Hildacmd/planting_pipeline

cropmonitor-11 countries-Africa # Crop-specific dekadal planting-window estimation — GHA / ICPAC (11 countries) Estimates **per-pixel planting dekad** for **maize, teff, wheat** inside **crop-specific masks**, by fusing **Sentinel-2 red-edge phenology + FPAR + Sentinel-1 SAR**, constrained by your **long-term-normal (LTN) phenology**, cross-checked against **WRSI rainfall onset**, and aggregated to admin units — **season-stratified** per country using a crop-specific calendar. Countries: Ethiopia, Kenya, Uganda, Tanzania, Rwanda, Burundi, South Sudan, Sudan, Somalia, Eritrea (Djibouti excluded — negligible cropping). **Reference runs (maize 2024):** Kenya Long rains + Short rains, and Ethiopia Meher (unimodal). **Production scale is 250 m** — two concurrent 10 m country exports do not sustain on GEE, so 250 m is the operational scale for the multi-country rollout; 10 m is the single-AOI reference ceiling. ## Where this fits your existing stack You already run: **generic crop mask, LTN phenology, FPAR, WRSI**. This pipeline adds the two missing pieces — **crop-specific masks** and **red-edge + SAR fusion** — and turns the outputs into a **crop-specific cropping calendar** that then *feeds back* into WRSI as a better onset. ``` generic crop mask ─┐ ┌─ LTN phenology (prior + anomaly baseline) crop-specific mask ─┼─► [S2 red-edge + FPAR + S1 SAR fusion] ─► SOS ─► planting dekad ─┐ (maize/teff/wheat)│ (season-stratified, per crop) (─offset) │ FPAR (existing) ────┘ ▼ CHIRPS ─► WRSI onset ──────────────► cross-validate ◄────────────────────► WRSI (better onset in) └─► zonal admin planting-window CSV + anomaly map ``` ## Pipeline stages (code map) | Stage | File | What it does | |---|---|---| | Config | `config/datasets.yaml`, `config/season_calendar.csv` | dataset IDs+APIs; crop-specific calendar | | Optical | `src/s2_preprocess.py` | dekadal cloud-masked NDRE/VI composites (10 m) | | …