A new multitemporal analysis of satellite images: The residuals from from principal Component Analysis. A series of landsat Thematic mapper images from the Camargue, France. A study case

Numerous methods exist for the analysis of changes applied to time series of satellite images. After a quick review of these methods, a new approach is proposed. This approach is based on residuals computed from PCA (Principal Component Analysis) on a NDVI table. It consists of: computing the NDVI variable for each date; building a space x time table which joins NDVI variables; carrying out a PCA on this table; choosing the number, k , of factors of PCA which explain time invariant landscape structures; and computing residuals between NDVI table and the table computed from the k factors. This method applied to the analysis of a series of three Landsat TM scenes acquired in 1983, 1984 and 1993 on the Camargue region (France) allows separation of the permanent land use structure from its annual variations. The specificity of the so called 'island Camargue' is clearly shown; control of water exchanges by man alters the land use every year.