Underlying Surface Monitoring And Thermal Environment Analysis In Highly Heterogeneous Urban Areas

Despite the rapid advancement of satellite-based earth observation technologies and remote sensing applications nowadays, how to accurately identify land cover classes in spectrally complex urban areas still remains a core issue in the remote sensing research community. This issue is particularly serious in urban areas of China. Because of their long history of development, most Chinese cities have experienced the rise and fall of numerous dynasties and different cultures over thousands of years, making their both spatial structures and construction materials far more complex than the western cities of comparable scales. Such extremely complex spectral characteristics have posed a huge technical challenge for urban remote sensing in China. Moreover, the spectral heterogeneity of the urban surface is further enhanced by China’s rapid urbanization over the past three decades. Large-scale rapid transformation of agricultural lands to urban land use, the conversion of natural environment into impervious surface-based residential, industrial or commercial land, the existence of historic buildings in central city, internal swapping and reconstruction of existing land uses, as well as water feature modification and pollution, have all contributed to the highly fragmented and spectrally heterogeneous urban underlying surface.Focusing on the issues of strong landscape fragmentation, high spatial variability and low spectral homogeneity specifically inherent in Chinese urban areas, this study uses Shanghai as a case study to design and evaluate a series of novel digital image analysis methods for accurate urban underlying surface characterization, urban growth pattern exploration, and inferential analyses of urban thermal environments through the use of multi-temporal and multi-source remote sensing data. The main objective of the study is to estimate highly heterogeneous urban land covers, spatial distributions, land surface components, as well as the temporal and spatial variation in an accurate, efficient, and timely fashion. Results can benefit further understanding of the urbanization in China at macro- and meso-scales and provide a scientific basis for land cover and land use planning and sustainable urban development. The major achievements of this study include the following five aspects.(1) A coarse-to-fine (CTF) land cover decomposition method was proposed to extract water bodies, which combined optical and thermal spectra from different seasons to cope with the issue of water bodies’spectral complexity. Traditional methods mostly based on a pixel based water extraction fail to accurately discriminate these water bodies with small features and are easily confused by the low albedo impervious surface and shadow. Moreover, the proposed CTF method can accurately identify these water bodies at subpixel level by integrating linear mixture spectral analysis (LSMA) and multi-temporal change analysis and seems particularly suitable for urban areas with a complex river network. This design was tested with two-date Landsat multispectral data for Shanghai. Results showed that the proposed CTF method was capable of consistently estimating perennial and seasonal water bodies in highly complex urban environments with acceptable accuracy.(2) A tetrahedron based model (vegetation, high-albedo, low-albedo, and soil, shortened as V-H-L-S) was adopted to solve the mismatch between the classical Ridd’s VIS conceptual model and selected endmembers for urban impervious surface extraction in practice. First, the tetrahedron-based model was used to define the distribution of minimum noise fraction (MNF)-transformed pixels in a 3-D space. Then, the tetrahedron was optimized via a multi-objective genetic algorithm to determine four tetrahedral vertices, i.e. the selected endmembers were located in the four small tetrahedron. Finally, the endmember identification was achieved by matching corresponding images. A case study with Landsat satellite imagery in Shanghai, China indicated that the V-H-L-S based tetrahedron method performed better than the traditional pixel purity index (PPI) and the two-dimension (2-D) scatter plots approach for identifying urban surface components with four endmembers.(3) NDVI time series with high spatial and temporal resolutions were generated for urban vegetation mapping by fusing Landsat and MODIS data with an enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM). Temporal mixture analysis (TMA) was then applied to unmix the temporal profile of NDVI for each pixel as a linear combination of NDVI temporal profiles of endmembers and to estimate the agriculture, evergreen vegetation and deciduous vegetation fractions at the subpixel level. Results provided a good agreementto available reference data (R2>0.79, RMSE0.11, MAE<0.84), indicating that the ESTARFM-based TMA approach was effective in reducing endmember variability and separating more vegetation types for mapping vegetation covers in cities with complex spectral confusion.(4) A change detection method based on multi-temporal fraction images was designed to detect impervious surface change in Shanghai, China. First, the annual impervious surface fractions were determined by the TMA of 30 m Landsat NDVI time series acquired in around 1900,1995,2000,2003, 2007 and 2013. Then, change detection was implemented at the sub-pixel level by using the Z-score analysis of multi-temporal impervious surface fractions. This method not only provides the binary change information, but also obtains the characterization about change direction and intensity. Finally, accuracy assessment was performed to evaluate the estimation and change of impervious surface by the historic images derived from Google Earth. Results indicated the proposed method is effective in impervious surface estimation, and can provide accurate change information both in heterogeneous urban and peri-urban environments.(5) A novel technique of extreme learning machine (ELM) was explored to analysis the nonlinearity existing in the relationship between urban LST and impervious surface. Quantitative comparisons of the modeling results from multi-year summer images in Shanghai, China showed that ELM models generally performed much better than linear models and were more efficient than traditional nonlinear models across all images, and adding neighboring pixel impervious surface information could provide improved accuracy for predicting LST. These findings provided robust evidence of strong nonlinearity existing in the causal relationship between complex urban landscapes and LST and might confirm the relationship associated with the surrounding landscapes.