Classification of scene evolution patterns from Satellite Image Time Series based on spectro-temporal signatures

Huge quantities of medium resolution satellite images are available from various Earth observation sites. These archives enable the creation of long-term, medium resolution Satellite Image Time Series (SITS). Such SITS are large, complex data sets, embedding spatial, spectral and temporal information. The development of effective methodologies for analysis of SITS is a challenging issue. In this paper we propose a new method to analyze medium spatial resolution SITS at pixel level by means of multi-sensor spectro-temporal signatures. The temporal resolution of SITS is improved by combining scenes captured with different remote sensing sensors. The proposed method is applied to a series of 40 Landsat TM and ETM+ scenes covering the area of Bucharest, Romania. Spectro-temporal signatures are extracted by means of the first two tasseled cap features and a multiclass SVM classification was used to create maps that show land use and transitions in land use over time.