Breaking

Google DeepMind satellite models map farm fields and crops on a 15-day cycle

September 6, 2026 at 02:29 EDT

Foundation Models
Sales & Business

Google DeepMind has built two satellite-based AI models that map individual agricultural fields and monitor crops in near real time, refreshing in-season estimates about every 15 days rather than waiting until after harvest.

September 2026 · Google DeepMind

Satellite AI maps farms in near real time

AnthroKrishi’s ALU and AMED models turn satellite imagery into field boundaries and in-season crop monitoring, with AMED refreshing about every 15 days while a season is still underway.

15 days

AMED in-season refresh

15 yrs

ALU landscape history

140M ha

Terrastack coverage

Supported crop classes

Update cadence

In-season refreshes vs a year-later report

A 15-day cycle yields about 24 in-season AMED updates a year, versus a single post-season statistical report.

AMED / year

~24 updates

Post-season

1 annual report

Takeaway: AMED is built for decisions during the growing season, not a year-later census.

How the stack works

1 · Imagery

Public satellite scenes, including high-frequency and high-resolution captures

→

2 · ALU

Segment fields, trees, and water; assign Plus Codes and area

→

3 · AMED

Crop class, sowing, and harvest windows every 15 days

→

4 · Delivery

Partner APIs and a Google Earth landscape layer

Where it is used

Partners overlay ALU and AMED on land records, irrigation, and carbon programs — including more than 140 million hectares in India and rice monitoring across 12 countries.

Where it stops

The stack reconstructs past seasons; it does not forecast yields. Only 12 crops are supported, and clouds or missing imagery can return no prediction.

Looking backward, with a confidence score attached

Talekar has described the models as reconstructing what grew, not predicting the next harvest. Each feature ships a 0–1 confidence value so downstream tools can filter weak calls.

ALU high confidence

> 0.9

AMED high confidence

> 0.7

Telangana field check

82% accurate

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.