🌾 I’ve been building SafraWatch, a working prototype that turns open satellite archives into a per-field, plain-language picture of what a farm has been growing — click any field over the study municipality and it reconstructs the field’s whole cropping history.

It rebuilds a field’s multi-year Sentinel-2 NDVI/EVI record, fills the rainy-season cloud gaps with Sentinel-1 radar, and automatically detects each crop cycle — estimated sowing, start / peak / end of season (SOS / POS / EOS), and how many crops were grown per agricultural year. The season-detection engine extends my published Start-of-Season method (RSASE, 2025) from a single sowing date into a full per-field history. The study region is Medianeira, Paraná.

🚀 Launch the live demo

Early prototype served over HTTP on a research server — your browser will flag it “Not secure”; it collects nothing. For research and teaching only, not for agronomic recommendations.

Whole-municipality cloud-masked NDVI for a chosen date — one click loads any field’s full history.

What it does

  • Automatic season detection — estimated sowing, SOS / POS / EOS and cropping intensity (single / double / triple crop) per agricultural year, straight from the NDVI time series.
  • Optical + radar fusion — Sentinel-1 backscatter fills the cloud gaps that break optical NDVI through the wet season.
  • Year-over-year field profiles — every season overlaid on a shared agricultural calendar and compared to the municipal mean.
  • Regional context — MapBiomas land cover, Hansen forest loss, and IBGE (PAM) municipal crop statistics.
  • Downloadable — per-field time series, detected seasons, and the field boundary as CSV / GeoJSON.

Year-over-year NDVI for a single field, with the detected sowing / SOS / POS / EOS markers.

Detected crop seasons by agricultural year — the summer + winter (safrinha) double-crop pattern the region is known for.

Built in Python (FastAPI + DuckDB) over a Sentinel-2 / Sentinel-1 data-cube pipeline, with a Leaflet + Chart.js front end, running on a small cloud deployment. It is an early proof of concept — the source code will be released once the demo is validated.