Characterizing extreme rainfall using Max-Stable Processes under changing climate in India

Abstract

Climate change has markedly intensified the frequency and intensity of extreme rainfall events globally over recent decades. The present investigation introduces a novel approach to modeling Intensity-Duration-Frequency (IDF) curves for major river basins in India using max-stable processes (MSPs). In contrast to earlier studies that mainly dealt with univariate extreme value theory and point-based IDF curves, this work uses a variety of MSP characterizations, such as Brown-Resnick, Schlather, Geometric Gaussian, and Extremal-t, to capture the spatial dependencies and non-stationary characteristics of extreme rainfall. This comprehensive two-stage modeling approach incorporates geographical covariates to capture spatial variation in extreme rainfall, followed by additional climate-informed covariates. One hundred fifty-six surface response models are analyzed across nine hourly extreme rainfall durations over 11 river basins.

The Brown-Resnick process effectively captured spatiotemporal dependencies across all durations in the annual timeframe, while the Geometric Gaussian process also demonstrated strong performance. During the Indian Monsoon season, distinct covariates such as the Southern Oscillation Index (SOI) and Global Temperature Anomaly (GTA) significantly influenced extreme rainfall patterns. The analysis reveals that the Brahmaputra basin consistently exhibits the highest short-duration extreme rainfall, while the Indus basin shows the lowest. Long-term projections indicate alarming trends, with potential short-duration extreme rainfall reaching 338.9 mm for a 100-year return period in the Godavari basin. The findings highlight the importance of updating IDF relationships in climate variability, providing insights that could lead to disaster preparedness and resilience planning for vulnerable communities across India.