SAR Multi-Crop Acreage — Season Explorer
Set the crop acreage of a hypothetical village and watch how its multi-temporal X-band backscatter changes across the Kharif season. The tool then runs the same linear unmixing the pipeline uses to recover crop areas back from the (noisy) curves — the inverse problem Round 1 must solve.
Co-pol (VV) backscatter — dB
Cross-pol (VH) backscatter — dB
Season summary
–cultivated ha (set)
–agri-extent score
Agricultural-extent score is the temporal coefficient-of-variation indicator: cropland (bare→canopy→harvest) scores high; water and built-up score low.
Recovered acreage (linear unmixing)
| Crop | Set (ha) | Recovered (ha) | Error |
|---|
Regularized least squares against the known land-cover signatures, clamped to ≥0. Single-village unmixing is ill-posed and noise-sensitive (similar crop calendars are hard to separate) — recovery is approximate. The full pipeline reduces this by learning the inverse map from many labelled villages rather than assuming the signatures.
Static verification snapshot from verify.mjs on the synthetic Kharif benchmark: skill ≈ 59% vs naive baseline, pooled R² ≈ 0.59, total-area recovery r ≈ 0.90, submission schema exact. This page inlines the same physical forward model used to generate that benchmark (drift guard: edit src/synth.mjs and this file together).