Integrating Urban Landscapes And Social Sensing For Urban Function Identification

Urban function identification plays an important role in natural and social resources management and urban function organization.Conventional approaches rely heavily on establishing physical and socioeconomic factors using statistical and survey data.However,these approaches are time-consuming and labour-intensive,and the established factors are not accurate to depite natural and social characteristics.Urban landscapes containing spatial structures can reflect natural characteristics,while social sensing shows advantages to depict socioeconomics and human activities.Integrating urban landscapes and social sensing is an effective way to identify urban functions.Therefore,this thesis proposes a multi-level urban landscape quantification model,quatifies urban functional semantics,and further integrates urban landscapes and functional topics for automatic urban function identification.The main contributions are as follows:(1)Proposing a multi-level urban landscape quantification model.First,single element-based characteristics are depicted using urban landscape elements including green land,water bodies,buildings,roads,urban facilities and other areas and descriptions including areas,length,number and height.Then,multiple landscape-based characteristics are quantified using the ratio between area of buildings and number of urban facilities,the ratio between area of buildings and area of others,the ratio between area of buildings and the length of roads as well as the ratio between number of urban facilities and the length of roads.Finally,landscape-based characteristics are quantified using urban landscape diversity,urban landscape shape index,mean landscape fractal dimension and degree of urban landscape division.(2)Quantifying urban functional semantics using spatio-temporal topic modelling.First,faced with the issue that conventional approaches cannot model spatio-temporal information,the spatial and temporal documents are constructed to establish urban functional topics by improving the traditional LDA topic model.Then,considering the temporal and semantic differences,the temporal similarity and semantic similarity among topics are calculated using TWED model and semantic similarity model.Finally,by proposing the hierarchical clustering algorithm,urban functional topics are clustered based on temporal similarity and semantic similarity.(3)Integrating urban landscapes and urban functional semantics for automatic urban function identification.First,the modifiable areal unit problem and the delineation of urban functions are discussed.On this basis,by integrating urban landscapes and urban functional semantics,a margin tree algorithm is applied to construct urban function identification model by involving complete linkage and greedy algorithm.Finally,the validation of urban function identification is proposed to ensure the the identification accuracy.Using Futian District,Shenzhen as a study area,experimental data including road networks.Internet maps,points of interest and review data are utilized to support the proposed approaches.The experimental results show that:the overall accuracy of urban function identification based on urban landscapes is 0.637.The overall accuracy of urban function identification based on urban functional semantics is 0.664.The overall accuracy of urban function identification by integrating urban landscapes and urban functional semantics is 0.823,which shows much higher identification accuracy.Accordingly,the approaches including multi-level urban landscape delineation,urban functional semantics quantification and automatic urban function identification can meet the needs for urban function identification with higher effiectiveness and accuracy.It provides a new perspective for urban function identification and classification.