A Multiscale Analysis On Urban Area And Spatial Structure Based On Nighttime Light Data

Urban areas are human settlements that gather most of anthropogenic activities.Their appearance and development are driven by and impact the dynamics of human activity,natural environment,and social-economic activity.Therefore,the research of urban geography,especially urban area and spatial structure,is urgent and meaningful.With the development of geographic information science and remote sensing,related technologies and data,such as nighttime light data,have been extensively applied to urban spatial analysis.However,these applications suffer from three issues.Firstly,the extraction of the long-term urban area is the foundation of spatial and temporal urban structure analysis.Due to different definitions of "urban area" adopted in each existing urban area dataset,these urban area datasets are commonly incomparable while a single urban area dataset is discontinuous in the time series.Secondly,the traditional studies of urban spatial structure relied on the census data to detect urban centers,which limits the urban centers’ extent detection and hierarchy analysis.So far,there is no efficient and published study to overcome these limitations by using remote sensing technology.Thirdly,even though house price is widely used to model the urban residential spatial structure,some studies have questioned the model accuracy,as house price is mainly impacted by policy and the laws of supply and demand,but cannot reflect the real occupancy rate.The house vacancy rate has been proved that it has the capacity to indicate the house price and is also able to represent the real occupancy rate.Thus,the house vacancy rate could be utilized to analyze the urban residential spatial structure.But few countries collect or publish the statistical house vacancy rate data.And the existing statistical house vacancy rate data are recorded under a large spatial scale,which is not fine enough to analyze its spatial pattern.To fill the previous research gaps mentioned before,this dissertation proposed a framework to study urban area and spatial structure under different spatial scales based on nighttime light imagery.The framework includes the long-term global urban area extraction,urban spatial structure detection,and urban residential spatial structure analysis.Main contents and significant findings of this dissertation include:(1)This dissertation defines the "urban area" as the land-used urban area,also referred as urban built-up area and makes a detailed introduction of the proposed RSVM-BMRF(Region-growing Support Vector Machine and Bidirectional Markov Random Field)model.Then the model is applied to DMSP-OLS(Defense Meteorological Satellite Program-Operational Linescan System)stable nighttime light data and three MODIS(Moderate-resolution Imaging Spectroradiometer)products to extract long-term(2000 to 2012)global urban area.The RSVM model is conducted to extract the initial global urban area for each year.The BMRF model integrates the spatial and temporal contextual information of the initial global urban area to modify and optimize the global urban area results.By comparing with existing global land-cover products,the result implies that the human activity indeed appeared in the area where are not impervious surface area.The performance of RSVM-BMRF model is evaluated by using Landsat 7 ETM+ imagery of 2000,2005 and 2010.The overall accuracy and Kappa of these three years are all higher than 85%and 0.7,respectively.And the extracted long-term global urban area has a more stable performance in the time series.(2)The nighttime light intensity is represented as a continuous mathematical surface of human activities,and the elemental features of urban structure are identified by analogy with Earth’s topography.A topographical metaphor of a mount to identify an urban center or sub-center and the surface slope to indicate urban land-use intensity gradient was proposed in this dissertation as the theoretical basis.A localized contour tree method is developed to indicate the urban spatial structure.This model is applied to NPP-VIIRS(Suomi National Polar-Orbiting Partnership-Visible Infrared Imaging Radiometer Suite)nighttime light data and detected 33 urban centers and their extents and hierarchical relations in Shanghai,China.In addition,several useful properties of urban centers have been derived to illustrate the urban centers’ development and polycentricity degree.Owing to the detected hierarchy of urban centers,this model is suitable for multiscale analysis.The analogy between urban spatial structure and topography provides a novel prospect for urban spatial structure research.(3)This dissertation presents a new way to analyze urban residential spatial structure by estimating house vacancy rate from NPP-VIIRS nighttime light data and NLCD(National Land Cover Database)data.After removing the nighttime light intensity from non-residential sources and the difference of nighttime light caused by the different urban area ratio,the nighttime light intensity for non-vacancy area is estimated and then such value is used to calculate the house vacancy rate for the spatial pattern analysis.Because of the limitation of accessing statistical house vacancy data,fifteen metropolitan areas in the U.S.have been selected for this study.The estimated house vacancy rate values are quantitatively validated using corresponding statistical data.The estimated house vacancy rate values have a strong correlation with the statistical data,with high Pearson correlation coefficient(Rpcc = 0.856)and determination coefficient(R2 = 0.734),which is acceptable to obtain house vacancy rate with a fine spatial resolution for the other regions.By visualizing the spatial distribution of estimated house vacancy rate on maps,the results show that the urban planning in the U.S.is reasonable in principle,especially in the area with a high degree of urban development.The urban residential spatial structure is also influenced by natural situations,such as mountain and bay.