And the predictions have turned out to be accurate, with the robust evolution of a strong El Niño by the summer. Climate models predicted the 2026 El Niño during early spring, just as they had in 2023.
But the real cause for concern is the erratic, and unprecedented, evolution of the monsoon across space and time in 2026, which the models have not been able to capture even a few days in advance.
The El Niño is one of the most predictable modes of natural variability. And this is for the all-Indian monsoon rainfall (AIMR); long-lead forecasting of monsoon onset and space-time evolution are even greater challenges. The monsoon domain is blessed with more than a century’s worth of data. Experts have extracted significant knowledge from this rich dataset of monsoon variabilities in space and time, and have made considerable advances in understanding the intrinsic timescales of active-break cycles, the onset becoming more erratic, delayed withdrawal, switching between extreme dry and wet events, more frequent extreme wet spells over Northwest India, and so on. However, their understanding of the mechanisms underlying these events remains incomplete. The hope is that complete climate models with coupled land-ocean-atmosphere components can shed light on the missing links and help scientists make better, more skillful predictions. The multi-tiered approach taken to develop separate tools for short, medium, and extended range predictions has yielded many impressive advances in useful predictions for various sectors, including agriculture, water, energy, health, and so on. To this end, hybrid dynamic and AI models could deliver sector-specific predictions at the requisite space-time scales for users to better manage disasters, farm-related decisions, and water and energy use. Extensively engaging users with sustained feedback can pave the way for decision-support products to be co-produced with the people who make those decisions.
Models’ ability to predict the monsoon is much lower, hovering at around 60%, which means prediction failures will also occur more often. But models being able to predict its occurrence to the tune of 80% still means they will be wrong one-fifth of the time. The failed La Niña prediction of 2024 is a good example.
