A Comparative Analysis of Random Forest and Support Vector Machines for Classifying Irrigated Cropping Areas in The Upper-Comoé Basin, Burkina Faso

Objectives: This study investigates the performance of two machine-learning algorithms in classifying land areas across the Upper-Comoé basin in Burkina Faso. Methods: Within the Google Earth Engine data processing environment, Support Vector Machine (SVM) and the Random Forest (RF) algorithms were applied to a Landsat-8 OLI image of March 2019, to discriminate agricultural land areas, with an emphasis on irrigated areas. Findings: The results indicated good to excellent classification performance, with overall accuracies and Kappa coefficients between 71% and 99%, and 0.66 and 0.99, respectively. The RF method outperformed the SVM in terms of mapping "accuracy", but in terms of spatial distribution of classes, the SVM method provided a mapping close to reality, due to the density of the classes generated. Novelty: Our findings suggest that remote sensing can constitute a tool fully adapted to the needs of services in charge of agricultural water management in Burkina Faso.