| The impervious surface is an important environmental indicator, because it is related to many environmental issues. Environmental effects of impervious surface coverage can be roughly categorized into six categories. First of all, impervious surfaces cause increases in volume and velocity of surface water runoff, which increases the risk of flooding during storms. Impervious surfaces also reduce the amount of underground water, and further reduce the supply of water to stream flow, which may lead to dry stream beds during dry seasons. Second, increased volume and velocity of surface water runoff also results in erosion of construction sites, stream banks, and degradation of related habitats. Frequent floods cause channel instability, which can be characterized as the loss of both in-stream and riparian eco-structures. Third, impervious surfaces are very efficient in transporting pollutants into streams, causing non-point source pollution and jeopardizing water quality. Non-point source pollution now becomes the leading threat to water quality. Fourth, impervious surfaces have a thermal impact on the receiving stream, because impervious surfaces absorb more heat. The increased temperature of a stream is correlated to the size of the urban impervious surface. The degradation of stream quality first occurs at the level of 10% of impervious surface cover in the drainage basin, and becomes severe and inevitable when imperviousness reaches 30%. Fifth, urban development causes the closure of shellfish beds, and also reduces the biodiversity in a wetland. Finally, several previous studies have pointed out that urban heat island (UHI) effect is positively related to impervious surface coverage.;Numerous techniques have been developed for impervious surface estimation using satellite remote sensing imagery, including linear spectral mixture analysis (LSMA), regression tree, multi-layer perceptron (MLP), and multiple regression. However, the limitations of some of those techniques significantly affect the classification accuracy and can not be neglected. For example, LSMA has problems related to linear structure and end member selection. MLP contains the limitations of hidden nodes and local minimum. Traditional image classification methods neglect the important image characteristics, such as spatial information, texture, and context. In this dissertation new techniques which can tackle these problems will be developed to obtain better estimation of impervious surfaces. Three hypotheses will be examined in this dissertation, which include: (1) non-linear methods have better performance than linear methods for impervious surface extraction at sub-pixel level due to the non-linear nature of the interactions between photons and features; (2) self-organizing map (SOM) can be a promising alternative to Multi-layer perceptron (MLP) for impervious surface estimation because it has no hidden nodes and local minimum problems, requires short learning time, and yields more consistent results; and (3) the object-based classification using fuzzy rules can yield better extraction of impervious features (e.g. roads and buildings) due to its advantages over traditional image classification and segmentation techniques. |
