India Meteorological Department’s (IMD) first monsoon forecast — it was released on April 15 — has projected above normal monsoon rainfall in 2025. This is one of the most keenly watched numbers not just by those who follow the weather but also economic analysts because of its implications for crop production and inflation. 75% of India’s annual rainfall, after all, is recorded in the June-September period, which is the official monsoon season in the country.
However, an HT analysis of IMD’s first forecasts since 2001 shows that there is good reason to be circumspect about these for two reasons. One, their statistical accuracy is worse than predicting a coin toss, and two, increasingly, even the headline number being right does not help the cause of economic analysis.
The headline number of the monsoon forecast is the estimate of the range of the season’s rainfall relative to the Long Period Average (LPA). The LPA is simply a historical average of rain and is currently defined as the average rain in the 1971-2020 period. The IMD classifies deviations from this period into various categories of monsoon. Currently, a deviation within 4% of the LPA is defined as normal monsoon for the country. This year, for instance, IMD expects the rainfall to be 105% of LPA (which is above normal).
The Centre for Monitoring Indian Economy (CMIE) database has compiled data on IMD’s first forecast for monsoons from 2001 onwards. In the 24 monsoon seasons up to 2024, the first rainfall forecast differed from the season’s actual performance by more than five percentage points in 14 years. IMD’s forecast comes with an error margin of up to five percent. This gives an accuracy of 42% to IMD’s first forecasts, which is worse than the statistical probability of predicting a coin toss (0.5 or 50%).
A breakup of the 14 times IMD got the first forecast of the season wrong by more than five percentage points shows that actual rain (as % of LPA) was higher than the forecast in six seasons and lower in eight. In other words, IMD’s first forecast is more likely to err on the side of optimism.
To be sure, IMD has done slightly better in getting its forecasts right or less wrong in the recent past. Only two of 14 forecasts that were off the mark by more than five percentage points were in the five years ending 2024. The five-year periods ending 2019, 2014 and 2009 had four, three, and three forecasts off the mark by at least five percentage points. This means the agency has been getting better with its forecasts.
The second and more important aspect of monsoons, one that warrants greater circumspection, at least vis-à-vis economic implications of the monsoon, is the increasing geographical and temporal variation of rain for a given amount of headline rainfall in the monsoon season.
For example, the IMD forecast 922 mm rain in the monsoon season in 2024. While the actual rainfall in 2024 monsoon was 936 mm (107.6% of LPA compared to 106% in the first forecast) there was a large geographical variation in the rain. It was 120% of LPA or more for 41% of India and 80% of LPA or less for 19% of India. Rain was within 20% of LPA (an arbitrary threshold considered the normal range for individual places) for 40% of India. This means that more than half of the country received either excess or deficient rainfall. An HT analysis published in these pages showed that the geographical and temporal distribution of rainfall has been becoming more skewed in India. This is likely a fallout of the climate crisis and makes projections more difficult.
To be sure, IMD has incorporated at least regional variations in its first forecast. The first forecast also gives the most probable event for individual places: whether rain is expected to be within the normal range or outside it. However, this data is not given in a format that can be tested for its accuracy.
