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

Accurate mapping of smallholder irrigated agriculture in sub-Saharan Africa using remote sensing is challenging due to the diverse and fragmented nature of these farming systems. This study evaluates how training data size and composition affect the accuracy of mapping irrigated agriculture in Mozambique using machine learning algorithms. We found that larger training datasets generally improve classification accuracy, but with diminishing returns beyond a certain size. Models trained in different regions showed varying generalization capabilities, with some regions producing more robust models than others. Our findings provide insights for optimizing field data collection strategies in similar agricultural landscapes.

Journal Article

Remote Sensing