Analytical framework for modeling scale-related variabilities in remote sensing

A general analytical framework was established to investigate the scale-related variabilities in remote sensing. The variabilities were studied first by investigating canopy structure, canopy interaction with light, relation between spectral reflectance and plant phenological parameters. The variabilities simulated by the plant model were compared with the actual spectral data acquired by ground spectroradiometer and satellite sensors. The theoretical relation between orthogonal-based transform and Kahunen-Loeve transform was investigated in the vector space. The role of spectral indices in identifying the status of phenological parameters was briefly studied.; The radiometric corrections of the remotely sensed data were carefully controlled to avoid the unwanted noise introduced by typical resampling/correction procedures from commercial operation. The non-linearity and sensor response corrections were applied to the spectral data as necessary. Variability analysis was conducted to illustrate the complexities of spectral variability embedded in the remotely sensed data.; The information extraction in spatial frequency domain was investigated with emphasis in Fourier domain feature extraction. The Radon transform was introduced as the potential tool to enhance the spatial information of the Fourier transformed image. The adequacy of entropy and fractal dimension as image information measures was proved. A functional link between entropy and fractal dimension was established. The image information content was extracted using various first and second order statistics, entropy, and fractal dimension. Results were presented for different remote sensors based on the full image information content and specific agricultural ground features. The quality of spatial resampling algorithms was tested by investigating the capability to maintain image information in the resampled image. Finally, two applications utilizing this analytical framework were presented to show its potential in land-use classification and multiscale data fusion.