Physics-based 18-month lead forecast of Indian Summer Monsoon Rainfall

Abstract

Long-lead forecasts (6, 12, 18-month leads) of the Indian Summer Monsoon Rainfall (ISMR) are crucial for adaptive planning and damage minimization against climate change-induced increasing threats of higher frequency and intensity of hydrological disasters in the coming decades1–4. A recent study5 indicates that the ISMR is potentially predictable up to 24-months in advance as its variability is closely associated with the slowly varying Global El Niño and Southern Oscillation (G-ENSO). However, the growth of small initial errors and the chaotic nature of the climate system makes it formidable to achieve the ‘potential skill’ of long-lead ISMR forecasts by global Atmosphere-Ocean General Circulation Models (AOGCMs). Overcoming the challenge, here, we unravel a novel ‘hybrid long-lead forecast system’ for ISMR where predictor discovery is done using a deep learning Artificial Intelligence (AI) model trained on the nonlinear relationship between ISMR and the global depth of 20O isotherm (D20) from a large ensemble of AOGCMs and past observations. Our hybrid model demonstrates successful prediction capability of almost all ISMR droughts/floods at 18 and 5-month leads between 1980-2011. Our findings establish the physical basis and methodology for long-lead seasonal prediction of tropical systems like the ISMR and ENSO with unlimited socio-economic benefits.