Characterization of land cover by multi-temporal biophysical variables in fused images

Nowadays, it is very common to have readily available remotely-sensed spatial information, at different resolutions, thanks to the different satellite sensors that are acquiring multispectral images at both low and high resolutions. Fusion techniques have then arisen as an alternative to integrate this information, which result in new images that contain better spectral and spatial information in terms of contents and resolution. Several fusion techniques based on the Wavelet transformation have been developed, in which the "a trous" algorithm stands out as one of the most important tool that is able to preserve spectral and spatial properties. As an alternative, we have introduced an algorithm based on an undecimated Hermite transform (HT) that preserves these properties, with. better image quality. In this paper, fused images are analyzed in the framework of biophysical-variables such as leaf-area-index and sparse-fractional-vegetation-cover, all of them derived from reflectance values in the visible-red and near-infrared bands, from multi-temporal SPOT-5 images [2005-2007]. Multi-temporal analyses are conducted to test the consistency of these variables for different illumination conditions, and vegetation amount, in order to determine indicators of land-cover-change. Results were used to characterize a change vector analysis, by differentiating land transformation from modifications based on the results with fused and original images. Results also showed how the HT algorithm resulted in the smallest modification of the bi-dimensional space of the vegetation and soil isolines after fusion. This method also preserves the information integrity necessitated to obtain similar biophysical variable values. By improving spatial resolution, while preserving spectral characteristics of the resulting images, the HT-based algorithm is able to better characterize land-cover-change.