This thesis proposes a system for monitoring farmer compliance with spatial regulatory requirements specified in the European Union’s Common Agricultural Policy (CAP). An application for combining satellite data from different sources is created to identify areas of potential non-compliance. Selected imagery from this application is then analysed using remote sensing and machine learning methods.

First, the spatial requirements contained within the CAP are reviewed and then followed by an examination of relevant methods for monitoring these requirements. The thesis investigates publicly available data portals for managing satellite imagery and metadata from the Landsat, Copernicus and ASTER programmes. An application to combine data from these sources and monitor agricultural sites is developed and discussed. The application is hosted online using cloud services and is available at https://monitor.trig.dk

The thesis then considers methods for extending the application to monitor a subset of the regulations specified in the CAP. Two approaches are tested; (1) a method based on vegetation indices and k-means clustering (2) a machine learning approach based on random forest machine learning algorithms. For all tests sentinel, 2B data is used along with field data supplied by the Danish AgriFish Agency.

The result is an online application and alert system for monitoring farmland and a two-tier workflow to determine areas of possible non-compliance by looking at statistical outliers, crop classification and heterogeneity. Conclusively, the thesis suggests the establishment of an open European-wide dataset for multi-season crop ground truth samples and minor changes to the current monitoring workflow.

The project area consists of two Danish islands: Lolland and Falster, located just south of Zealand.

Keywords: Copernicus, API, NodeJS, Agriculture, Regulation, Remote Sensing, Machine Learning, Red Edge, ESA, NASA