Temporal sequences of images called Satellite Image Time Series (SITS) allow land cover monitoring and classification by affording a large amount of images. Many approaches attempt to exploit this multi-temporal data in order to extract relevant information such as classification-based techniques. In this paper we compare low and high levels classification-based approaches that aim to reveal the SITS pixels or regions spatio-temporal evolutions through a temporal map. The first approach is a graph region-based classification approach that proved its performances for Very High Resolution (VHR) optical SITS analysis. The second one is a change detection pixel-based approach that has been successfully applied for SAR SITS multi-temporal classification. The experimental results have been conducted on both synthesized and real data in order to compare those approaches and conclude about their accuracies.