Science Service System
You are here : Home : Proposals_Summary

Summary of Proposal LAN2280

TitleRemote Sensing of Impervious Surfaces in Tropical and Subtropical Areas
Investigator Hongsheng, Zhang - The Chinese University of Hong Kong, Institute of Space and Earth Information Science
Team Member
Professor Qihao, Weng - Indiana State University, Department of Earth & Environmental Systems
Professor Hui, Lin - The Chinese University of Hong Kong, Institute of Space and Earth Information Science
Professor Yuanzhi, Zhang - The Chinese University of Hong Kong, Institute of Space and Earth Information Science
SummaryThis is a book project entitled “Remote Sensing of Impervious Surfaces in Tropical and Subtropical Areas”, which has been approved to be published by CRC Press, Taylor & Francis Group in the United Statesof America. Firstly, this book investigates the state-of-the-art of remote sensing of impervious surfaces by summarizing the environmental and socio-economic impacts of impervious surface, the methods of impervious surface estimation using remote sensing technology, and challenges of remote sensing in humid subtropical regions, where urbanization process has been accelerating. Secondly, this book pays a special attention to the seasonal effects of impervious surface estimation in humid subtropical areas, which is different from that in mid-latitude areas. In the subtropicalmonsoon regions, winter is believed the best season to estimate impervioussurfaces from satellite images. There are little clouds, and most of theVariable Source Areas (VSA) is not filled with water. On the other hand, autumnimages may obtain the lowest accuracy of IS due to the clouds coverage and thewater in VSA. Autumn is a rainy season in a subtropical monsoon region, forwhich clouds occur very often and VSA areas are always filled with water.Consequently, clouds are confused with bright impervious surfaces due to theirsimilar high reflectance, and more water in VSA is confused with dark impervioussurfaces due to their similar low reflectance. Thirdly, a methodological framework of fusingthe optical and SAR images is proposed in this book for impervious surfaceestimation in humid subtropical regions. Three different data sets are used toassess the effectiveness of this methodological framework, including theLandsat TM and ASAR images, the SPOT-5 and ASAR images, and the SPOT-5 andTerraSAR-X images. In addition, different methods (e.g. ANN, SVM and RandomForest) are employed and compared to fusion the two data sources at a mixedlevel fusion of pixel and feature levels. Experimental results showed that thecombined use of optical and SAR image is able to effectively improve theaccuracy of impervious surface estimation by reducing the spectral confusionsthat happen easily in optical images. Moreover, Random Forest (RF) demonstratesas a promising performance for fusing optical and SAR images as it treats thetwo data sources differently through a random selection procedure of variablesfrom different data sources. Finally, major conclusions and recommendationsare provided for the future study of this field. Three issues are identifiedfor further studies: (1) effective feature extraction from both optical and SARdata, (2) comprehensive validation design via study area selection and fieldwork, and (3) effective fusion strategies such as fusion levels and fusionmethods. This book provides both retrospective andprospective views over remote sensing of impervious surface and its applications,challenges and solutions in humid subtropical urban areas. In addition to itsnovel contributions urban remote sensing, this book will also provide importantinsights into remote sensing applications in tropical and temperature regionsby comparing and contrasting.
Final ReportDramatic urbanization processes have happened in many regions and thus have created a number of metropolises in the world, especially in tropical and subtropical regions where most developing countries are being rapidly urbanized. As one of the most important implications, a large increment of impervious surface turned out to be one of the features of fast urbanization process. impervious surfaces has been widely recognized as the most important land cover type in urban areas, and serves as a key environmental indicator of many environment issues such as urban flooding, urban climate, water pollutions, and air pollutions. Moreover, impervious surfaces was also reported to be a significant factor in many socio-economic studies, including urban growth, estimation of population distribution, variation of housing prices, etc. In this project, both optical and SAR data, including ENVISAT ASAR and Terrasar-X data, were employed and assessed to extract the urban impervious surfaces. Some results have been obtained using the data sets of both optical and SAR satellites. An experiment was designed and conducted to evaluate the efficiency of different image features of optical and SAR images for ISE. Spectral, texture and shape features were extracted from optical and SAR images. Two sets of optical and SAR images in Hong Kong and Sao Paulo were tested with different spatial resolutions. Both LULC classification and ISE were conducted separately to investigate the effectiveness of different feature combinations. Some interesting results were found from the experiments. Firstly, with various features extracted from the images, the accuracy can be improved compared with using only the original image data of optical images. This indicates the effectiveness of feature extraction for remote sensing classifications for LULC and ISE. Secondly, with the spectral, texture and shape feature extraction, the combination of optical and SAR images obtained better results than using optical data alone. This is consistent to the results in previous experiments, and also proves the effectiveness of the synergistic use of optical and SAR data. Thirdly, edge effects located on the images edges of the study areas can be found in both the study cases due to the texture extraction using GLCM technique as it applies the moving window with a certain window size. The additional use of SAN texture and shape features was able to reduce this edge effect to some extent. When calculating GLCM features, a fixed size of rectangular neighborhood (moving window) is compulsory, and thus brings the edge effect on the image boundaries. However, as the size and shape of neighborhood is feasible when calculating SAN features, there is no edge effect on the image boundaries. Moreover, the SAN shape feature was able to enhance the edge information on the boundaries between different land objects, which is the main reason why SAN is able to improve the classification accuracy in the Hong Kong study case. However, shape features are much more significant in high spatial resolution images (e.g. SPOT-5 image) than low and medium spatial resolution images (e.g. Landsat TM image). In contrast, the use of shape features in low and medium resolution images may bring some noises and consequently cause some negative impacts. This is exactly the case showed in the Sao Paulo study case. Furthermore, this study presents our effort to synergistically combine the two data sources to improve the mapping of impervious surfaces using the Random Forest algorithm. Four combinations of optical and SAR images, Landsat TM/ETM+ and ENVISAT ASAR, Landsat TM/ETM+ and TerraSAR-X, SPOT-5 and ENVISAR ASAR, and SPOT-5 and TerraSAR-X, were selected in various study areas including Guangzhou, Shenzhen, Hong Kong, Mumbai and Sao Paulo to validate the effectiveness of the methods in this study. Results indicate some interesting findings about the application of Random Forest to the fusion of optical and SAR data. Firstly, the built-in out-of-bag (OOB) error is insufficient for accuracy assessment, and assessment with additional reference data is required for combining optical and SAR images using RF. In this study, the overall accuracy (OA) and Kappa coefficient were employed as an additional assessment. The OA and Kappa value show a consistent but slightly different from the OOB error. Secondly, the optimal number of variables (m) for splitting the decision tree nodes in RF should be some different from the previously reported principle, which indicates m as the root number of the total variables. In this study, an empirical relationship (Equation 7.1) was provided for determining the parameter m. Thirdly, the optimal number of decision trees (T) in RF is not sensitive to the resolutions and sensor types of optical and SAR images, and the optimal T in this study is 20. Fourthly, the combined use of optical and SAR images using RF is effective to improve the land cover classification and impervious surface estimation, by reducing the confusions between bright impervious surface and bare soil, dark impervious surface and bare soil, as well as shaded area and water surface. Fifthly, two SAR data sets, ASAR and TSX, were comparatively employed in this study, with interesting results indicating that higher resolution of SAR data does not guarantee higher improvement compared to lower resolution SAR data. Moreover, the effectiveness of various resolution SAR data may also depend on the classification modes such as LULC classification and impervious surface mapping. Finally, even though the easily-confused land classes tend to be different in different resolutions of images, the effectiveness of combining optical and SAR images is consistent. This improvement is more noticeable for the fusion of optical and SAR images with lower resolutions. The conclusions of this study would serve as an important reference for further applications of optical and SAR images, and as a potential reference for the applications of RF to the fusion of other multi-source remote sensing data.

Back to list of proposals

© DLR 2004-2016