Spatio-temporal characterization in satellite image time series

Last years have witnessed an increased interest in the analysis of evolution of spatio-temporal structures in large data volumes. The current satellite remote sensing missions allow the recording of long Satellite Image Time Series (SITS) with passive and active sensors. However, the spatio-temporal characteristics of SITS imply different approaches. This paper aims at presenting an overview of these issues, and also some recent results based on Gibbs Markov Random Fields models that are used to describe the spatio-temporal patterns. In addition, in order to obtain an automatic analysis of these patterns, the problem of determining the optimal number of spatio-temporal clusters is also discussed. The experiments are carried on Landsat 7 multi-temporal and multi-spectral images and on Envisat ASAR images, both at 30 meters spatial resolution.