Sown area, measured from orbit.

Challenge

Knowing how much land is actually planted, and with what, is a question governments, insurers, traders and agribusinesses spend heavily to answer, usually through field surveys and self-reported declarations that arrive late and disagree with each other. Satellites see every field every few days, but turning raw multispectral imagery into a trustworthy hectare count is the hard part: crops look different by region, season and growth stage, and clouds, water and bare soil all masquerade as noise.

The question this project set out to answer: could open satellite data alone reproduce official sown-area figures closely enough to be operationally useful?

Solution

Antenor built a supervised classification pipeline on Sentinel-2 L2A imagery that identifies crops and quantifies the area planted, region by region. The classifier was trained and evaluated across three distinct regions, then validated not against a held-out slice of its own data but against an external, authoritative benchmark: the sown-hectare figures reported by Argentina’s national agriculture secretariat. The same approach extends to other land categories, mapping water bodies and flood zones by adding labeled polygons.

The work was developed as applied research within the University of Buenos Aires Data Science master’s program and is peer-visible on ResearchGate.

Stack

• Sentinel-2 L2A multispectral satellite imagery (open data)
• Supervised classification trained across three geographic regions
• Geographic information system polygons for crop, water-body and flood-zone labeling
• External validation against official government sown-area statistics

Results

• Over 90% accuracy in planted-area estimation, measured against the sown hectares officially reported by the national agriculture secretariat
• Built entirely on open Sentinel-2 imagery, with no proprietary sensor data
• Extends beyond crops to water-body and flood-zone mapping, using the same pipeline
• Demonstrates the remote-sensing and computer-vision capability Antenor brings to agriculture, environment and land-use problems: measurable accuracy, validated against real-world ground truth