Data fine, interpretation faulty: Why IMD gets forecasts wrong

On January 7, in its three-day forecast, the India Meteorological Department (IMD) warned of “moderate rainfall across central and northwest India” on January 8 and 9 and issued an orange alert for Haryana, Punjab, Delhi, Madhya Pradesh, and parts of Rajasthan, predicting rain and thunderstorms in parts of these states.

January 8 came and went – with no sign of rain in most parts of these states. Ditto with January 9 (although, after the January 8 debacle, IMD tweaked its forecast to “a possibility of isolated showers”).

The problem, experts say, is not the technology or the models – but simply the interpretation of data by weathermen.

Defending IMD’s forecast, a senior weather department official said that a cyclonic circulation was developing over the region at the time the warning was issued.

“Rainfall activity has been reported in parts of Rajasthan, Madhya Pradesh and Uttar Pradesh, but it did not reach Delhi and NCR. Such error margins are completely normal and is likely with the best of forecasters,” added this person who asked not to be named.

But this was not the first time India’s weather office got its forecast wrong. Last month, IMD was also unable to accurately predict the “extremely heavy” rainfall that inundated Tamil Nadu, killing at least 10 people in a span of two days. The state officials mentioned in subsequent press briefings how a warning from the agency could have resulted in better preparedness and less damage.

IMD director general M Mohapatra said, without referencing any specific event, that no weather agency can be accurate all the time and that small misses in the prediction of weather developments such as non-seasonal rains should not be categorised as “mistakes”.

“Over the last few years our forecasts have improved significantly, and we are striving to make it better in the coming years,” he added.

Documents shared by IMD show that the agency has almost completed work on a high-density meso network and high-resolution modelling framework for major cities for early weather and air pollution monitoring/forecast/warning under urban meteorological services.

A mesonet or mesoscale network is a chain if weather stations, usually automated, that observe weather in a range of 10km to 1,000km. It is finer than the synoptic scale, which involves measurements in ranges exceeding 1,000km. And by 2025, the ministry of earth sciences will expand its Doppler radar network with the addition of 33 radars, ensuring comprehensive observational coverage across the country, the documents show.

Which is all fine, but the problem isn’t really of technology or equipment.

M Rajeevan, former secretary of the ministry of earth sciences, explained that the models being used by IMD have improved, and are currently at par with the technology being used in the US, UK and Japan — countries that are known to produce the most accurate weather forecasts in the world. But he pointed out that interpretation of data and satellite images by weathermen is an important aspect of forecasting – and one where IMD lags.

“IMD has access to all the state-of-the-art models along with its own models. The resolution of IMD’s models have also significantly improved over the last few years… An accurate weather forecast depends on two aspects — the model and the interpretation of forecasters… models are just tools and if we are unable to look at multiple satellite images, radars and pick up hints from the models, we will still be missing the mark with forecasts. That is what seems to be happening.”

Rajeevan also said that while there has been some improvement with short- and medium-range forecasts, IMD’s weakness seems to be long-range, seasonal forecasts.

“Seasonal forecasts are more difficult to make. There is a term called ‘predictability’, which means the accuracy with which we can forecast, it is higher for three-day and five-day forecasts, but for seasonal forecasts, the predictability becomes low.”

Mahesh Palawat, vice-president (meteorology and climate change), at private forecaster Skymet agreed and said forecasts can be improved by training weathermen along with improving the technology and increasing ground stations.

“Models without a trained weathermen mean nothing. There is always scope for making forecasts better and more accurate.”

But the IMD official cited in the first instance believes that it makes no sense to compare the forecasting accuracy of weather offices across the US, UK, and India. India, he added, has tropical climate, and predicting weather in such countries is trickier. He also said that the climate crisis is another variable that poses a challenge, but admitted there was room for improvement.

“Tropical climates are more unpredictable and thereby more difficult to predict, as opposed to US and UK which have more systematic weather systems. Climate change is also making the task more difficult for forecasters, because going forward, weather will become erratic and break away from past patterns, but that is not to say that we do not need to adapt to the challenges.”

The problem, meanwhile, persists.