Top News

Scientists of IIT Delhi have developed a new Artificial Intelligence (AI) based model, which can predict the arrival of monsoons 18 days in advance. This technology is faster than the current weather forecast models.

This year, monsoon arrived in Kerala 8 days before its forecast, while there was no rain in Delhi despite the yellow alert being issued. At the same time, cloudburst caused havoc in Kullu and Dharamshala of Himachal Pradesh. In this accident, two people lost their lives, 9 people are missing and more than 2000 tourists are stranded there. In such a situation, if the accurate weather forecast was available in advance, then perhaps so much loss would not have happened.

Now this will also be possible, because scientists of IIT Delhi have developed a new Artificial Intelligence (AI) based model, which can predict the arrival of monsoon 18 days in advance. This technology is not only faster than the current weather forecast models, but is also more accurate. Transformer neural network technology has been used in this, which is also used in modern systems like ChatGPT. Professor Dr. Sandeep Sukumaran, Professor Hariprasad Kodamana and two PhD researchers KM Anirudh and Pankaj have worked together in this research.

How is the AI model better?

Professor Sandeep Sukumaran said that this transformer model is very advanced and modern. During the research, he found that the error rate (Error Growth) in this model is much lower than traditional models. When this model was compared with the weather forecast models of America, Britain and other countries, it proved to be better. He told that large supercomputers and heavy resources are required to run the traditional model, whereas the new AI model can also work with a simple GPU, as used in gaming laptops. This will not only save time, but will also save a lot of government resources. This project has been funded by the Ministry of Earth Sciences of the Government of India and the team is working in collaboration with the Indian Meteorological Department (IMD).

Study done with 5 years of data

PhD scholar Anirudh told that he has compared the AI based transformer model with traditional models by studying the weather data of the last 5 years. Comparing it with the existing models of America and Britain, he explained through the graph that the AI model is not only more accurate, but it saves both time and money. In the traditional model, it takes a lot of time to make predictions with the help of supercomputers, whereas the new AI model can give results in a few seconds.

Research was also done on 40 cyclones

PhD scholar Pankaj told that during his research, he examined 40 cyclones that occurred during four years. Traditional models predict the direction and position of the cyclone within a radius of about 500 kilometers, but the AI model was successful in reducing this distance to a great extent. According to Pankaj, traditional models or machine learning models are able to predict only up to 4 days in advance and even in that the track prediction is not accurate. At the same time, accurate prediction of the track has been possible with the help of AI model. IIT Delhi's new AI based model is being considered a revolutionary step towards weather forecasting in the country. This will provide people with accurate weather information on time. At the same time, government resources will also be saved. In future, such models can prove to be very important in giving warning before disasters.