Crop classification

Netherlands Enterprise Agency

Detecting a crop on a defined geometry.

Last change on 18th of July 2025, at 11:54 (CET) | Publication Standard 1.0

• Publication category: Impactful algorithms
• Impact assessment: DPIA
• Status: In use

General information

Theme

Nature and Environment

Begin date

2023-01

Contact information

https://www.rvo.nl/onderwerpen/contact/formulier

Link to publication website

https://www.rvo.nl/onderwerpen/glb-2024/ams

Responsible use

Goal and impact

The algorithm helps make the annual CAP grant application better. It checks whether the applicant's information is correct. If something is incorrect, the applicant is notified.

Considerations

Since 2023, the Area Monitoring System (AMS) has been a mandatory component for monitoring

of land-based schemes, as stipulated in Article 7 of Implementing Regulation (EU)

2022/1173. Automated monitoring using satellite data is the most

cost-effective way. The relation itself is also able to provide evidence and

ultimately, it can also object.

Human intervention

If the algorithm finds a discrepancy, an employee first looks at the result. If the clerk agrees with the deviation, he will let the requester know. The applicant can then send evidence via a special app (the geotag photo app) to show that it might be right after all.

Risk management

The quality of the algorithm is tested annually. This is done on the basis of samples.

And a mandatory quality test that has to be reported to the European

Commission.

Legal basis

Since 2023, the Area Monitoring System (AMS) has been a mandatory component for monitoring

of land-based schemes, as stipulated in Article 7 of Implementing Regulation (EU)

2022/1173.

Links to legal bases

UITVOERINGSVERORDENING (EU) 2024/2202 VAN DE COMMISSIE van 4 september 2024: UITVOERINGSVERORDENING (EU) 2024/2202 VAN DE COMMISSIE van 4 september 2024

Impact assessment

Data Protection Impact Assessment (DPIA)

Operations

Data

Functional_id, specified crop code, geometry, application number (specify data of the

relation for a CAP grant application, Satellite data).

Links to data sources

• BRP Gewaspercelen : https://www.pdok.nl/-/brp-gewaspercelen
• Copernicus Data Space Ecosystem : https://dataspace.copernicus.eu/

Technical design

The crop recognition algorithm aims to confirm farmers' crop declarations. For this purpose, it uses a Random Forest model that classifies crops based on different (interpolated) time series of satellite signals. The model operates at the level of geometries (Features of Interest, FOI), not at the pixel level. Training is based on farmer statements, with outliers removed first.

The model uses crop groups (superclasses), which can be further split later in the season (early versus late groups). To accommodate regional variations (such as climate and soil type), the model is trained by agronomic zone.

All FOIs with an explanation in a zone are used as training data, except FOIs marked as outliers. Outliers are identified with a simple Random Forest model that looks at how often FOIs with the same crop type end up in the same leaves of the model. Only crop groups with at least 30 statements per zone are included to avoid bias. Although the model can deal with missing data, reliability decreases with too few inputs. Therefore, only FOIs with sufficient data (above a certain threshold) are included in training and validation.

This is not a self-learning algorithm.

External provider

NEO BV

Link to code base

N.V.T.