CONSTRUCTION AND UNCERTAINTY ANALYSIS OF URBAN CLIMATE MODELS UNDER MULTI-SOURCE DATA FUSION

Abstract

Urban climate modeling faces significant challenges in accurately representing complex urban environments while maintaining computational efficiency and quantifying prediction uncertainties. Traditional approaches struggle with heterogeneous data integration and fail to provide reliable uncertainty bounds essential for urban planning decisions. This study develops a novel Adaptive Bayesian Hierarchical Multi-source Fusion (ABHMF) framework that systematically integrates satellite remote sensing, ground observations, IoT sensor networks, urban morphology databases, and numerical weather predictions. The framework employs an adaptive fusion algorithm with dynamic weight adjustment based on real-time data quality assessment, coupled with comprehensive uncertainty propagation through Bayesian hierarchical modeling. Validation across multiple urban environments demonstrates superior performance, achieving an RMSE of 0.51°C with only 10 seconds of computation time per day, representing a 180-fold efficiency improvement over traditional WRF-Urban models. The uncertainty quantification reveals measurement uncertainty as the dominant component (32.5%), followed by model structure (28.3%) and parameter uncertainty (24.7%). During extreme heat events exceeding 35°C, the framework maintains robust performance with an RMSE of 0.68°C. Cross-city transferability assessment shows consistent accuracy (average RMSE: 0.72°C) without site-specific recalibration. The proposed methodology significantly advances urban climate modeling capabilities, providing reliable predictions with quantified uncertainties for climate-resilient urban planning and real-time monitoring applications.