Comparison of Google Earth Engine Machine Learning Algorithms for Mapping Smallholder Irrigated Areas in a Mountainous Watershed, Upper Blue Nile Basin, Ethiopia

Irrigated area mapping with improved accuracy is essential to enhance the monitoring and management of the limited water resources of a country. This study compared machine learning algorithms, classification and regression trees (CART), gradient tree boost (GTB), random forest (RF), and support vector machine (SVM) on the Google Earth Engine platform. The study used Sentinel-2 multispectral and Sentinel-1 synthetic aperture radar satellite data to classify smallholder irrigated areas during the 2021/22 irrigation season. The methods' accuracy was evaluated relative to the inputs and agroecology. The accuracy has been improved by incorporating monthly SAR and vegetation indices data. RF has been found to be the consistent classifier in different agroecological zones and inputs with overall accuracy (OA) of 0.89, followed by SVM and GTB with OA 0.88 and 0.87 at the watershed level respectively. The lowest OA (0.86) and Kappa coefficient (0.82) values were obtained using the CART algorithm. The normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI), normalized difference red edge (NDRE), and normalized difference water index (NDWI) were found to be important in mapping irrigated areas.