Evaluating the Effect of Training Data Size and Composition on the Accuracy of Smallholder Irrigated Agriculture Mapping in Mozambique Using Remote Sensing and Machine Learning Algorithms

Mapping smallholder irrigated agriculture in sub-Saharan Africa using remote sensing techniques is challenging due to its small and scattered areas heterogenous cropping practices. A study was conducted examine the impact of sample size composition on accuracy classifying Mozambique’s Manica Gaza provinces three algorithms: random forest (RF), support vector machine (SVM), artificial neural network (ANN). Four scenarios were considered, results showed that smaller datasets can achieve high sufficient accuracies, regardless their composition. However, user producer accuracies do increase when algorithms are trained with larger datasets. The also found training data important, too few or many samples “irrigated agriculture” class decreasing overall accuracy. algorithms’ robustness depends data’s composition, RF SVM showing less decrease spread than ANN. concludes more important for classification used. suitable task as they robust sensitive outliers Overall, provides valuable insights into mapping techniques.