A 6-Kilometre Forecast Grid Is Live Over India. What It Changes — and What It Does Not — for Agri Risk
India's weather model now resolves the country at a scale a district risk team can actually use. The Bharat Forecast System, built by the Indian Institute of Tropical Meteorology with the National Centre for Medium Range Weather Forecasting and the India Meteorological Department, operates at 6 km horizontal resolution — twice as fine as the roughly 12 km grid of the Global Forecast System it replaces, and finer than the 9–14 km at which most global operational models run. The Earth Sciences Ministry told the Lok Sabha in July that India is the only country running a global weather prediction model at 6 km in real time. That is a data-infrastructure fact, not a slogan, and it changes what can be built on top of Indian weather data.
What the finer grid actually buys
Resolution is the size of the box the model calculates for. At 12 km, one box can hold a town, its tank irrigation and the rain-shadow slope behind it, all averaged into a single number. At 6 km, those features start to separate. The ministry says the system's stated objective is forecasts at the level of a cluster of panchayats, and that it has shown better skill on extreme rainfall in the core monsoon region — the category of event that writes off a crop portfolio's worst week. The model carries rainfall predictions up to 10 days, covering the short and medium range that sowing, spraying and harvest decisions actually run on.
Two supercomputers make the runtime possible: Arka and Arunika, at IITM Pune and NCMRWF Noida respectively. Compute is the unglamorous half of this story. A 6 km global model that finishes after the weather has happened is a research artefact; one that finishes on the operational cycle is infrastructure.
The observing system underneath the model
A model is only as honest as what it ingests. Under Mission Mausam, the observing network has thickened with it: 47 Doppler Weather Radars are in operation, putting 87 per cent of the country's area under radar coverage, alongside expanded automatic weather station and rain-gauge networks. BharatFS does not sit alone, either. The February update to Parliament lists a small fleet:
| Model | Resolution | What it is for |
|---|---|---|
| BharatFS | 6 km, global | Operational short/medium-range forecasts, panchayat-cluster level |
| Predecessor GFS (T1534) | ~12 km, global | The grid BharatFS replaces |
| Typical global operational models | 9–14 km | International comparison band |
| Mithuna-FS | 12 km, global coupled | Atmosphere–ocean–land–sea-ice coupled forecasting |
| Mithuna-FS regional suite | 4 km | Monsoon dynamics, cyclones, mesoscale extremes over India |
| Mithuna-FS urban suite | 330 m | Fog, visibility and air quality over Delhi |
Source: Ministry of Earth Sciences statements to Parliament, PIB, 5 Feb and 22 Jul 2026. Table: Farmioc.
The pattern matters more than any single row. India is no longer buying one model at one scale; it is running a stack, from a 330-metre urban grid to a coupled global system, all assimilating radar, satellite and station data. Anyone building agricultural risk products on Indian weather data is building on this stack whether they name it or not.
Where the grid stops helping
Here is the part that gets skipped in the launch coverage. A 6 km forecast is still a forecast for a 36-square-kilometre box. A borrower's two hectares sit inside one box, next to a neighbour whose field drains differently. Parametric insurance triggers, crop-loan early-warning scores and reservoir decisions all live or die on verification against what actually fell on the ground — rain gauges, radar-derived rainfall, and eventually crop-cutting results. Resolution shrinks the model's blind spot; it does not remove it.
Access is the second limit. The raw model output is not directly disseminated to stakeholders; IMD works it into the operational forecasts and advisories it issues. For a bank or insurer, that means the procurement question is not "do we have BharatFS?" — nobody outside the ministry does, directly. The questions are about the layer in between: which products carry the 6 km signal down to a district trigger, how are those products verified, district by district and season by season, and what is the false-alarm rate on the extreme-rainfall warnings your payout rules depend on. A vendor who cannot answer the verification question is selling you the launch headline, not the model.
The decision this leaves on the table
Treat the 6 km grid as a raised floor, not a finished product. Portfolio teams should re-test their weather triggers against the finer-resolution guidance now feeding IMD advisories, and demand verification scores at the lead times their contracts use — a skill claim for the core monsoon region in aggregate says little about a specific district in the third week of October. This is the layer Farmioc's DataStore works at: satellite, weather-model and ground data joined into district-level signals that a credit or insurance team can audit, rather than a national forecast taken on trust. The model got sharper this year. The underwriting still has to be done.