Pixel-Based Classification of Land Cover and Land Use Incorporating External Modeling Products, Sampling Designs, and Multi-Type Features

Timely and accurate information on land cover and land use (LCLU) is fundamental to a wide range of environmental and socioeconomic studies and applications. In recent decades, remote sensing has been recognized as a major source of LCLU information, especially for large areas. Image classification is the underlying process for generating LCLU thematic maps from remotely sensed data. Although this process is affected by many factors, most existing studies focused on the development and refinement of classification algorithms.;This dissertation investigated three factors that may also exert a strong influence on classification performance, namely the utilization of ancillary modeling products, the selection of training samples, and the extraction of informative features in the pixel-based classification process of LCLU. Results indicated that the three factors examined have significant impacts on the performance of LCLU classification. More specifically:;1) Probabilities of change produced from urban growth prediction models were combined as additional information content with spectral data to monitor urban change. The fusion of spectral information and urban change probabilities led to classification accuracy improvements at both the pixel and block scales compared to the exclusive use of either data source.;2) Different sampling designs were tested for selecting a training sample for a binary classification of urban and non-urban cover. The allocation of sample size to classes had more of an effect on classifier performance than the spatial distribution of the training data.;3) Multiple types of features extracted from dual-polarized Synthetic Aperture Radar (SAR) imagery were integrated into land cover classification. Feature synergies provided more pronounced discriminative capability, which brought in considerable improvements in overall and class-specific accuracies, in particular for vegetation classes.