasonal Precipitation Trends in Ohio

Abstract

Global warming (GW) is a defining challenge of the 21st century, driving notable changes in weather patterns. Simultaneously, multi-decadal climate modes, including the Pacific Decadal Oscillation (PDO), Interdecadal Pacific Oscillation (IPO), and the Atlantic Multi-decadal Oscillation (AMO), shape decadal climate patterns and interact to influence regional climates. This study employs the extreme gradient boosting (XGBoost) gain-based feature importance metric to disentangle and rank the contributions of GW and these climate modes to seasonal precipitation changes in Ohio, US, a region known for its variable weather. Monthly precipitation data from 55 weather stations spanning 1960–2023 were analysed using Theil-Sen's Slope method, with statistical significance assessed at the 95% confidence level. Results revealed statistically significant increases in precipitation in winter (3.81 mm/decade) and summer (3.30 mm/decade), with no statistically significant changes in spring and autumn. For winter precipitation, 98% of stations exhibit a statistically significant negative correlation with PDO, while 51% show a significant positive correlation with GW; significant correlations with AMO and IPO are observed in fewer than 41% of stations. Analysing feature importance with XGBoost indicates that the GW signal ranks highest in 32.7% of stations—including the northeastern regions affected by lake-effect snow. In contrast, PDO dominates 58.2% of stations, and AMO in 9.1%. These findings highlight that Ohio's winters are becoming wetter and suggest that, among the variables we analysed, the net effects of PDO, followed by GW, are the strongest predictors of winter precipitation changes in Ohio.