Comparative analysis of auditory and olfactory complaints: spatiotemporal patterns and multi-scale built environment drivers in urban fringes

Urban fringes in rapidly urbanizing regions face multisensory environmental challenges, where auditory and olfactory nuisances interact and impair resident well-being. This study developed a data-driven framework based on approximately 560,000 civic complaint records from Bao’an district, Shenzhen in 2024, integrating multi-level built environment factors to analyse auditory complaint rates (ACR) and olfactory complaint rates (OCR). Using the Qwen large language model for multi-label classification with population-standardised complaint rates per 10,000 residents as dependent variables. Spatial and temporal patterns were examined through Getis-Ord Gi* for hotspot/coldspot identification, dynamic time warping (DTW) for temporal similarity quantification, and spatial similarity assessment. Driving factors were modelled with geographically weighted random forest (GWRF), with SHAP (SHapley Additive exPlanations) values and partial dependence plots (PDP) revealing nonlinear effects and spatial heterogeneity. Finally, government responses were contrasted with model-derived driver priorities to form a closed-loop validation. Results showed high similarity between auditory and olfactory complaints across diurnal and annual scales, with hotspots concentrated in southern mixed-function areas, though ACR hotspots were broader and more intense. Public function facilities contributed most strongly to both complaint types, while residential function, crowd attraction, and traffic flow exhibited positive nonlinear influences, with multiple variables displaying threshold and saturation effects. Government responses prioritise human activity optimisation yet deviated from model-identified functional facility optimisation priorities, while community management optimisation received greater policy attention than its actual driving strength warrants. This study established a multisensory comparative framework centred on urban fringes, achieving closed-loop validation of spatial and temporal dynamics, driving mechanisms, and policy responses. These insights provided targeted, threshold-based recommendations for multisensory urban governance and contributed to more human-centred urban planning.