A SAR-Driven Approach for Winter Catch Crop Classification in Germany

Growing winter catch crops is a key agricultural practice offering a number of advantages, particularly for soil fertility, soil health, nitrogen leaching and soil protection. The growing phase of these crops, spanning autumn and winter, coincides with dense cloud cover in Germany, leading to data gaps in optical remote sensing that hinder effective monitoring. To address this limitation, this study assesses the potential of Sentinel-1 SAR time series to identify and classify winter catch crops and their specific types. Here, we propose a two level hierarchical Random Forest (RF) classification framework and evaluate its spatio-temporal transferability. In level-1, the model distinguished between catch crops and non-catch crops using phenological contrasts and sowing window differences observed in the time series derived from radar parameters, including VV, VH, VH/VV, and dpRVI. In Level-2, the model classified six specific catch crop types using 22 defined crop specific descriptive features. The RF model, trained on 2021 data from Lower Saxony (LS), achieved overall accuracies of 94.9% in level-1 and 85.7% in level-2. The model maintained strong temporal transferability when applied to LS data from 2022 and 2023. Moreover, spatio-temporal transfer tests demonstrated consistently good performance in Brandenburg (BB), while North Rhine–Westphalia (NRW) showed satisfactory results with some misclassifications in non-catch crops likely due to phenological shifts. The results highlight the potential of SAR-based frameworks for identifying catch crop types and their applicability for scaling to other federal states of Germany and national-level mapping efforts.