IMD launches pilot weather forecast within 1 km radius in UP, national roll out in 2-3 years
The India Meteorological Department (IMD), in partnership with the National Centre for Medium Range Weather Forecasting (NCMRWF), launched a pilot high-resolution weather forecast system in Uttar Pradesh on May 12, 2026, capable of generating rainfall forecasts at a spatial resolution of 1 km up to 10 days in advance.
The system uses AI-driven downscaling of coarser numerical weather prediction (NWP) output, integrating data from automatic rain gauges, automatic weather stations (AWS), Doppler weather radars, and satellite-based rainfall datasets.
Forecasts will be issued at district and block levels, with a 10-day monsoon outlook released every Wednesday; dissemination channels include mobile apps, SMS alerts, WhatsApp, Kisan portals, television, digital displays at vegetable markets, and rural self-help group networks.
IMD officials stated that the service will be extended nationally in 2–3 years as observational infrastructure (AWS, radar networks) is expanded to other states.
Spatial Resolution in Meteorology and Its Agricultural Significance
Spatial resolution in weather forecasting refers to the size of the smallest geographic unit for which a distinct forecast value is computed. Global models like ECMWF's IFS resolve at ~9 km; India's operational IMD-GFS runs at ~12 km. District-level forecasts, the previous finest publicly available service in India, represent areas of hundreds to thousands of square kilometres. A 1 km resolution forecast resolves individual valleys, urban heat islands, irrigation-influenced microclimates, and topographically forced rainfall — all of which matter profoundly for smallholder farm decisions in a diverse landscape like UP (which alone spans plains, Vindhyan hills, and the Terai).
Key Details
- India's agricultural plots average below 1 hectare; a district-level forecast can cover tens of thousands of such plots with a single number — masking local variability entirely.
- At 1 km resolution, differential rainfall predictions become meaningful: a farmer in the irrigated belt near the Ganga can receive a different forecast from one 10 km away in the rain-shadow zone.
- UP's agricultural diversity — paddy in the east, wheat-sugarcane in the west, horticulture in the hills — makes spatially fine forecasts especially valuable.
- Urban heat island effects, fog formation in the Indo-Gangetic Plain (responsible for major aviation and road disruptions), and flood-triggering localised convective events are all better captured at 1 km scale.
The UP pilot is specifically designed to demonstrate the value of 1 km resolution forecasting in one of India's most agriculturally and demographically significant states, prior to national scaling.
NCMRWF and the AI Forecasting Ecosystem
The National Centre for Medium Range Weather Forecasting (NCMRWF), under the Ministry of Earth Sciences (MoES), is India's primary centre for medium-range (3–10 day) NWP. It operates the NCUM-G (global) and NCUM-R (regional) models and runs global AI foundation models — including Pangu-Weather, GraphCast, and FourCastNet — experimentally on its Arunika Supercomputer at 25 km resolution. The 1 km UP pilot represents a major leap in operational resolution for NCMRWF, achieved through AI super-resolution downscaling techniques including Generative Adversarial Networks (GANs) and CNN-based nowcasting.
Key Details
- AI downscaling (statistical/ML-based) learns from historical relationships between coarse NWP output and high-resolution observations; it is computationally far cheaper than running full NWP at 1 km resolution globally.
- Fusion of multi-source observational data (AWS, rain gauges, Doppler radars, satellite) corrects model biases and improves local accuracy in real time.
- AI-hybrid systems have demonstrated 20–30% improvement in rainfall accumulation skill at medium range compared to raw NWP baseline.
- NCMRWF's AI/ML integration roadmap includes bias correction, post-processing, downscaling, nowcasting, and multi-source data fusion — a comprehensive overhaul of the NWP-to-product chain.
The 1 km UP pilot is the first operational deployment of NCMRWF's AI downscaling capability; success here is the basis for national rollout.
India's Meteorological Observational Infrastructure
The quality of any weather forecast — AI or otherwise — is bounded by the density and quality of ground observations. IMD's observational network includes: ~900+ Doppler Weather Radars (DWR) sites (expanding under Mission Mausam), ~6,000+ Automatic Weather Stations (AWS), ~1,200+ rain gauge stations, INSAT/EOS satellite constellation for atmospheric sounding, and radiosonde balloon networks. The UP pilot's integration of all these sources reflects a "data fusion" approach that is essential for hyperlocal accuracy.
Key Details
- Doppler Weather Radar (DWR) can detect rainfall and wind within a ~250 km radius at fine temporal resolution (5-minute scans) — critical for nowcasting and short-range forecast verification.
- AWS data provides real-time surface observations every 15–30 minutes, feeding model initialisation and AI correction algorithms.
- Gaps in AWS density in rural/remote areas are the primary barrier to national 1 km rollout — the 2–3 year timeline reflects the pace of infrastructure expansion.
- Mission Mausam (announced in the Union Budget 2024–25) aims to modernise and densify India's entire weather observation network.
The 2–3 year national rollout timeline is not a technology constraint — the AI models are ready — but an infrastructure constraint. As AWS and radar coverage expands, the 1 km service can be replicated state by state.
- Forecast resolution: 1 km spatial grid (compared to 12–25 km for standard NWP)
- Forecast horizon: Up to 10 days in advance
- Issuance frequency: Weekly (every Wednesday)
- Pilot state: Uttar Pradesh — India's most populous state and a major agricultural hub
- Developed by: NCMRWF (National Centre for Medium Range Weather Forecasting) under MoES
- Data inputs: Automatic rain gauges, automatic weather stations (AWS), Doppler weather radars, satellite-based rainfall datasets
- Dissemination channels: Mobile apps, SMS, WhatsApp, Kisan portals, TV, marketplace digital boards, rural SHG networks
- National rollout timeline: 2–3 years, subject to observational infrastructure expansion
- Technology: AI-driven downscaling (GANs, CNNs, super-resolution techniques) applied to NWP output
- Parent models at NCMRWF: Pangu-Weather, GraphCast, FourCastNet (running at 25 km on Arunika Supercomputer)
- Key ministry: Ministry of Earth Sciences (MoES) — parent of both IMD and NCMRWF