'Nobody asked us to work on this. It was on our own that we decided to embark on this journey.'
It took five long years for a group of few scientists to develop a high-resolution (~6 km resolution over global tropics) weather model on their own in their lab at the Indian Institute of Tropical Meteorology (IITM), Pune, without much assistance from anyone.
It took another three years for them to get the model data evaluated and subsequently to get their paper published in a peer reviewed scientific journal, EGU (European Geophysical Union) Geophysical Model Development (GMD).
In the meantime, the lead scientist Dr Parthasarathi Mukhopadhyay retired from IITM, Pune.
Finally, the weather model is further customised and ready for the Indian Meterological Department to use. In fact, for anyone in the world to use.
The model -- named Indian Institute of Tropical Meteorology (IITM) High-Resolution Global Forecast Model in the scientific paper -- is now officially called the Bharat Weather Forecast System.
Till this model was developed, India had only used models developed by other countries and customised it for application in India.
Dr Mukhopadhyay tells Rediff's Shobha Warrier how IITM scientists achieved the unexpected, developing India's first km-scale weather forecast system, the world's first high-resolution model for operational purpose.
How different is the weather forecast system developed by you and your team from existing weather models?
The major difference is in the resolution.
While the existing systems could forecast weather patterns over a 12-kilometre grid which is ~144 sq.km, this new system can accurately predict climatic conditions in a 6-kilometre grid (~36 sq. km).
When the resolution of the model is 12 km, it cannot give any information about weather variability if it is less than 12 km. So, basically the rainfall prediction by the 12 km grid model has uniform value over the whole of 144 sq.km.
As you know, when we increase the resolution, we get more information, like the resolution of a camera.
Till now, India was using the Global Forecast System developed by the US and adopted by India.
In weather prediction, we were always a step behind the Americans at NOAA (National Oceanic and Atmospheric Administration), or the Europeans at ECMWF (European Centre for Medium-Range Weather Forecasts), or the British at the Met Office.
Is that why you decided to develop a new system with higher resolution?
In 2017, we had organised an international conference called INTROSPECT at IITM in Pune. I had invited many weather experts from across the world to introspect our current status and future ahead.
Peter Bechtold, the principal scientist at ECMWF, is one of the best in the world in weather forecast. He was a speaker at the conference.
He explained how reducing grid sizes in their weather model allows precise weather predictions in small areas. He said that it improves false weather information particularly spurious orographic rainfall near the mountainous region.
Another thing he said was, this model was very efficient in super computers and we could easily scale up too.
I then wondered whether we could also use the same system in GFS so that we could get a better resolution. It was just a thought.
As I had said earlier, we have not built any weather models in India, we have only adopted models from other countries.
It meant if we were to build one in India, we had to start from scratch. So, I knew it was a gigantic task.
What gave me confidence to go ahead with the idea was I had some very talented scientists in our lab like Dr R Phani Murali Krishna, Dr Siddharth Kumar and Dr A Gopinathan Prajeesh who were quite good in computer coding.
I discussed with them whether we could develop a model like the one Peter Bechtold spoke about.
Did you think of adopting the model they were using as India was adopting the models developed by other countries?
We couldn't adopt their model as ECMWF (the European Centre for Medium-Range Weather Forecasts) does not share their model with anyone. Although they generate forecast for the entire world, they share their resources only with the European Union.
So, there was no way we could adopt their model.
In fact, I had a discussion with Peter Bechtold and Nils Wedi, the head of their division having profound experience in model development and I asked them whether we could develop a model like theirs. They said, in principle, we could. When I asked whether we could expect any help from them, they said, 'It's not permitted'..
But they told me that there were some papers online which we could go through and if we had any difficulty in understanding them, they were ready to explain.
That is how we took the decision to build a high resolution weather model here at IITM in Pune.
When did you start building the model?
In 2017. We started with the basic dynamic code which itself took around 6 months.
The next step was to run the model in Aqua planet mode (with only sea and no land as an idealised prototype of Earth) and checking.
The third step was adding the land area including the mountain ranges, the vegetation and the topography and run the model again.
The problem was, when you start adding more complexities, the model slows down and shows more complications and errors. The moment you add mountains, for example, the model goes wild.
Finally, we added the physics part which is the process that produces clouds, rain, etc.
That is when you get the complete realistic model.
But the model did not run in any of the computers we had at the lab as they were not powerful enough.
Then, our scientists tried to optimise the code. And finally it started running in the earlier computer we had.
By the time we achieved this, it was 2022. Five years had passed.
So, all these five years, your team was working only on developing the weather model....
Yes. Yes.. We did nothing but work on the model.
Even when we ran the model in the old computer, it took 10 to 12 hours for generating a 10 day forecast. And the results were in delayed mode. It was like we got today's forecast tomorrow!
Only if we demonstrated the model worked could we ask for a super computer.
Once we felt we were successful, I shared the results with Peter Bechtold and Nils Wedi. They also felt the model was good enough except for a little bit of fine tuning.
From 2022 June to 2023, we were continuously running the model without any problem. In fact, the whole of 2022 went in testing the model.
Some more scientists who were experts in specific areas like cyclone, heavy rainfall, etc joined us to analyse them specifically, namely Dr Medha Deshpande, Dr Malay Ganai, Dr Tanmoy Goswami, Ms Snehlata Tirkey, Dr Sahadat Sarkar, Dr Radhika Kanase, Dr Kumar Roy, Dr Revanth Reddy.
Even after all those additions, we found that the model worked well.
That was when we decided to send the paper for publication to EGU (European Geophysical Union). The model was named Indian Institute of Tropical Meteorology (IITM) High-Resolution Global Forecast Model.
We submitted the paper in May 2024, and after all the discussions and testing of the model, it got published in March 2025.
Now, the paper is available free online for the entire world.
Anybody who wants to develop it or expand on it can do so. I am of the opinion that only when a scientific study is available free for others to work on, it will be improved upon.
And it enhances the prestige of the model developing country and centre if other scientists, other countries use it.
Once the new HPC 'Arka' was procured by IITM with generous support from the Government of India, the ministry of earth sciences and after substantial optimisation to make the model run in just three hours, it is thought the model was ready to be handed over to IMD for use.
I can say that with the 6 km model, India is at the top of the globe.
Is it true that India is the only country with such a high resolution 6km model?
Yes. It is true. Right now, India is the only country with a 6 km resolution weather model. Even the European Union model resolution is 9 km.
But I cannot say what will happen next year. Some other country may develop an even better model. But whatever it is, India has made a 6 km model for the first time in the world! Nobody can change that!
Another important feature of this model is, it has variable resolution. For example, when you are near the Poles, the resolution is not 6 km. It is ~10km. And when you move from the North Pole or South Pole to the Equator, the resolution changes to 9 km to 8 km to 7 km and near the Equator, it is 6 km.
It is advantageous because it is near the tropics where we experience monsoon, very heavy rain, cyclone etc.
How satisfying was it for you when the model worked perfectly?
It was very satisfying for all of us. When I showed the model to the international community, they were rather surprised as India was not a weather model developing country.
While countries like US, Germany, Japan, South Korea, UK, etc developed weather models, we were always users of the models they developed.
So, our model was the first weather model developed by India.
Now, it has been named Bharat Forecast System....
When we published the paper, we had given the name Indian Institute of Tropical Meteorology (IITM) High-Resolution Global Forecast Model.
The new name is an extension of the Make in India, Made in India theme as it is made by Indian scientists.
The name sounds good, I feel.
IMAGE: Dr Parthasarathi MukhopadhyayYou said you retired before the paper got published...
I retired on July 31, 2024 and at that time, it was fully ready.
We submitted the paper in May 2024 for publication when we made sure that the model was successful and ready.
We spent 5 years of our life building this model.
There were moments when it did not run and that was when I thought of those young scientists who were working with me.
My worry was, if it failed and did not work, they would lose five years of their time. I would be morally responsible for that.
I would retire in July 2024 but they had a long career ahead of them.
There were days of uncertainty.
Nobody asked us to work on this. It was on our own that we decided to embark on this journey.
Anyway, all's well that ends well!
Feature Presentation: Aslam Hunani/Rediff
© 2025