Urban Small Area Population Estimation Based On High-resolution Remote Sensing Data

Small area population statistics in an urban is essential for understanding social, political, economic, and environmental issues. As the rapid development of remote sensing technology, remote sensing has become an important means of population estimation in urban areas. At present, most of the estimation methods are aiming at the total population estimation in an urban area, and the small area population estimation methods are necessary to be addressed further. There are two main reasons. First, the population in small area cannot be estimation by median-low resolution images. Second, how to improve the accuracy and reliability of interpretation using high-resolution images is a hot issue. Meanwhile, how to reduce human intervention to improve the automation of population estimation is the bottleneck in small area population estimation.With the use of Airborne LiDAR data, accurate 3D information of earth surface is able to be obtained rapidly in recent years. And how to extract the related information for small area population estimation, with the combination of high-resolution remote sensing images and airborne LiDAR data, is a hot issue which is worth to be addressed further. In order to solve this problem, aiming at estimating population in small areas timely, accurately, and automatically, typical urban areas of the Unite States are taken as an example, and the key processing algorithms and applications involved are researched. The technical details of the proposed approach are listed below:(1) The approach to small area population estimation using high-resolution remote sensing data. The geometric attributes of residential buildings are used to describe the living space, and residential buildings are proposed as the link between remote sensing data and small area population. The automation of population estimation is improved by automatic extraction the residential buildings based on remote sensing data. And the small area population estimation model is generated according to the relationship analysis between population and geometric attributes of residential buildings.(2) Review of automatic building extraction methods based on remote sensing data and the method of accuracy assessment. In order to compare the building extraction methods and chose a suitable automatic building extraction method for small area population estimation, a method for accuracy assessment of building extraction results is proposed, which assess the quantitative extraction accuracy from three principle aspects of building description (building counts, building areas, and building volumes). First, the existing automatic building extraction methods using remote sensing data are reviewed, and three representative methods are chosen to extract buildings from study areas. It is proved that the accuracy assessment results can reveal the characteristics of different extraction methods, and find out the best extraction method.(3) Automatic detailed building classification method. A novel parcel-oriented detailed building classification method is developed based on remote sensing data. First, according to the formation of urban land use, the features of parcel are described from four aspects, which are geometry characteristics, vegetation characteristics, ground characteristics, and building characteristics. This feature description method can reveal the difference of detailed land use and improve the classification performance of urban land use. And the Multi-Class support vector machine classifier is employed to classify all the parcels. Finally, the buildings in urban area are classified according to the classification results of detailed land use, and all the extracted buildings are classified into three classes, which are single-family, multi-family, and non-residential, based on the requirement of small area population estimation.(4) Small area population estimation model. Considering the patterns of residential population distribution in the Unite States, the small area population estimation model is generated based on the geometric attributes of residential buildings. According to the multivariate linear regression analysis, this small area population estimation model is generated by optimization choice of geometric attributes of residential buildings, such as building counts, building areas, and building volumes. Finally, the estimation model is calibrated and validated, and the results demonstrated that proposed model yielded good small-area population estimation results.The experiment results indicate that the population in small urban area can be estimated accurately using the population estimation model proposed here, based on high-resolution remote sensing data. And the automation of estimation processes is improved by automatic building extraction and building classification.