A Study On Land Surface Temperature Retrieval Over Urbanized Region Based On Remote Sensing Data Mining

Land use / land cover transformations due to the accelerated development of urbanization not only result in a change of the Earth surface physical properties,but also influence the exchange processes of energe and water between land surface and atmosphere,and biological and geochemical circulation of the Earth,and generally have profound effect on structure and function of regional and even global ecosystem. Especially in Shanghai,the important economic central city in China,the change of urban landscape and land use pattern with high-speed social and economic development has brought far-reaching influence to the evolvement of urban ecological environment.Among varoius urbanized ecological environment effect,the formation and evolution of urban heat island(UHI)effect is closely related with urban land cover change and human social and economic activity,and can be used to generalize and embody the condition of urban ecological environment.The research presented in this paper focuses on the study of spatio-temporal evolvement pattern of land use in course of urbanization and mechanism of UHI effect in the selected study area,Shanghai central city,by the intergration of quantitative remote sensing(RS)method,Geographic Information System(GIS) spatial analytical technique,and spatial data mining technique.Land cover classification with high accuracy,by means of mixed-pixel classification and sub-pixel mapping for Landsat TM/ETM+ images,is used to represent the pattern of land use spatio-temporal evolvement and urban land spatial spraw with urbanization in Shanghai central city.Based on these,land surface temperature(LST)and land surface emissivity(LSE)are retrieved by means of modified mono-window algorithm for Landsat TM6 or ETM+6 data.Furthermore,Decision Tree method and Exploratory Spatial Data Analysis(ESDA)are applied to reveal the spatio-temporal evolution characteristics of LST field in Shanghai central city and mine the mechanism of UHI effect.The results are of important theoretical value in improving the accuracy of remote sensing imagery interpretation and quantitative retrieval and deep study of urban ecological environmental system,and are helpful to establish reasonable urban land use arrangement and planning,and manage and improve urban ecological environment in practice.Five chapters are included in this paper.Chapter one firstly discusses study background and significance,then summarizes recent study results of related field including remote sensing image data mining,mixed-pixel classification and sub-pixel mapping for remote sensing imagery, and remote sensing retrieval of LST.Based on these,the research content, methodology,technical route and innovation features of the dissertation are put forward.Chapter two sets up a Possibilistic C Repulsive Medoids(PCRMDD)clustering algorithm,based on possibility theory and basic principle of c-medoids clustering method.By utilizing initial cluster centers obtained through Subtractive Clustering, mixed-pixel classification is implemented on Landsat TM/ETM+ images of the study area by means of the algorithm,and class proportions of each endmember and spectral reflectance of endmember on images are automatically acquired.Accuracy analysis demonstrates that PCRMDD represents a robust and efficient tool to obtain reliable soft classification results and endmember spectral information in noisy environment. Furthermore,according to the obtained multi-temporal land cover classification of the study area,the pattern of spatio-temporal land use evolvement and urban land spatial sprawl with urbanization in Shanghai central city are explored with the application of spatial analytical function of GIS.By making up of time-frequency local property and multi-scale analytical capability of wavelet transformation and self-learning and prediction function of artificial neural network,chapter three develops a prediction model loose combining wavelet analysis and Radial Basis Functions(RBF)neural network,abbreviated as Wavelet-RBFNN.According to the proportion of each land cover class within each pixel from mixed-pixel classification,based on the assumption of neighbourhood dependence of wavelet coefficients,sub-pixel mapping on remote sensing image is accomplished through two steps,i.e.,prediction of proportion of each land cover class within sub-pixel and soft classification hardening.The experimental results obtained on artificial images,QuickBird image,and Landsat TM/ETM+ images indicate that the sub-pixel mapping method proposed in this paper,can successfully achieve remote sensing image super-resolution enhancement,outperforming cubic spline and Kriging interpolation method in visual effect and prediction accuracy.The sub-pixel mapping results of Wavelet-RBFNN model applied to Landsat TM/ETM+ image of study area at higher spatial resolution demonstrate that the model can be used to simulate higher spatial resolution imagery,and automatically identify and map land cover targets at the subpixel scale,when the cost and availability of fine spatial resolution imagery prohibit its use in many areas of work. Three methods to retrieve the land surface temperature(LST)from the Landsat thermal channel,including Radiative Transfer Equation(RTE),a generalized single-channel method developed by Jimenez-Munoz and Sobrino,and Qin et al.'s mono-window algorithm,are presented in chapter four.In this paper,the method to estimate land surface emissivity(LSE)is modified when the mono-window algorithm is applied to retrieve LST and LSE from Landsat TM6 and ETM+6 data of the study area in 1989,1997,2000 and 2002.Besides,the resultant multi-temporal LST images are normalized radiometrically through relative radiometric correction based on RBF neural network.Based on these,the quantitative relationship between the spatial distribution of LST field and its driving factors in Shanghai central city is set up and the mechanism of UHI effect is mined,by applying decision tree to developing a classification and prediction model of urban thermal environment system.The obtained classification rules are visually represented in the form of classification image of causes of thermal environment formation to reveal the spatial pattern difference of the thermal environment in Shanghai central city under compositive effect of various influencing factors.Furthermore,by utilizing Exploratory Spatial Data Analysis technique and global and local spatial autocorrelation analysis,several spatial statistical indices,such as Global Moran's I,Local Moran's I and Getis-Ord local G,and semivariogram are adopted to qualitatively describe the characteristics of spatial heterogeneity and temporal evolution of thermal landscape at different scales and periods in Shanghai central city.In chapter five,the research results are concluded.Furthermore,future research keys are discussed,too.