Nowadays, satellite technologies provide huge amounts of remote sensing images. Such sources of information describe geographical areas through time, producing time series of satellite images.A remote sensing time series usually contains spatio-temporal phenomena that are difficult to understand and monitor due to their intrinsic complexity. In this work, we pro- pose a new clustering framework to mine time series of satellite images. Our proposal firstly detects spatio-temporal entities, secondly it characterises their evolutions by a graph-based representation and finally it produces clusters of spatio-temporal entities sharing similar evolu- tion behaviours. Our approach is original in that it works at object-level (image segments) and not pixel-level as is usually the case in the remote sensing field.We experimentally validate our framework on a real world time series of satellite images w.r.t standard techniques employed in remote sensing analysis. We also highlight how the obtained results can be easily interpreted by domain experts.