The demonstrated workflow is primarily based on a further analysis of the crop type classification based on both Sentinel-1 and Sentinel-2 satellite imagery providing spectral and temporal information about crop characteristics and development. The original crop type nomenclature provided by the Czech Paying Agency (SZIF) including 515 individual crops was aggregated into more general categories based on biological relationships and similarities in temporal (NDVI profiles) and spectral behaviour. The crop classification is performed across three different levels of detail, from general categories (Level-1) to more detailed classifications (Level-3). The classification accuracy is measured using standard metrics like F-1 scores and overall accuracy. The results show that the classification is highly accurate at Level-1 (ca. 95 % overall accuracy) and Level-2 (ca. 93 % overall accuracy), but the accuracy diminishes at the more detailed Level-3 (falling to ca. 80 %). A probability map (indicating “how sure the classifier is in assignment of the given parcel into the target crop type class”) is generated on top of the thematic crop type layers. Detection of catch crop occurrence is finally done through analysis of NDVI profiles (identification of the secondary growing season peak and analysis of its relative significance compared to the primary season) together with crop type classification.

The crop rotation rules are based on three main conditions including: 1) the main crop cultivated at the given parcel up to 31.8. of the actual year must differ from the main crop cultivated in the previous year, 2) the same main crops can be cultivated at the given parcel only if a catch crop is grown between the two growing cycles of the same main crop, and 3) the same main crop cannot be cultivated at the given parcel for more than 4 consecutive years (even with considering catch crops). Thus, the analyses compares the results of crop classifications from two consecutive years trying to identify parcels with the different crops/groups of crops at the most general Level-1. The parcels where the different crops cannot be confirmed proceed to the second iteration of the analysis applied at the Level-2 detail. Similarly as in the previous steps, the parcels where different crop types are identified in the two compared years are considered as compliant, whereas the others proceed into the third iteration of the analysis done on the most detailed Level-3 crop type classification. The parcels where different crops are not proven even at the most detailed Level-3 are then checked for the catch crop occurrence considering those with catch crop presence as compliant. The rest of the parcels (i.e. same crops identified in both of the compared years but with no catch crop presence) are then analysed in terms of the crop type classification probability removing the cases where the combined probability (i.e. probability that the result of crop classification is correct in both of the compared years) is less than 85 %. The remaining parcels are finally flagged as non-compliant. As a result, 4 % of the agricultural parcels in Czechia seem to be non-compliant with the crop rotation rules applied in the two years 2022/2023. This value further drops to 1 % if the longer period (2021 – 2023) is taken into account.

Crop diversification is analysed by calculating the number of different crops grown on each farm and their relative shares. Farms are categorised based on their total arable land, and different diversification rules apply depending on farm size. For instance, no crop diversification rules are applied on the farms with less than 10 ha of arable land. At least two different crops (with maximal 75% share of the most frequent crop) are required in case of the farms with 10 – 30 ha of arable land. Three different crops with maximal share of 75 % for the most frequent crop and 95% share of the two most frequent crops are required in case of farms with 30 – 150 ha of arable land. Finally, if the given farm manages more than 150 ha of arable land, then at least four different crops are requested for fulfilling the crop diversification rules. In such a case, the most frequent crop can have up to 75% share and the three most frequent crops have not exceed 95% share. The results of crop classification are first aggregated from parcel to farm level calculating the number of cultivated crops and their relative share. The decision of compliance/non-compliance of the given farm is then based on application of the appropriate rules corresponding to the given farm category. As a result, 56 % of the analysed farms were found as compliant with the crop diversification rules in 2022/2023. Another 38 % of farms manage no arable land and the crop diversification analysis is not relevant in such a case. 6 % of the farms seem to be non-compliant with the rules, whereas no information on cultivated crops were available for another 0.5 % of the farms.

The last part of the use case focuses on meeting the rules established for erosion-prone areas. Each parcel is first categorised into one erosion risk category including: 1) no erosion risk (NEO), 2) medium erosion risk (MEO) and 3) high erosion risk (SEO). A list of the non-allowed crops (“forbidden crops”) is defined for each level of the erosion risk. Parcels with erosion risk are cross-referenced with a list of restricted crops to ensure compliance with anti-erosion measures. Non-compliance is flagged when prohibited crops are detected on erosion-prone land. As a result, 6 % of the analysed parcels seems to be non-compliant with the “anti-erosion rules”. However, this value drops to 3 % if only the cases with high probability results of the crop type classification are taken into account.