GroundWise is a weather downscaling and agro-meteorological advisory prototype by Team ZenTechs. It explores how a regional forecast can be inspected as a local 5 km field, with Panchayat context, uncertainty-aware readings, and practical farm-planning prompts.
Open the deployed experience at groundwise.vercel.app.
| Overview | Local grid inspection |
|---|---|
 |
 |
| Panchayat-context map | Field-planning evidence |
|---|---|
 |
 |
This repository deliberately contains no application code, model weights, credentials, API implementation, or private-repository material. It is the public record for the prototype: its intended use, result summaries, research basis, limitations, and next milestones.
GroundWise is aligned to SIH problem statement 26074: Downscaling of weather forecast from Block level to Panchayat level for agro-meteorological advisory services (Ministry of Earth Sciences / India Meteorological Department; Agriculture, FoodTech & Rural Development).
The prototype presents the decision trail from regional input to local detail:
- Choose a place, a Panchayat-context point, or a 5 km grid cell.
- Inspect the parent 25 km block alongside its native 5 km child cells.
- Read the local estimate, available range, source context, and map position together.
- Select crop, growth stage, and reported field-moisture context for dated planning guidance.
| Area | Demonstrated behaviour |
|---|---|
| Local forecast inspection | Side-by-side 25 km parent-block and 5 km native-cell views, with geographic context and exact cell reading on hover/select. |
| India map interaction | India-only basemap, restrained zoom, location search, local weather patches anchored to the selected model cell, and responsive mobile layouts. |
| Administrative context | Real Pune pilot Panchayat/block identity plus a viewport-loaded LGD-derived 2024 national context layer that distinguishes coded Panchayats, explicit urban records, missing mapping, and provider failure. |
| Uncertainty | Selected-cell ranges attach only when coordinate, date, forecast lead, and source run match. A range is never copied from a neighbouring cell. |
| Field planning | Dated crop, maturity, weather, and reported-moisture-aware prompts for spraying, irrigation, harvesting, heavy rain, heat, and humid conditions. |
| Transparency | Research, validation metrics, rule evidence, loading states, and a downloadable plain-text field bulletin are exposed in the product. |
The current GroundWise-R regression report evaluates a 122-day 2025 perfect-prognosis benchmark using ERA5 coarse input and CHIRPS / ERA5-Land targets. These are offline benchmark results; they are not a claim of live operational Panchayat accuracy.
| Variable | GroundWise-R MAE | Bilinear MAE | Relative change |
|---|---|---|---|
| Rainfall | 2.636 mm/day | 3.864 mm/day | 31.8% lower MAE |
| Maximum temperature | 0.568 °C | 1.235 °C | 54.0% lower MAE |
| Minimum temperature | 0.598 °C | 1.317 °C | 54.6% lower MAE |
| Relative humidity | 1.784 percentage points | 3.249 percentage points | 45.1% lower MAE |
| Wind speed | 0.176 m/s | 0.242 m/s | 27.2% lower MAE |
Rainfall performance varies substantially by event intensity. At the ≥2.5 mm/day event threshold, CSI is 0.505 for GroundWise-R versus 0.434 for bilinear interpolation. At ≥64.5 mm/day, CSI is 0.0587 versus 0.0968. Heavy-rain skill needs further work and should not be overstated.
Read the full public material:
- Prototype experience
- Method and uncertainty
- Validation results
- Advisory approach
- Problem fit, innovation, and roadmap
- Evidence and source record
GroundWise does not yet publish an area-weighted forecast for every Panchayat polygon. The visible weather value is a selected 5 km model cell with Panchayat context where source geometry is available. The prototype still needs IMD administrative-block input, verified Panchayat polygon aggregation, local observation validation, reviewed agronomic rules, and operational-language delivery before it can be presented as a Panchayat-level public service.
Daily forecasts do not establish an hourly spray window, measured soil moisture, crop disease, field gusts, or guaranteed outcomes. Product labels, local observations, and agronomist guidance remain necessary for any farm operation.
Made by Team ZenTechs.