A robust and scalable crop mapping framework using advanced machine learning and optical and SAR imageries

Monitoring agricultural systems is increasingly essential as we address the pressing challenges of climate change, biodiversity loss, population growth, and rising food demands. High-resolution, large-scale maps of agricultural lands are fundamental for creating sustainable strategies but mapping extensive and diverse croplands over time remains complex. To tackle this, our study presents an efficient and reproducible framework for annual crop type mapping using multi-temporal satellite data and deep learning.We integrate 10-day composites from Sentinel-1 SAR and Sentinel-2 MSI data with machine learning (XGBoost, CatBoost) and deep learning models, including a Bidirectional LSTM (BiLSTM) and a Self-Attention-enhanced architecture. The approach focuses on five major crops—winter wheat, winter rapeseed, winter barley, silage maize, and sugar beet—and was applied across three German states (Lower Saxony, North Rhine-Westphalia, and Brandenburg) for 2021 and 2023.The BiLSTM model achieved the best performance among the tested approaches, with an overall accuracy of approximately 93 %. Data fusion improved classification accuracy by 2–3 % compared to single-sensor inputs. Feature importance analysis highlighted key temporal intervals related to crop phenology. To ensure consistency, we implemented linear interpolation for gap filling and tested multiple scaling techniques, with percentile-based normalization offering a good balance of simplicity and effectiveness.Our framework also demonstrated strong spatial transferability and adaptability, achieving high performance in new regions even with limited training data, and outperforming established benchmark datasets. Its integration with open platforms like Google Earth Engine enables scalable, field-level monitoring across Europe. These results support the development of robust, transferable tools for agricultural decision-making and long-term agroecosystem monitoring.