LAUSR

Abstract Imbalance of residential environments in cities (e.g., slums and wealthy districts) causes serious social inequality, negatively impacting livable city development and attracting worldwide attentions. Previous studies on residential environmental quality (REQ) mainly focused on general evaluations at city level, while ignored the spatial heterogeneity of REQ inside cities and failed to survey REQ at local scales. This study recognizes the heterogeneity of REQ strongly related to land-use patterns, and aims to explore how local land-use patterns influence REQ. Firstly, a multimodal semantic segmentation method is presented to classify land uses by using satellite images, building data and points of interests. Secondly, a feature system is defined to characterize land-use patterns, which can be extracted at multiple scales based on land-use classification results. Thirdly, these features are fitted with REQ survey data by random forest regressions, which can predict REQ scores across Beijing and give deep insights into how land-use patterns influence REQ. Experimental results indicate that 1) REQ of Beijing is strongly heterogeneous, and our method can generate a REQ map revealing REQ's imbalance across the city; 2) land-use patterns within 700 m have significant impacts on the local REQ; 3) spatial allocations of land uses are more important than proportions for influencing REQ; and 4) our method visualizes the rules that land-use patterns influence REQ, thus can assist urban land-use planning to balance and improve REQ.