SATIKAS — Estonia's Deep-Learning Satellite System Monitors 100% of Grasslands for CAP Mowing Compliance

Since 2017 Estonia's Agricultural Registers and Information Board (ARIB, now part of PRIA) has used SATIKAS, a satellite-based mowing-detection system built with the University of Tartu's Tartu Observatory and its spin-off KappaZeta, to check compliance with the EU Common Agricultural Policy's grassland-mowing rule. It feeds Sentinel-1 radar and Sentinel-2 optical time series into a deep-learning classifier, replacing on-site inspections with continuous remote monitoring of Estonia's roughly 101,000 registered grassland parcels.

A peer-reviewed evaluation (Preidane et al., Scientific Reports, 2022) tested the method on 2,000 parcels through the 2018 growing season, finding 94.8% end-of-season accuracy and an AUC of 0.97; flagging the 12.8% most ambiguous cases for manual review raised overall accuracy to 92.4%. The same study reports successful trial extensions of the method to Sweden, Denmark and Poland.

The reject-region approach defers about one case in eight to human review, and the published accuracy figures come from a single growing season, so year-to-year robustness under varying weather conditions is not yet independently documented at the same depth.