How it works
- NISAR L-band GCOV, track 061 descending, 11 dates 18 Jun to 28 Oct 2026, averaged to 60 m.
- Each date is normalised against the scene so rain and calibration shifts cancel.
- A sequential omnibus test (Conradsen 2016; Canty 2020) flags pixels whose HH+HV backscatter changed, at a 1% false-alarm rate; HV must stay down for two dates.
- Patches of 1 ha or more are checked against NASA FIRMS VIIRS and MODIS hotspots within 1 km, between the previous radar date and the change date.
- The range comes from a Monte Carlo run over hotspot position, buffer and time window; the area interval from a labelled sample (Olofsson 2014).
Research prototype using provisional NISAR data. Shows where land changed, not who caused it. Not for legal or enforcement use without verification on the ground.
Full sources: see the README in the repository.