Mixed-use urban land parcels identification integrating geospatial data and machine learning

Abstract

Urban land use information is essential for urban planning and sustainable development. However, the increasing prevalence of Mixed-Use Urban Land (MUUL) introduces uncertainty in land use mapping. Due to the compatibility of different human activities within MUUL, identifying MUUL is still a challenge for scholars. In this study, we proposed a novel framework by integrating geographic big data and machine learning to identify it. First, to be consistent with urban planning practices, we delineated mapping units from urban planning map. Second, based on the variability of the spatial pattern of POIs, two indices, the Amount of POI (AP) and the Deviation Index (DI), were developed. Finally, the random forest model was employed. The empirical study in the Jianghan District of Wuhan, China, showed that the MUUL and non-MUUL exhibited significant spatial pattern of POIs separability, the average AP and DI of the MUUL were much larger than those of the non-MUUL. Moreover, the relatively high identification accuracy (Kappa Coefficient (KC) = 0.85, Hellden’s Mean Accuracy (MA) = 0.87) further demonstrated the effectiveness of the developed indices. The proposed framework in this study can reduce the adverse impacts of MUUL on urban land identification by avoiding the MUUL from being incorrectly identified as a single function. Moreover, by extracting MUUL in advance, it can improve the research efficiency by avoiding the application of the segmentation algorithm to all mapping units as in the previous study. Overall, this study provides important references for scholars to increase the accuracy of urban land use identification and allows authorities and policymakers to monitor the dynamics of MUUL that are essential to urban vibrancy. The spatial pattern indices proposed in this study can also be applied to urban land use identification studies as important features.

Keywords

• Mixed-use urban land
• spatial pattern index
• points of interest
• random forest