Extracting characteristics of satellite image time series with decision trees

The use of SITS improves the accuracy of the mapping of landcover. Nonetheless, SITS are complex datasets and the classification algorithm may be difficult to set up. In this article, we propose to learn an explicit model, a decision tree, from labelled time series. Our decision trees model enables to identify which time series and which time periods are the most discriminant for a classification task and thus, it provides insightful knowledge to the expert. We illustrate this method with the characterisation of agro-ecological areas of the Senegal.