Development of spatial-temporal satellite image fusion models for complex changes of the environment

Ph.D. ; Satellite remote sensing can provide solid and indispensable data for multi-scale environmental change monitoring and analysis. Nevertheless, existing satellite sensors or retrieved products compromise between spatial resolution and temporal resolution/spatial coverage. Under the growing demand for high spatial resolution data on temporal-coverage dimensions, data fusion is entailed to blend the advantages of multiple data sources to generate synthetic datasets with more useful information. This thesis proposes two Spatial-Temporal satellite Image Fusion (STIF) algorithms for the two compromises, thereby providing better datasets for different applications. ; For the fusion between spatial resolution and temporal resolution, a robust and adaptive STIF model is developed for complex land surface changes, which are reorganized into the newly defined non-spatial and spatial changes in terms of their changes in the spatial domain. Its robustness and adaptivity are guaranteed by the proposed Non-Local Linear Regression (NL-LR) theory, which can obtain precise similar neighbors and weights of a target pixel by a non-local searching and constrained linear regression scheme. The non-spatial and spatial changes are predicted by two modules, and the more challenging spatial changes are handled by a two-layer fusion framework. Additionally, a regression based high pass modulation is devised to get predictions with clear spatial details and high spectral fidelity. Remarkably, the proposed model forms a unified fusion framework for non-spatial and spatial environmental changes. ; Based on the proposed STIF model, the STIF for various satellite datasets with complex temporal changes in homogeneous or heterogeneous landscapes is performed, which include three types of typical and widely used data, i.e., land surface reflectance, Brightness Temperature (BT), and Aerosol Optical Depth (AOD) at the wavelength of 550 nm. Meanwhile, each dataset is featured by different changes from seasonal variation, anthropogenic ...