Estimation Of Impervious Surfaces At Regional Scale Using Landsat Time Series Imagery

Impervious surface is a significant indicator of the degree of urbanization and the quality of urban eco-environment.The rapid development of urbanization brings massive expansion of impervious surfaces,influencing regional eco-environment and restricting regional sustainable development.The spatial distribution and dynamics of impervious surfaces contribute to better understand urbanization and its impacts on regional hydrological environment,surface temperature balance and biodiversity,etc.With the development and improvement of remote sensing,extracting impervious surfaces fast and accurately has become the focus of remote sensing information processing and applications.Since high spatial resolution imagery had difficulty of data acquisition and limited coverage,coarse or medium spatial resolution imagery was often used for estimating impervious surfaces at regional scale.Previous methods of impervious surface estimation mainly focused on the spectral and spatial information from remote sensing imagery.However,they were often limited by spectral and spatial resolution and encountered the problem of spectral confusion and difficulty of spatial features description.The accuracy of estimation could not meet the requirement of practical applications and caused uncertainty in estimating regional impervious surfaces.Furthermore,it restricted the accuracy of regional impervious surface estimation and influenced the accuracy of monitoring impervious surface dynamics.In order to extract regional impervious surfaces precisely and meet the application requirement of monitoring impervious surface dynamics,this study proposed a new method for extracting impervious surfaces at finer time scale.Intra-annual and inter-annual phenological change information in temporal domain was extracted to explore temporal profiles of land cover,and then impervious surfaces were estimated based on differences of temporal characteristics of land cover.The spatial resolution of Landsat data was suitable for regional-scale study,and Landsat data could provide time series observations.Thus,this study used Landsat time series data to present the new approach to address the problem of low accuracy and extract impervious surface dynamics at finer time scale.The main work is as follows:(1)Extracting temporal spectral features of impervious surfacesThis study applied modified neighborhood similar pixel interpolator to remove cloud cover,applied geostatistical neighborhood similar pixel interpolator to fill gaps,and applied modified transfinite pixel smoothing to remove noise to produce high temporal resolution Landsat images.However,due to unevenly distribute of observations,annual and seasonal temporal features derived from dense Landsat time series of spectral indices had unequal length.This study fused high temporal resolution MODIS data to generate Landsat time series of spectral indices with equal time intervals at monthly time scale.(2)Similarity indices of temporal spectral features of impervious surfacesThis study proposed a series of similarity indices to measure similarity of temporal spectral features based on spatial-temporal correlation,using long-term trend change,seasonal change and irregular change in temporal spectral features.These indices represented temporal changes of spectral values to measure similarity of temporal spectral features of different land covers.They aimed at maximizing spatial-temporal differences between impervious surfaces and pervious surfaces and minimizing spatial-temporal differences between different types of impervious surfaces.(3)Semi-supervised fuzzy clustering method for estimating impervious surfacesBased on semi-supervised learning,this study used similarity information between limited labeled data and large amount of unlabeled data to estimate impervious surfaces.Dynamic time warping distance instead of Euclidean distance was introduced in the proposed method,and the objective function was redefined based on similarity indices of temporal spectral features.Then,temporal filtering was applied to check unreasonable land cover changes and correct classification errors for improving performance of clustering and accuracies of impervious surface estimation and monitoring.(4)Extracting time series impervious surfaces and change analysis in the Pearl River DeltaThis study selected the Pearl River Delta as study area,using available Landsat images with cloud cover less than 90%,for the period from 2000 to 2015,to develop high temporal resolution Landsat data.Then,impervious surface dynamics in the Pearl River Delta was combined with social economy and population to analyze urbanization and verify the effectiveness of techniques of proposed method.