PIB TrackerIMD launches first AI monsoon forecast system
India's weather agency rolls out an AI-driven monsoon-advance forecast and a 1-km rainfall pilot under the Earth Sciences ministry.
What happened
- The Minister of State (Independent Charge) for Science & Technology and Earth Sciences launched two AI-based weather products under the Ministry of Earth Sciences (MoES).
- The first is an AI-enabled "Forecast of Monsoon Advance over Different Parts of the Country" — described as the first-ever AI-driven forecasting system built by the India Meteorological Department (IMD).
- The second is a "High Spatial Resolution Rainfall Forecast for Uttar Pradesh", offered as a pilot service at 1-km resolution.
- Both were developed jointly by the IMD, the Indian Institute of Tropical Meteorology (IITM) Pune, and the National Centre for Medium Range Weather Forecasting (NCMRWF).
- The monsoon system issues probabilistic forecasts every Wednesday, up to four weeks ahead, serving farmers across 16 States and more than 3,000 sub-districts.
- Dissemination runs through the framework of the Ministry of Agriculture and Farmers' Welfare, linking the science output directly to the field-level advisory chain.
Background & context
The IMD is India's national meteorological service, founded in 1875, and it functions as a subordinate office of the Ministry of Earth Sciences (MoES) — not of any agriculture or disaster body, a distinction worth holding for the exam. MoES is the umbrella ministry that also houses the IITM Pune (the country's lead institute for monsoon and climate modelling), the NCMRWF (which runs the operational numerical weather-prediction models), the Indian National Centre for Ocean Information Services (INCOIS), and the National Centre for Polar and Ocean Research (NCPOR). The two products launched here are the joint output of the first three of these — IMD as the forecasting authority, IITM Pune as the modelling brain, and NCMRWF as the medium-range prediction engine.
The launch sits inside a larger programme called Mission Mausam, the MoES flagship approved in 2024 to make India "weather-ready and climate-smart" by deepening observation networks, expanding radar and satellite coverage, and improving forecasting through better physics and now artificial intelligence. The radar build-out cited at the launch — moving from roughly 16–17 Doppler Weather Radars a decade ago to about 50 today, with another 50 planned — is the observational backbone that AI downscaling depends on: a model can only sharpen a forecast to 1 km if it has dense enough ground and remote-sensing data to learn from. The monsoon-advance product, in turn, addresses a long-standing weakness in Indian forecasting: predicting not just how much rain a season will bring, but exactly when and where the monsoon front will arrive — the single date that governs sowing decisions for hundreds of millions of cultivators.
This is also why the work is collaborative rather than the output of a single lab. Monsoon onset is a coupled ocean–atmosphere phenomenon shaped by the Bay of Bengal and Arabian Sea branches, by El Niño–Southern Oscillation and the Indian Ocean Dipole, and by intra-seasonal oscillations that play out over weeks. A purely physics-based numerical model struggles at the extended (two-to-four-week) range; this is precisely the gap where machine-learning systems, trained on decades of historical fields, have shown the most promise globally. The IMD product is India's operational entry into that extended-range, AI-assisted space.
It helps to place the new system inside the forecasting ladder the IMD already runs, because the exam frequently tests where a product sits on that ladder. The IMD issues nowcasts (next few hours, for thunderstorms and squalls), short-range forecasts (one-to-three days), medium-range forecasts (up to about ten days, the NCMRWF domain), extended-range forecasts (about two-to-four weeks), and seasonal/long-range forecasts (the headline monsoon outlook). The AI monsoon-advance product is an extended-range tool — it fills the awkward two-to-four-week window that is too far out for medium-range numerical models yet too near-term for the seasonal statistical models. The 1-km UP rainfall service, by contrast, is a medium-range, high-resolution tool living in the ten-day window but sharpened spatially through AI downscaling. Reading the two products as occupying different rungs of the same ladder is the cleanest way to keep their specifications straight.
"Downscaling" is the technical idea doing the heavy lifting in the UP pilot and deserves a plain definition: large-scale weather models produce output on a coarse grid (tens of kilometres per cell), and downscaling is the process of inferring a finer, local-scale field from that coarse output plus dense local observations. Traditionally this was done with high-resolution physics sub-models, which are computationally expensive; the AI approach instead learns the statistical relationship between coarse fields and observed local rainfall from years of paired data, then applies it cheaply and quickly. That is why the launch stresses the observation network — the ARGs, AWSs, Doppler radars and satellite datasets are the "ground truth" the AI is trained and conditioned on. More radars and rain gauges are not a side note; they are the precondition that makes 1-km AI rainfall forecasting credible at all.
For Prelims
- Entity: AI-enabled "Forecast of Monsoon Advance" — IMD's first AI-driven forecasting system; plus a 1-km "High Spatial Resolution Rainfall Forecast for Uttar Pradesh" (pilot). [source-anchored]
- Nodal ministry: Ministry of Earth Sciences (MoES); disseminated via the Ministry of Agriculture and Farmers' Welfare. [source-anchored]
- Built by: IMD + IITM Pune + NCMRWF — the three MoES weather institutions. [source-anchored]
- Monsoon product specs: probabilistic, issued every Wednesday, valid up to four weeks ahead; covers 16 States and 3,000+ sub-districts. [source-anchored]
- UP rainfall pilot: 1-km spatial resolution, up to 10 days ahead, using AI downscaling of Automatic Rain Gauges (ARGs), Automatic Weather Stations (AWSs), Doppler Weather Radars and satellite rainfall data. [source-anchored]
- Radar network: ~16–17 Doppler Weather Radars a decade ago → ~50 now → +50 planned under Mission Mausam. [source-anchored]
- Accuracy gains cited: ~40% improvement in severe-weather forecast accuracy over the last decade; cyclone track/intensity/landfall (72-hour) forecasts improved ~30–35% over five years. [source-anchored]
- IMD itself: India's national meteorological service, established 1875; a subordinate office under MoES. [curator-added, established public fact]
- The MoES family it belongs to: IMD, IITM Pune, NCMRWF, INCOIS, NCPOR — the institutions you may be asked to match to their function. [curator-added, established public fact]
What it is NOT: This is not the IMD's well-known Long Range Forecast (LRF) of seasonal monsoon rainfall (the April/May "above/below normal" percentage-of-the-long-period-average statement) — that is a seasonal quantity forecast. The new product is an extended-range forecast of monsoon advance (timing and progression), issued weekly. It is also not a Mission Mausam in itself — Mission Mausam is the umbrella programme; these AI products and the radar expansion are activities under it. And the 1-km UP service is a pilot, not a nationwide rollout — a precision the "consider the following statements" pattern loves to test.
Why it matters
The economic stakes of monsoon timing are immense: a large share of India's net sown area remains rain-fed, and the date the monsoon reaches a district decides when farmers sow, which crop they choose, and whether they risk a failed germination from sowing too early. An accurate four-week-ahead advance forecast turns that gamble into a planned decision, and routing it through the agriculture ministry's advisory network means the science reaches the cultivator rather than staying in a bulletin. The 1-km rainfall pilot attacks a second problem — the coarse resolution of conventional forecasts, which can flag "rain over the region" but cannot tell one tehsil from its neighbour. For flood-prone, densely farmed States, sub-district precision is what makes a forecast actionable for irrigation scheduling, urban drainage and disaster pre-positioning. The work also illustrates a wider shift: weather services worldwide are moving from purely physics-based numerical models to hybrid AI-assisted systems that can run faster and extend skill into the extended range, and India entering this space with an operational product keeps its forecasting current with the global frontier.
There is a disaster-management dimension too. Better cyclone-track and severe-weather skill — the ~30–35% gain in 72-hour cyclone forecasts and the ~40% improvement in severe-weather accuracy cited at the launch — feeds directly into the early-warning chain that lets State authorities evacuate ahead of landfall. India's sharp fall in cyclone fatalities over the past two decades is widely attributed to exactly this improvement in lead time and accuracy, and the AI products are a continuation of that trajectory rather than a break from it. By embedding precision at the sub-district scale, the new services also support the "last-mile" goal of early warning: a warning is only useful if it is specific enough for a particular block or town to act on, and a 1-km forecast moves the system closer to that. For an aspirant, the release is therefore a single, well-sourced node that touches agriculture, science policy, and disaster preparedness at once — which is why it carries cross-tags across three GS-III sub-heads.