Mapping Irrigated Agriculture in Fragmented Landscapes of Sub-Saharan Africa: An Examination of Algorithm and Composite Length Effectiveness

May 15, 2023·
T. Weitkamp
,
G.J. Veldwisch
,
P. Karimi
,
C. De Fraiture
· 0 min read
Abstract
Which machine learning algorithm and which composite length work best for mapping smallholder irrigation? We compared random forest, support vector machines, neural networks and k-nearest neighbours, each with 1x12-, 2x6-, 4x3- and 6x2-month composites, in study areas in Mozambique. No single algorithm was best: neural networks, SVM and random forest all did well. Shorter composites suited complex, mixed landscapes, longer composites were enough for more uniform ones. Combining the models into agreement maps showed core areas of irrigated agriculture with high certainty, surrounded by a zone where the models disagree.
Type
Publication
International Journal of Applied Earth Observation and Geoinformation