Objectives of the service
Paying Agencies across the EU face growing pressure to automate Common Agricultural Policy (CAP) controls. Manual field inspections are costly, slow and do not scale. CropCloud addresses this by delivering automated field boundary delineation and agricultural event marker detection from Copernicus Sentinel-1 and Sentinel-2 imagery. The service uses deep learning combined with proprietary super-resolution (enhancing 10 m imagery to approximately 1 m) to produce parcel-level intelligence accessible through scalable APIs and web map services (WMTS). The CCN2 extension added eight operational event markers – including seeding, mowing, harvest, bare soil/tillage, homogeneity, similarity, soil erosion zones and ecological indicators – aligned to the Lithuanian National Paying Agency (NMA) Area Monitoring System (AMS) requirements. The project demonstrated that satellite-derived event markers can replace a significant share of manual field inspections while meeting the NMA’s 95%+ accuracy target.
Users and their needs
The primary pilot user is the Lithuanian National Paying Agency (NMA), responsible for administering CAP payments across Lithuania. NMA operates an Area Monitoring System (AMS) that relies on satellite imagery to verify farmer declarations and detect agricultural events such as seeding, mowing and harvest. NMA needs: (1) automated detection of key agricultural activities to reduce manual inspector workload; (2) high accuracy (target 95%+) to maintain payment control integrity; (3) coverage of 2.8 million hectares of delineated field boundaries; (4) integration with existing GIS/IACS workflows via standard web services. Beyond NMA, the service targets EU Paying Agencies broadly and enterprise agri-platforms (B2B). DigiFarm currently serves more than 55 clients across 21+ countries. The main challenge is geographic variability in agricultural landscapes, requiring substantial region-specific training data and model adaptation. Targeted countries: Lithuania (pilot), with commercial deployments across the EU, North America and other regions.
Service/ system concept
CropCloud ingests Copernicus Sentinel-1 (SAR) and Sentinel-2 (optical) imagery and applies proprietary super-resolution to enhance spatial detail to approximately 1 m. Deep learning models then perform two core tasks: (1) automated field boundary delineation, producing vector polygons of agricultural parcels; and (2) event marker detection, classifying eight agricultural activities (seeding, mowing, harvest/greening, homogeneity, similarity/Euclidean distance, bare soil/tillage, soil erosion zones and non-pesticide/ecological indicators) per parcel and time step. Results are delivered through RESTful APIs returning GeoJSON geometries with classification metadata, a Web Map Tile Service (WMTS) for GIS integration (tested in QGIS), and a web dashboard for visual exploration and quality assurance. A traffic-light system flags parcels for inspector follow-up. The processing pipeline runs on GPU-accelerated cloud infrastructure with automated scheduling, achieving API latency below 300 ms and service uptime above 99.9%. A usage and billing module tracks consumption by product layer and version.
Space Added Value
CropCloud relies on Copernicus Sentinel-2 multispectral imagery (13 bands, 10–60 m resolution, 5-day revisit) as its primary Earth Observation input, complemented by Sentinel-1 SAR data for cloud-independent monitoring. DigiFarm’s proprietary super-resolution processing enhances the native 10 m optical imagery to approximately 1 m effective resolution, enabling parcel-scale interpretation that would otherwise require commercial very-high-resolution satellites at significantly higher cost. The dense temporal revisit of Sentinel-2 is essential for detecting time-sensitive agricultural events such as seeding, mowing and harvest – activities that can occur within days and would be missed by less frequent observation. Combining optical and SAR data improves robustness in cloud-prone regions. Compared to current methods – which rely heavily on manual field inspections or aerial photography campaigns – the satellite-based approach offers wall-to-wall national coverage at a fraction of the cost, with automated and repeatable analysis. No competing
service combines Sentinel super-resolution with deep-learning event marker detection at this operational scale. The Harmonised Landsat Sentinel (HLS) dataset is proposed as an additional input to further increase temporal density.
Current Status
The project is completed. All eight CCN2 event markers were developed, integrated and validated against the Lithuanian NMA’s ground-truth data, achieving the 95%+ accuracy target across all markers. A three-month pilot (July–September 2024) demonstrated operational readiness: API polygon delivery latency averaged below 239 ms (target 300 ms), boundary delineation IoU reached
0.96, and service uptime remained above 99.99%. Factory Acceptance Testing (FAT) and Site Acceptance Testing (SAT) were completed on 11 December 2024, with live WMTS integration verified in the NMA’s QGIS environment. DigiFarm delivered 2.8 million hectares of delineated field boundaries to NMA. The service reached Service Readiness Level (SRL) 7. Commercially, DigiFarm reports more than 55 active clients, over €2M in annual Paying Agency contracts, and deployments across 21+ countries.
Prime Contractor(s)
