What happened
Since 2018, paying agencies of the EU's common agricultural policy may check aid claims by monitoring with Sentinel satellite data. On 28 January 2020, the European Court of Auditors described the method of the paying agency of Belgium (Flanders) in a special report. An artificial intelligence algorithm, trained on series of Sentinel 1 and 2 images and on farmers' declarations, predicts each parcel's class. Parcels whose predicted class does not match the declaration get a field visit. The agency reported that it monitored 397,568 parcels in 2019 and inspected 1.0% of them in the field. In its annual report for 2020, published on 26 October 2021, the Court reviewed checks by monitoring at two paying agencies. They covered the basic payment scheme in Flanders and coupled support for tomato farmers in Malta. The Court found that both agencies had used the checks to prevent a number of overpayments. It gives no count or amount. Malta's system compares crop growth markers with an expected curve. The Court does not describe it as trained on data.
Sources
-
Audit report, by authors with no tie to the system.
European Court of Auditors. 2020 annual reports on the implementation of the EU budget for the 2020 financial year, Chapter 6 'Natural resources', paragraph 6.20 and Box 6.3. Published 26 October 2021.
During our 2020 Statement of Assurance work, we reviewed two paying agencies' use of checks by monitoring, focusing on the voluntary coupled support scheme for tomato farmers in Malta and the basic payment scheme in Belgium (Flanders). We found that these paying agencies had used checks by monitoring to prevent a number of overpayments.
-
Audit report, by authors with no tie to the system. Context or contrary evidence.
European Court of Auditors. Special report 04/2020: Using new imaging technologies to monitor the Common Agricultural Policy: steady progress overall, but slower for climate and environment monitoring. Box 1 and Table 3. Published 28 January 2020.
An artificial intelligence algorithm, trained using a time series of Sentinel 1 and 2 images (in combination with the information contained in the farmers’ declarations), predicts the probability that each parcel belongs to one of five classes (arable land, grassland, leguminous crops, fallow and non-eligible).
Details
AI system
- Workflow
-
An algorithm trained on Sentinel 1 and 2 image series and farmers' declarations predicted whether each parcel was arable land, grassland, leguminous crops, fallow or non-eligible. Parcels that did not match the declaration got a field visit. Inconclusive parcels were checked on screen first and visited if needed.
Follow-up
- Conditions for replication
-
Every eligibility criterion of the scheme must be checkable from satellite images, since other criteria still need on-the-spot checks of a sample of farmers. Small parcels need other data or follow-up, since Sentinel images give too few pixels for them.
Entry details
- Sector
-
Agriculture (control of EU farm aid claims).
- Purpose
-
The paying agency of Belgium (Flanders) built its checks by monitoring to verify from Sentinel satellite images whether each declared parcel is used as the farmer declared.
31 more questions not yet checked
model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). Checks by monitoring help Flanders prevent overpayments of farm aid (AIBD-1034). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1034.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)