Updating Scarce High Resolution Images with Time Series of Coarser Images : a Bayesian Data Fusion Solution

As a consequence of the great variability between sensors, the characteristics of remotely sensed data widely differ with respect to spectral and spatial resolutions. Additionally to their respective technical characteristics and peculiarities, sensors also have different temporal frequencies of acquisition. Coarser sensors (e.g. SPOT VEGETATION or TERRA MODIS) have generally close to daily acquisition rates while high spatial resolution sensors (e.g. SPOT HRVIR or IKONOS) have lower acquisition rates. Cloud-free high resolution imagery may therefore not be available at the required period unlike coarser resolution images. On top of this, high resolution images are sometimes so highly priced that updating past high resolution images with recent coarse images can be cost effective. For these reasons, there is a real need for a sound theoretical framework that aims at merging information coming from two or more different sensors while taking explicitly into account the spatial resolution discrepancies between images. Typically, for cost effective applications, this could involve predicting a high resolution image by updating a past one with more recent but coarser images. It is a common fact that remote sensors have different spatial resolution. This change of resolution is thus a typical issue in remote sensing applications. Depending on users’ needs and the heterogenity of the study areas, different algorithms of fusion were proposed for the spatial enhancement of remotely sensed images. These include Brovey method (Pohl & van Genderen, 1998), Intensity-Hue-Saturation (IHS; Harrison & Jupp, 1990), Principal Component Analysis (PCA; Pohl & van Genderen, 1998), wavelet-based Multi-Resolution Analyses (MRA; Zhou et al., 1998; Garzelli & Nencini, 2005; Ranchin et al., 2003), High-Pass Filter (HPF; Chavez et al., 1991)