Mumbai: In 2023, the world experienced the hottest global temperatures in over 100,000 years, and heat records were broken on every continent. India experienced extreme weather events on 235 of the 273 days (a little over 86%) in the first nine months of 2023--from heat and cold waves, cyclones and lightning to heavy rain, floods and landslides, according to the Centre for Science and Environment. These events killed 2,923 people and over 92,000 animals, affected 1.84 million hectares of crops, and destroyed 80,000 homes, it estimated.

The India Meteorological Department (IMD), the national meteorological service of the country which was formed in the year 1875, enters its 150th year on January 15. The milestone comes on the backdrop of India experiencing a weak winter so far and overall large variations in weather events year on year. For example, from 2013-2022, an average of 29% of India’s land area experienced over three months of extreme drought per year. The amount of land experiencing at least one month of extreme drought per year has increased 138% from 1951-1960 to 2013-2022. The intensity and frequency of heatwaves, extreme rainfall, floods, droughts, landslides, cyclones and every other weather event is changing annually.

IndiaSpend spoke with IMD’s Director General Mrutyunjay Mohapatra, 58, on concerns around the agency’s weather forecasting accuracy, the need for improvement in technology, the role of climate change and the IMD’s future plans.

Edited excerpts from the interview:

As IMD enters its 150th year, what are your reflections and what plans does it have for the coming year?

With a very humble beginning in 1875, IMD has undergone several phases of evolution. It so happened that on October 5, 1864, a massive cyclone came and devastated Calcutta wherein 80,000 people died. In the same year, 40,000 people died in Andhra Pradesh when a cyclone hit Machilipatnam. This event was followed by the 1866 famine because of monsoon failure followed by another famine in 1871. All this led the then government to form a meteorological committee. Henry Francis Blanford was appointed as the Imperial Meteorological Reporter. He joined on January 15, 1875 in Mumbai, and that’s why we celebrate the day as the formation day of the IMD.

As of today, we have more than 1,000 automatic weather stations, 6,000 rain gauges and 550 departmental observatories.

With all our scientific interventions, we have come a long way but there are certain science gaps. We need to improve forecasts for lightning and cloudbursts--right now, there is no forecast in the world. We have increased the number of doppler radars but we have to improve our infrastructure further to improve forecasts. That requires models to be improved which is possible only with higher computing power, and improved resolution of models. Our ambitious plan is that each household and each person should get forecast.

Is that where Artificial Intelligence comes in the picture for IMD?

In the next five years, AI and machine learning will help improve our decision making, forecast accuracy and it will help in sectoral applications. For example, how do we downscale agromet services [forecasts for farming] from block to village level? Similarly in urban areas, how do we reach out to each ward for urban flooding? AI will compliment numerical models. And to do all these things, you need computing platforms which we are developing within the IMD. Also, we have identified eight startup agencies and will be supporting them to develop tools and technology in the future.

How will AI, our models and our human forecasting abilities come together--especially because a former Ministry of Earth Sciences secretary had said that our model has a certain bias: It normally under-predicts heavier rainfall and can only give an indication of heavy to very heavy rainfall.

Every day we analyse all our model guidance. Our 26 state offices analyse observational products, forecast products and come to a consensus. So objective consensus is modulated by the subjective consensus derived from the knowledge and experience of forecasters across the country. Only then are bulletins prepared. In future also, these processes will be helped by machine learning.

Coming to the question of bias, you are correct. Model always underestimates rainfall. That’s why we have forecasters. If we compare model performance vis-à-vis forecasters’ performance, there is improvement in forecast accuracy by 10-15% by the forecasters from the model performance.

In December, after parts of Tamil Nadu received as much as 950 mm rainfall in 24 hours, chief minister M.K. Stalin had criticised the IMD saying its forecast failed to convey the magnitude of rain in Tamil Nadu and that a red alert was issued too late.

There are two aspects here. The first thing is, what was the quantum of rainfall? It was around 94 cm [940 mm]. Such an exceptionally high rainfall is not predicted or cannot be predicted by the models. At present, whatever models are there in the world, it is not possible to predict this type of rainfall.

We had issued the warning for extremely heavy rainfall but until now, there is no such modelling system in the world where you can predict 90 cm rainfall in 24 hours. It is a cloudburst and worldwide, it cannot be predicted. We are improving our observational systems; [for that] you need mesoscale observation network which can provide data [mesoscale meteorology pertains to atmospheric phenomena having horizontal scales ranging from a few to several hundred kilometres]. At the same time, you need computing systems for the processing of the data and you need some kind of understanding of the physical processes shown in a testbed format [an environment for experimentation]. So, all these are the science gaps for which the IMD and various agencies are working together. But if you compare the toll from the 2005 Mumbai floods [also 944 mm rainfall in 24 hours], when there was no such forecast either, to 2023, there is a significant improvement.

There is an improvement in lead period [early warning] in 2023; we have been giving the damage expected from weather events, and there has been forecast not just from state level but also from the district level, the block level and the location-specific level.

Also, in the case of Tamil Nadu cloud burst, we provided a yellow alert five days ahead, then an orange alert three days ahead. We said be prepared to take action with the orange warning. So, I will request all the disaster managers and stakeholders to pay attention to these warnings also [and not just wait for a red alert]. Early warning needs to be supported by early action.

The government had said that there has been a 40% to 50% improvement in severe weather forecasts with a lead period of five days in the last five years. Can this be improved further?

In the last five years, the IMD has come up with a seamless weather forecasting system. There has been a 40% to 50% improvement in forecast accuracy for all types of severe weather compared to the previous five years. There was no ‘nowcasting’ [weather forecast for next few hours] before 2013 which was started in 2013 at 120 stations and now, we have it in about 1,200 cities and towns.

If I look at the forecast for heavy rainfall, it is accurate up to about 80% for 24 hours ahead which used to be about 60% five years ago. If you consider the five-day forecast, the forecast accuracy is about 60% at present. Also, our lead period has gone up from one day to five days with similar accuracy.

Then in terms of cyclones, our forecast accuracy is better than all leading countries. Thunderstorm forecast accuracy is about 86% which was around 58% or so in 2013. Similarly heatwave and coldwave accuracy is 92%; there has been significant improvement.

But still there are challenges with respect to cloudburst and lightning. The number of deaths due to lightning strikes has not decreased. So, we are going to establish a research testbed, where we will be studying and trying to improve lightning forecast. Only five countries in the world are doing this because there are still gap areas here.

Our target is we want to reach out to every panchayat in India and involve all village-level stakeholders. For example, if a person is constructing a house or a contractor is building a road, they should consider what is the climate of the place, what are the extremes of that particular place, what materials should they use. So what I want is we should mainstream weather and climate information in each and every activity of the human being and the society.

In the last decade, we saw that our long range forecasts [monsoon season forecast] did not meet the 5% margin of error in all years except two. Why is that, and how can that be improved?

We have the five best models in the world and that has improved the forecast accuracy and service delivery at a very small-scale error. Between recent 15 years and previous 15 years, there is a 17% improvement and I am hoping we will improve it further.

Critics still say that IMD is unable to forecast the exact regions that will be affected by a certain weather event and the number of houses, structures, fishing boats, people who will be affected. Your comments?

We are open to criticism. Our attempt is to show people what the weather will do tomorrow, to the population, to the infrastructure, to the various communities, to the people in the slums, the people in cities, villages. We have done the risk assessment at the village level where you can find out how many houses will be damaged, how many electric poles, how many telephone poles, how much agricultural area will be affected. All these things are calculated nowadays in a dynamic platform and we also convey the expected losses to the government. All the socio-economic data with respect to heatwaves, cold waves, heavy rainfall, thunderstorms has to be integrated in the system.

As the impact of climate change on Indian weather will become more prominent in coming decades, how is IMD preparing itself?

One is the detection as to whether climate change is occurring or not. Every year in the month of January, the IMD issues a statement on our weather features, what is our long term data, what are the extremes and which records were broken. But climate change limits the predictability [of weather]. If I could forecast a particular heavy rainfall event five days in advance, I'll be able to predict it only two-and-a-half days in advance. The science gets limited because of climate change. But if I look at the past 10 years, our accuracy and lead period has improved. That is how IMD has taken care of the impact of climate change and we will continue to improve on these.

(Misha Vaid, intern with IndiaSpend, contributed to this report)

We welcome feedback. Please write to We reserve the right to edit responses for language and grammar.