Sub-seasonal Modulation and Predictability of Indian monsoon hourly Rainfall Extremes

TL;DR

This paper uncovers the physical factors behind the predictability of Indian monsoon hourly rainfall extremes, develops a tracking algorithm for storm systems, and suggests data-driven models could forecast these events over a week in advance.

Contribution

It establishes a physical basis for the medium-range predictability of rainfall extremes and introduces a novel storm tracking algorithm for improved forecasting.

Findings

Rainfall extremes cluster with temperature, MISO phases, and vapor.

Storm systems embedded in mesoscale clusters influence extremes.

Deep learning models can potentially forecast organized rainfall extremes over a week.

Abstract

Hourly rainfall extremes cause some of the most destructive weather disasters, yet numerical weather prediction models still struggle to forecast them, and a physical basis for their predictability remains unclear. Here, we identify a trivariate clustering of hourly rainfall extremes with surface temperature, phases of the Monsoon Intraseasonal Oscillation (MISO), and precipitable water vapor, establishing a physical foundation for the medium range predictability of these events. This clustering arises from multiscale interactions in which extremes organize into storm systems embedded within mesoscale convective clusters and synoptic low-pressure systems during active MISO phases. We develop an algorithm to identify, track, and monitor these storm systems. Although rapid error growth limits the prediction of isolated hourly extremes, our results provide basis for a physics informed training of deep learning, data driven models to forecast organized clusters of hourly rainfall extremes more than a week in advance, offering substantial potential to reduce losses from extreme rainfall.