Micro-scale land-use functional patterns and driving mechanisms in historic districts: a multi-source GIS and interpretable machine learning approach

Historic cultural preservation districts, as carriers of urban memory and traditional spatial patterns, are increasingly confronted with tensions among conservation, utilization, and vitality under the acceleration of urbanization and tourism-oriented consumption. With the introduction of commercial and tourism activities, district functions tend to exhibit localized agglomeration, resulting in pronounced spatial differentiation among street segments. Taking the historic cultural preservation district of Tongcheng City as a case study, this research integrates POI data, population vitality indicators, and street-view imagery to characterize the internal spatial structure of the district from three dimensions: functional supply, population vitality, and the built environment. Using spatial statistical techniques and interpretable modeling approaches, the study examines the spatial clustering of functions and vitality and their impacts on land-use functional patterns. The results indicate that: (1) the land-use functional pattern of the district demonstrates significant spatial differentiation, with high functional density and strong functional concentration primarily distributed along main streets and core nodes, while higher levels of functional mix are more prevalent in internal secondary streets; (2) population vitality varies markedly across time periods and day types, with both the intensity and spatial coverage of vitality during holidays significantly exceeding those observed on weekdays and weekends; and (3) driving mechanism analysis reveals that population vitality and functional structure exert significant influences on land-use patterns, with the strength of these effects varying across different spatial units. From a micro-scale perspective, this study elucidates the structural characteristics of functions and vitality within historic cultural preservation districts and proposes an analytical framework that integrates multi-source GIS data with interpretable machine learning. The findings provide quantitative support for fine-grained renewal strategies and segment-based functional guidance in historic districts.