Machine learning implementation for mapping irrigated areas at fine temporal and spatial resolutions in the Limpopo River Basin

Most of the global available freshwater for food production is utilized for irrigation. Irrigation expansion is crucial for agriculture production as it can increase crop yields and be a dependable adaptation measure against climate change. Accurate information on the spatial extent of irrigated areas and their dynamic shifts is therefore essential for efficiently managing already pressured water resources. The multiplicity of remotely sensed data sources and state-of-the-art machine techniques offer new avenues for producing more accurate irrigation maps. This study presents the results from a monthly monitoring framework for fine-scale mapping of irrigated areas in the Limpopo River Basin. The proposed framework uses high to moderate-resolution earth observation data, the extra-tree classifier, and a series of land cover masks in differentiating rain-fed and irrigated areas. We found that the area of irrigated land during the dry season in 2021 varied from 356589 ha to 612738 ha between and September. The overall accuracy of classified maps varied from 98 to 100%.  The proposed framework offers an automatic and replicable cost-effective means of mapping irrigated areas using Google Earth engine, multisource data, and machine learning algorithms.