CLEAR DATA. CLEAR DECISIONS. · Market Intelligence methodology
This page is the reference behind the Market Intelligence dashboard: where each number comes from, how it is checked, and where it is less certain.
The product has two halves: a crop map — a 10 m deep-learning mask that finds the crop and measures its area — and in-season crop condition, read daily from weather and satellite. With a yield record they combine into a supply forecast (yield × area); without one, condition stands alone as a stress signal.
Ten classes of signal — normalised, time-aligned, assimilated with explicit bias and uncertainty — feed the two halves: the crop map finds and measures the crop; crop condition reads how it is doing.
Most of any region isn't the crop. A deep land-cover model finds the pixels that are, at 10 m, from optical and radar imagery - SAR reads through cloud, deep learning catches tree crops under shade canopy.
Every reading that follows comes only from these pixels - and the same mask measures planted area on current-year imagery, a year ahead of official surveys.
Reanalysis weather (0.25°, daily) gives heat, frost and moisture stress; optical satellite (10 m, ~5-day revisit) gives vegetation health - read over the masked pixels only.
Each signal sits against the 10th–90th percentile band of every season since 1980: is today normal, and by how much?
A hot week during flowering costs far more than one early in the season. PhenoWeight weights each signal by how much a stress on that day matters to final yield.
The dashboard marks where the season is now, so a shock reads against how much it matters today - and damaging stress registers weeks before official reports.
A region baseline - normal-year yield - is adjusted by this season's stage-weighted signals, only in ways the crop's biology allows. Official reports enter as biased observations, weighted by track record.
Stress counts events past crop-specific thresholds; confidence is a calibrated spread that tightens as the season runs (2020 US corn: the 95% band narrowed 8.8 → 6.3 bu/ac).
No reliable yield record? We don't fake a number - the same machinery tracks a weather-stress percentile over the production area instead (India sugar, below).
The same pipeline runs for every region the mask shows the crop - Brazil coffee alone spans 5,570 municipalities - then rolls up: yields production-weighted, production by summing.
The mask that selected the signal also measured the area - current-year, forecast forward - giving the production number markets price on, a year fresher than surveys.
Every forecast is judged against data it never saw - benchmarked to USDA WASDE, tested walk-forward, out of sample. Late July 2022: 174.8 bu/ac for US corn, 0.07 off the realized 174.9; WASDE sat 2.1 high for two more months. Area is cross-checked against MapBiomas. Live since June 2026; backtests are labelled.
An earlier read on supply is an earlier read on price. Desks use it two ways:
It is not a release-cadence product; the cadence is fixed:
| Component | Updates | Notes |
|---|---|---|
| Futures | ~1 min | front-month, live feed on the dashboard |
| Weather & signals | daily | reanalysis + satellite over the mask |
| Model outputs | weekly | initialised Monday (data as of Monday) · released via API Wednesday after QA · history back-dated to the same cadence |
India has no reliable sugarcane yield series, so we don't publish one. The product is a standing-area map plus a cane-area-weighted weekly stress score against climatology.
The read splits by water source: irrigated Uttar Pradesh shrugs off heat that rainfed states cannot. Imperfect, stated as such - and still the only weekly in-season read.
ReasonFlooding pooled in the Krong Ana river valley, ~600 m below the coffee belt - no overlap with the highland arabica and robusta.
Market reaction+4.5%. Traders reacted to TikTok and social-media flood footage.
Treefera signalNo impact on supply. The footprint is intact - no sourcing action required.
Intersect the flood polygons with the step-1 mask and exposure reads out in hectares, not headlines.
Mask, observe, weight, model, scale, check. The method is fixed; the commodity is the mask you point it at.