Field crop mapping using machine learning and multi-sensor satellite fusion: toward dynamic agricultural monitoring

Monitoring crop dynamics with precision is vital for food security and sustainable agricultural management. Yet, conventional monitoring approaches often lack sufficient spatial resolution, revisit frequency, and scalability, particularly in heterogeneous and fragmented agricultural landscapes. This study introduces a novel hierarchical remote sensing framework that integrates Sentinel-1 SAR and Sentinel-2 multispectral data with machine learning to generate high-resolution, multi-season crop type maps across extensive agricultural regions. Focusing on the western Negev in Israel (2018–2024), we developed a three-tier Random Forest classification workflow specifically designed for field crop mapping. The workflow integrates multi-temporal spectral, phenological, and SAR-derived features in a stepwise approach: (1) agricultural land cover types (94% overall accuracy), (2) wheat identification (95% accuracy), and (3) classification of 13 field crop types (81% accuracy), including key crops such as wheat, corn, and cotton. The hierarchical structure improved classification precision and enabled robust generalization across years, facilitating tracking of crop rotation, land-use intensity, and field-level management practices. Importantly, the model captured climate-driven phenological gradients in wheat, revealing spatial variability in crop development patterns that conventional methods often overlook. The novelty of this study lies in its scalable, multi-sensor classification framework that fuses radar, optical, and phenological information to support operational, dynamic crop monitoring. By coupling remote sensing outputs with national GIS infrastructure, this approach offers a cost-effective, transferable solution to advance precision agriculture, promote climate adaptation, and guide sustainable land-use planning at regional and national scales.