Research Projects

The First Revolution in Weather Forecasting emerged through Global Forecast Models grounded in physical laws and solved using Numerical Weather Prediction (NWP). It began with the first successful forecast in 1950 (Charney et al., 1950) and reached its peak around 2020. Although the theoretical predictability of weather extends to about 15 days (Lorenz 1963), the intrinsic chaos of the atmosphere has limited forecast skill. Despite substantial advances in model resolution, physics, and data assimilation, useful skill remains stuck at roughly 7 days. NWP has thus reached a critical crossroad. A Second Revolution was therefore essential to push forecast skill and extend range. This new era began in 2023 when 3-D deep learning AI models of the atmosphere demonstrated the capacity to improve skill and extend forecasts beyond the best NWP systems (Bi et al., 2023; Lam et al., 2023; Price et al., 2024). India largely missed the First Revolution, unable to anticipate its potential, and risks missing the Second as well. Despite the enormous benefits such advances could bring to farmers, fishermen, and other vulnerable communities, national priority on this issue appears limited. The PI, during a SERB Distinguished Fellowship at Cotton University, foresaw this Revolution nearly five years ago and trained a PhD student on applying deep learning AI to Indian monsoon prediction. Two pioneering papers on long-range AI-based monsoon forecasting were subsequently published (Sharma et al., 2025a, 2025b). Given his academic contributions and leadership record, the PI is uniquely positioned to guide the Second Revolution in Weather and Climate Forecasting in India. With this foundation, the Proposal seeks to lead the Second Revolution in Climate Forecasting in the country. It outlines nine objectives, including the development of deep learning AI models for long-range monsoon rainfall prediction; extending skill for active–break phases; accurate prediction of onsets over Kerala and the North East; and forecasting the frequency and intensity of daily extreme rainfall. Models will be designed, tested, and validated at Gauhati University and later transferred to IMD for operational use. Alongside these task-specific systems, a more general 3-D global Sub-seasonal-to-Seasonal (S2S) model is also proposed, subject to availability of computational resources. These models will be trained on CMIP6 simulations and fine-tuned with observations or reanalyses. Training will follow a ‘physics-guided’ strategy devised by the PI, ensuring consistency with established physical principles. Establishing the physical basis for predictability will form the foundation for each application. The PI is well prepared, having recently developed, with his student, the theoretical framework for long-lead predictability of the Indian monsoon (Sharma et al., 2022).