| At present, the worldwide rapid development of urbanization has led the natural vegetation and water dominated landscapes to have been gradually replaced by high thermal storage impervious surfaces such as buildings, roads and parking lots. This has a significant negative impact on urban environment, and hence has attracted increasing concern all over the world. Timely understanding of spatiotemporal information of impervious surface has become more urgent as it is important for monitoring dynamics of impervious surface and evaluating the corresponding impact on urban environment. Fortunately, the advantages of the repetitive, synoptic, and real-time view of the satellite observation over other methods offer a promising method to map large impervious surface areas. Various multispectral- and hyperspectral-resolution remote sensing imagery provides rich sources of data for automatically monitoring impervious surface change on multiple scales.However, current researches on impervious surface retrieval are mainly based on moderate-spectral resolution satellite imagery. As impermeable materials have high heterogeneity in spectrum, the extracted impervious surface information from the moderate spectral resolution satellite images is often mixed with non-impervious surface, which causes the lowering of the accuracy of impervious surface extraction. In order to address the problem, this thesis carries out the study on information extraction of urban impervious surface in several test areas such as Fuzhou, Guangzhou and Hangzhou, China. Apart from moderate spectral remote sensing imagery, the study also used hyperspectral remote sensing data (EO-1 Hyperion) and ground-measured hyperspectral data as data sources. Impervious surface features were derived from remote sensing images by using linear spectral mixture analysis (LSMA).Firstly, to compare the performance of impervious surface extraction between different multispectral images, the study uses two date-coincident images of Landsat ETM+ and EO-1 ALI dated on 26th March,2003. The accuracy of retrieved impervious surface information of the two sensors was assessed. The results show that the ALI image has higher accuracy than ETM+. The overall accuracy of ALI is nearly 9 percentage point higher than that of ETM+, and Kappa coefficient is 0.179 higher than ETM+. Moreover, the root mean square error (RMSE) and systematic error (SE) of ALI are lower than ETM+, with discrepancy of 0.037 and 0.027, respectively. The differences in spectral resolution and radiometric resolution between the two sensors are believed to be the main factors causing these differences in retrieving impervious surface. An increase in spectral information in ALI sensor data can be of help in distinguishing differences between impervious surface and non-impervious ground objects (e.g. soil and vegetation shadow), while the enhancement in radiometric resolution in the ALI sensor can make the sensor have a higher sensitivity in detecting ground surface features.Secondly, a hyperspectral models for separating impervious from non-impervious ground objects was built up based on ground-measured hyperspectral data. The spectral differences of main impervious and non-impervious ground objects were compared by the original in-situ spectrum, first order differential spectrum and removing envelope spectrum. The correlation analysis between ground-measured spectrum and image spectrum was performed. The results suggest that the reflectance of impervious surface material rises with the increase of the wavelength. The near infrared and infrared bands are the main wavelength ranges in the detection of impervious feature. The main wavelengths for distinguishing high albedo impervious from soil are 610nm,693nm,940nm,940nm,1487nm,1794nm,1991nm and 2215nm, and the wavelengths for distinguishing between impervious surface and vegetation are 560 nm,610nm,685nm,942nm,990 nm,1124 nm,1225 nm,1487 nm,1700 nm, 1960 nm and 2150 nm. The correlation analysis between ground-measured spectrum and image spectrum show that the ground-measured hyperspectral data have the very high relevance with the satellite-measured spectrum of ETM+ and Hyperion. The coefficients of determination of the fitted functions of the ground-measured spectrum with respect to the two image spectrum are both greater than 0.8. Moreover, the spectral values of the Hyperion sensor data are more close to the ground-measured spectrum than those of the ETM+ data.Thirdly, as the hyperspectral image of Hyperion has 158 valid spectral bands, there is a high correlation among the spectral bands resulting in strong data redundancy. Therefore, the information analysis, correlation analysis, and principal component analysis for each Hyperion band were carried out and a stepwise discriminate analysis was then conducted to select feature bands for impervious surface retrieving from the 158 bands of the Hyperion image. As a result, eleven feature bands were selected and a new image named Hyperion’was thus composed. The eleven feature bands are at 447nm,942nm,1124nm,1154nm,1245nm,1477nm,1487nm,1699nm,1991nm, 2072nm and 2345nm. To verify the effectiveness of such selected eleven feature bands, the new Hyperion’image was used to retrieve impervious surface The results indicate that the endmembers extracted from the Hyperion’image yields high accuracy in the LSMA model, where the coefficients of determination of the fitted function of image endmembers vs. its pure samples are larger than 0.9, and image endmembers vs. ground-measured endmembers are larger than 0.75. Furthermore, the selected eleven bands were applied to a MODIS image to extract impervious surface features. This produced accuracy of 81.94%.Lastly, the comparison based on the retrieved impervious surface information among Hyperion’, Hyperion and TM/ETM+data was performed. The new Hyperion’ image was used to investigate whether this band-reduced image could obtain higher accuracy in retrieving impervious surface. The three test regions were selected from Fuzhou, Guangzhou and Hangzhou of China, with date-coincident or nearly-coincident image pairs of the used sensors. The LSMA was employed to retrieve impervious surface and the results were accessed for their accuracy. The comparison shows that the Hyperion image has higher accuracy than TM/ETM+ in all of the three test areas, the overall accuracy of the hyperspectral image Hyperion is 4 to 8 percentage point higher than multi spectral TM/ETM+. The advantages of Hyperion in spectral and radiometric resolutions over TM/ETM+ are believed to be the main factors contributing to the higher accuracy. The high spectral and radiometric resolutions of Hyperion image allow the sensor to have higher sensitivity in distinguish subtle spectral changes of ground objects. Especially, the improvement of spectral resolution in near infrared and infrared bands well embodies the spectral characteristics of impervious and non-impervious ground objects, which played a key role to address the mixing problem of impervious surface and other ground surfaces such as soil and vegetation shadow. While, the highest accuracy the 11-band Hyperion’ image achieved is owing to the significant reduction of the band dimension of the image and thus the band redundancy. The overall accuracy of Hyperion’ is 3. to 6 percentage point higher than Hyperion and 7 to 14 percentage point higher than TM/ETM+. The tests of the slected eleven feature band combination in different areas and different images show a very good representativeness of the band combination in retrieving impervious surface, and thus it can be used as an optimal band combination to retrieve impervious surface.
|
PDF Full Text Request