Accurate crop classification is essential for global food security under climate uncertainty. This systematic review, following PRISMA 2020 guidelines, analyzes 111 peer-reviewed studies (2015–2025) on Sentinel-based crop mapping using artificial intelligence. Our meta-analysis reveals a clear methodological evolution: traditional machine learning achieved ~85% accuracy, deep learning pushed performance to 87–95%, while hybrid CNN-Transformer architectures attain 95–98% an 11–13% improvement over baselines. Multi-sensor fusion of Sentinel-1 SAR and Sentinel-2 optical data consistently yields 5–8% accuracy gains, particularly valuable in cloud-prone tropical regions. However, research concentrates on temperate systems in Asia and Europe, leaving Sub-Saharan Africa and South America critically understudied. We demonstrate that contextual factors dataset size, crop complexity, and validation protocols influence accuracies as substantially as algorithm choice, demanding standardized benchmarks. Future priorities include transfer learning and foundation models to address data scarcity, expanding coverage to tropical smallholder systems, and adopting uncertainty quantification for scalable global crop monitoring.
Benzhair et al. (Tue,) studied this question.