4 min readPuneMar 27, 2026 10:23 PM IST
At the India Meteorological Department (IMD) headquarters in Delhi, Director-General of Meteorology, Dr Mrutyunjay Mohapatra says that “AI will not replace physical models”. “What AI cannot yet do is take into account the physics of the atmosphere. It goes entirely by statistical data. It is found that, if you can complement Physics-based models with AI weather forecasting models, you can generate forecast faster way. We are using about 10 weather forecasting models including AI-based data-driven models.” he says.
He adds that weather forecasting is defined as an initial value problem, i.e. if you have very good observational data, such as the pressure, temperature, wind speed, wind direction, low-pressure systems or high pressure systems, then weather forecasting models can generate accurate forecast.
“The science of meteorology is data driven. AI/ML (machine learning), too, is nothing but a data-driven model. The only difference is that, earlier, we were using the physical model. These physical observations were converted into digital data. It needed high computing power and many hours. Initially AI/ML weather forecasting models are trained on huge historical observational data. This initial training is typically performed only once. Once trained, the model is used to generate weather forecast based on current observational data in a few minutes.” says Dr Mohapatra.
IMD has established a dedicated functional group on the “Application of AI/ML in weather and climate,” and the scientists of this group are working in this domain.
There are three AI/ML models being employed at the NCMRWF and these models are in experimental stage at IMD. These models are Pangu, GraphCast and FourCastNet and used for generating weather forecast. IMD also uses AI/ML-based Advanced Dvorak Technique (AiDT) to estimate the intensity of cyclones. “The AI improve forecast of the cyclone track, but not the intensity forecast. When we go into extreme weather, AI has limitations,” says Dr Mohapatra.
Until then, and even now, weather forecasting is dominated by human effort. In any typical observatory, meteorologists work round the clock monitoring instruments, such as anemometers, wet and dry bulb thermometers, hair hygrometer, thermometers to measure soil temperature, gauges and the Campbell–Stokes Bright Sunshine Hours Recorder, among others. Every morning and evening, a hydrogen balloon carrying sensitive instruments is sent up into the atmosphere to understand the upper air conditions, such as temperature and wind velocity.
Weather observations are essential to understand the current state of the atmosphere, which helps in drawing scientific conclusions. They are useful for meteorological analysis and short-range weather forecasting up to 24 hours. “Beyond 24 hours, we rely mainly on Numerical Weather Prediction (NWP) models. These models simulate geophysical fluid dynamics, thermodynamics, and complex atmospheric processes. NWP models have their own limitations. Therefore, we do not depend on a single model to prepare the final forecast,” says SD Sanap, Scientist, IMD Pune.
Story continues below this ad
He adds that running these models is not possible on a normal PC or servers. “They require high-performance computing facilities (supercomputers). The Ministry of Earth Sciences provides such facilities at IITM Pune and National Center for Medium Range Weather Forecasting Centre (NCMRWF) in Noida. Data from surface, upper-air, radar, ship, buoy, and satellite observations are used as inputs for NWP models. These models are run centrally, and their outputs are available to all users,” he says.
The outputs act as guidance for predicting weather over the next few days. As mentioned earlier, models have limitations, but they still provide the best possible guidance. “In addition, Multimodel Ensemble (MME) forecasts are prepared by averaging outputs from different models to reduce errors. Subject knowledge and expert judgment are always necessary. We cannot use model outputs directly without analysis. AI models learn from past weather data but do not fully consider physical processes. Therefore, human expertise and intervention are always required in weather forecasting,” says Sanap.
Click here to join Express Pune WhatsApp channel and get a curated list of our stories
© The Indian Express Pvt Ltd
Source link
#weatherWhy #weather #forecasting #human #effort
