Multi-scale drivers of spontaneous plant diversity in urban residential green spaces across dispersal modes: Implications for ecological planning.

Residential green spaces (RGSs) are ubiquitous elements of the urban fabric and can function as low-maintenance, nature-based solutions that reconnect fragmented habitats and deliver ecosystem services. Yet the mechanisms by which multi-scale environments shape spontaneous plant diversity across seed-dispersal modes remain poorly understood. We surveyed 138 RGSs in Fuzhou, China, recording 382 spontaneous plant species, and classifying them into wind-dispersed (anemochory), animal-dispersed (zoochory), water-dispersed (hydrochory), and self-dispersed (autochory). Using Multiscale Geographically Weighted Regression (MGWR) and explainable machine learning (XGBoost-SHAP), we related α-diversity (Shannon index) and local contributions to β-diversity (LCBD) to socio-environmental and connectivity metrics across multiple spatial extents for each dispersal mode and for all species combined. Along the urban-rural gradient, α-diversity declined towards the city center, whereas LCBD increased, indicating fewer but compositionally more distinct communities in highly urbanized cores. Diversity responses were strongly scale-dependent and nonlinear across dispersal groups. Zoochorous species were most prevalent (43.2 %) and were jointly shaped by thermal stress and large-scale connectivity. Anemochorous species (28.8 %) responded mainly to green space connectivity at medium to large spatial scales. In contrast to anemochorous and zoochorous species, which were more diverse in peripheral neighborhoods, autochorous species (25.4 %) were relatively more dominant in densely urbanized cores and were highly responsive to fine-scale patch configuration and intermediate population density. Hydrochorous species were least represented (8.4 %) and largely confined to peripheral sites retaining aquatic habitat. Across groups, external neighborhood conditions generally exerted stronger effects on within-site diversity than internal attributes, indicating that biodiversity in RGSs is structured at neighborhood-to-city scales rather than by plot-scale design alone. Based on these dispersal-specific mechanisms, we propose a multi-scale framework integrating socio-environmental management, ecological corridors, and hierarchical green-space networks. This framework aims to maintain spontaneous vegetation and support sustainable, equitable urban development through resource-efficient, low-input management.