Infrastructure

The Modelled Harvest

The official size of a harvest is moving from farmer reports to satellite-model estimates. AI and machine learning are being added next, after the old counts lost market trust.

What is it?

The official number for how much a country harvested is increasingly estimated by a satellite model before anyone counts it. Canada already publishes one, India pays farmers from one, and the US now plans to test AI and machine learning for its own.

Why now?

US corn estimates jumped between first and final figures in 2025, and grain prices fell more than 5% on the revision. USDA answered in September 2026 with a satellite and AI pilot, plus a promised accuracy scorecard. Cocoa’s official count went quiet the same month.

Who’s building it?

National statistics offices, crop-insurance schemes and climate researchers are replacing hand counts with models.

Statistics officesSpace agenciesCrop insurersClimate researchersCommodity bodies

The full read

Why now, who is building it, where the market goes, and what stays unsettled.

Get weekly →

01 · The market

Who’s in it

Builders — statistics offices turning satellite imagery into official harvest figures.

Buyers — grain traders, insurers and farmers who act on the first crop number published.

Gatekeepers — agriculture ministries deciding which method counts as the official number.

What’s at stake

Whether markets will trust a modelled harvest more than the survey counts that moved grain prices on revision.

A farmer’s insurance payout set by a satellite.

Official cocoa figures withheld just as weather risk rises.

What’s unsettled

Whether USDA’s pilot turns into machine learning used in official estimates.

Whether an annual scorecard shows models beating the old surveys.

Whether farmers disputing modelled payouts win a right to a field count.

02 · How the demand matureda progression, not a timeline

Survey

Farmers report and inspectors cut sample plots by hand to size the harvest.

Satellite

Statistics offices publish a preliminary harvest modelled from imagery and weather data.

Payout

Insurance schemes pay farmers on modelled yields, and one state drops the hand count entirely.

Machine learning

The largest crop statistician tests AI and machine learning after its counts lost trust.

03 · Where the market goesthe forks, not a prediction

→Model or survey: offices may make the modelled figure the headline number, or keep it as a first read under the survey.

→USDA’s pilot may put machine learning into official estimates, or stay a test.

→Farmers may accept modelled payouts as faster and fairer, or win back the field count.

→Cocoa and coffee outlooks may adopt yield models, or stay weather commentary.

Satellites can already size a harvest faster than surveys; what decides the shift is whether traders and farmers accept a number no one counted by hand.

04 · The Resilience Tech deskadjacent markets, moving now

The Orbital Lookout

Satellites now spot fires from orbit.

Greece deployed four EU-funded nanosatellites with thermal sensors to spot fires.

Euronews

Adaptation Economy

Building for the heat becomes a market.

Climate adaptation spending could reach $1.3 trillion a year by 2030.

Eco-Business

The Overridden Forecast

Store managers veto the ordering model’s smaller order.

ReFED finds AI ordering advice frequently overridden by store managers.

ReFED

05 · The Wire

USDA will pilot satellite imagery in its acreage and yield estimates

U.S. Department of Agriculture ↗

USDA plans an annual review comparing crop forecasts with final numbers

RFD-TV ↗

Statistics Canada publishes its preliminary harvest from a model built on satellite imagery and agroclimatic data, in use since 2016.

Statistics Canada ↗

Madhya Pradesh gives technology-based yield estimates 100% weight over manual crop-cutting for insurance payouts, and soybean farmers dispute the settlements.

Down To Earth ↗

In India’s crop insurance, yield estimates of insurance units are the basis for indemnity payouts, and machine learning is among the model families piloted.

ICRISAT ↗

ICCO temporarily withheld 2025/26 cocoa production and grindings data and flagged an El Niño expected later in 2026.

ICCO ↗

A random-forest model driven by 12 climate models projects cocoa yield losses of about 5% in southern and coastal Ghana and Côte d’Ivoire, and up to 20% under severe drought, under high emissions.

PLOS Climate ↗

Philippine statistics office has begun piloting satellite and AI crop estimates

GMA News Online ↗

The panel on this signal

They agreed on

–Official harvest figures are moving from surveys and hand crop-cuts to satellite models, led by grains and staples.

–USDA’s AI and machine learning work is still a pilot.

–Lost trust, not new technology, pushed USDA to act after the 2025 corn revision.

The remote-sensing agronomist disagreed on

–Satellites see canopy, not grain: without field crop-cuts to calibrate against, a model drifts exactly in the unusual years that matter.

Know more about the method →

✦How this signal was built

01 · Panel

A Delphi panel of 4 AI analysts, each reasoning independently.

→

02 · Debate

They challenge and revise each other’s reads over two rounds.

→

03 · Evidence

Every claim is tied to a cited, linked source on The Wire.

→

04 · Synthesis

A moderator writes it into one read, disagreement and all.