Mapping cropland in smallholder farmer systems in South-Africa using Sentinel-2 imagery

Details

Supervisors

Defourny , Pierre

Faculty

Faculté des bioingénieurs

Degree label

Master [120] : bioingénieur en sciences et technologies de l'environnement

Abstract

This thesis investigated three different classification methods for mapping small scale farming systems in the north-west of the Mpumalanga province in South Africa. With the Sen2Agri’s RF classifier being used as baseline method, different sampling strategies and availability of time series were evaluated on the classifiers performance. The adequateness of the 10 m resolution of Sentinel-2 was also evaluated through a size stratified error analysis. In parallel two other classification methods were used, one using the SVM classifier and the other one using an NDVI threshold method. While the results presented in this thesis aren’t concluding for evaluating the general performance of Sen2Agri for mapping smallholders systems on a larger scale, they do set a baseline of comparison for future studies.