Operational Sentinel-2 Based System for Near Real-Time Irrigated Area Monitoring in the Limpopo River Basin

Monitoring irrigated agriculture is critical in water-scarce regions like the Limpopo River Basin (LRB), where irrigation consumes significant freshwater. This study presents a scalable, semi-supervised machine-learning framework for monthly mapping of irrigated croplands and water use in the LRB from 2019 to 2024. The framework combines Sentinel-2 imagery, Random Forest classification, time-lagged precipitation–vegetation analysis, and slope masking for monthly irrigation mapping. FAO’s WaPOR evapotranspiration data was used to estimate water use. At 10 m resolution, the framework achieves 80% accuracy (κ = 0.60), capturing smallholder plots and seasonal dynamics. Dry-season irrigated area declined from ~ 211,281 ha in 2019 to ~ 184,771 ha in 2024, while water use rose from ~ 103 to ~ 134 million m³, indicating intensifying demand. Irrigation concentrates in key sub-basins, with potential for sustainable expansion if water is available. The methodology highlights the effectiveness of combining high-resolution imagery with time-lagged analysis for accurate irrigation mapping. Its open-access implementation via Google Earth Engine offers a replicable model for water-stressed transboundary basins, enhancing resource management and resilience in climatically vulnerable regions.