Projected Changes in the Temporal Organization of Heavy Precipitation under Future Climate Scenarios

Abstract

This study evaluates future changes in the temporal organization of heavy precipitation using CMIP6 simulations (30 models, daily resolution, 1950–2100, ~3.8 million samples) under multiple emission scenarios. Temporal organization is quantified using a persistence index, event clustering coefficient, and a graph-based temporal connectivity metric. A gradient boosting regression model (GBDT, 800 trees) is used to relate climate drivers (temperature, humidity, circulation indices) to clustering behavior. Results show that clustered heavy precipitation events increase by 29–37% under SSP5-8.5, with the clustering coefficient rising from 0.34 to 0.48 by 2100. Mean inter-event intervals decrease from 6.3 days to 4.5 days, while event duration increases by 0.9–1.6 days. Variance decomposition indicates that thermodynamic factors explain 61% of clustering changes, while dynamic circulation contributes 39%. Model performance achieves R² = 0.79 and RMSE = 0.12 in reproducing clustering metrics. These findings indicate a structural shift toward more temporally concentrated and persistent precipitation extremes in a warming climate.