Chishang Rice Decision Report
Turning one paddy in Chishang, Taiwan into decisions you can act on — straight from Sentinel-2 imagery. The indices are real, computed live; the recommendations are illustrative.
This is how the satellite reads the field.
Crop vigor · NDVI
0.65
Vegetation index mean (peak 0.92)
SENTINEL-2 · 2026 Q2
Uniformity · σ
0.23
NDVI std-dev — lower is more even
Dispersion
Water · MNDWI
−0.56
Irrigation / flooding signal (peak 0.83)
Green vs SWIR
Chlorophyll · NDCI
0.37
Canopy N / photosynthetic-activity proxy
SENTINEL-2 · 10 m · 2026 Q2
snapshot · 2026 Q2 · pulled live from service-system's satellite MCP (Google Earth Engine · Sentinel-2 Surface Reflectance). Values are current-period composites, not a live stream.
Three findings, one move.
The point isn't the numbers — it's what to do next.
It's growing well. NDVI mean 0.65, peak 0.92 — most of the area is in vigorous vegetative growth.
But it's uneven. σ=0.23 is high — clear field-to-field variation, with laggard low-NDVI plots.
Mostly drained. MNDWI mean −0.56 (canopy, not open water); a few high spots (0.83) still hold water.
Illustrative action: Scout the low-NDVI plots first (nutrient / drainage / pest?) and consider spot top-dressing; check drainage on the wet spots; near heading, sequence harvest by the high-vigor zones.
What does 0.65 mean?
NDVI (Normalized Difference Vegetation Index) measures photosynthetic vigor — the closer to 1, the more chlorophyll.
NDVI scale
This field’s mean sits in the 0.6–0.7 “vigorous growth” band — consistent with rice before heading.
This period's distribution
The green band is mean ±1 std-dev (≈0.42–0.88). Wider band = less even; the −0.28 minimum is mostly bunds, water or bare ground.
Not one snapshot — a whole season's trajectory.
Same field, same parameters, re-scanned across time windows — you see the growth curve AND when to act. Anyone can re-run the same parameters and get the same result.
NDVI seasonal curve (real, two windows)
NDVI peaks in April (0.69, vegetative max), then eases in May–Jun (0.64) as the crop heads and ripens; chlorophyll NDCI 0.37 shows the canopy is still N-rich and photosynthesizing.
The dip is normal phenology, not disease — what matters is the laggard plots (σ=0.23).
What is this scan worth?
Farmers / co-ops: σ pinpoints the laggard plots → targeted top-dressing and scouting instead of blanket spraying — saving fertiliser and labour.
Buyers / processors: the heading-to-ripening curve → estimate harvest timing and quality grade, schedule better.
Insurers / lenders: objective, reproducible indices → loss adjustment and crop-loan risk control.
Indices are real satellite computations; the business translations above are illustrative, not a guarantee or formal advice.
The water drained; the canopy came up.
MNDWI (Modified Normalized Difference Water Index) averages −0.56 — most of the field is no longer open water. The canopy has closed and the paddies have drained: exactly the look of mid-to-late vegetative growth. A few high values (peak 0.83) flag plots still holding water — the priority spots for a drainage check.
MNDWI = (Green B3 − SWIR B11) / (B3 + B11).
One field, two ways to see it.
True color looks like a photo; false color (near-infrared) makes healthy vegetation glow red — the differences jump out.
How it was computed
| Data source | Sentinel-2 Surface Reflectance (COPERNICUS/S2_SR, 10 m) |
|---|---|
| Extent | 121.17–121.25°E, 23.08–23.16°N (Chishang paddy belt, Taitung) |
| Period | 2026-03-15 → 2026-06-14 (≤25–30% cloud composite) |
| Indices | NDVI = (B8 − B4)/(B8 + B4); MNDWI = (B3 − B11)/(B3 + B11); NDCI (chlorophyll) = (B5 − B4)/(B5 + B4) |
| Seasonal windows | NDVI April window (2026-04-01→04-30) = mean 0.69; May–Jun window (2026-05-15→06-14) = mean 0.64. The early Feb–Mar window had no cloud-free scene, so it is excluded (not fabricated). |
| Reproducible | Anyone re-running the same collection_key + bbox + period + cloud threshold in GEE gets the same indices — a third-party-auditable basis for decisions, not a black box. |
| Pipeline | service-system satellite MCP → Google Earth Engine (10+ indices: NDVI/EVI/chlorophyll/turbidity/water/SST/time-series…) |
| Output | Pulled, computed and assembled automatically by an agent |