Abstract. Air temperature (Ta) has critical implications for various socioeconomic sectors, yet its dynamics are particularly complex in urban areas due to heterogeneous built environments, landscapes, and diverse anthropogenic activities. Physics-based models struggle with intra-city Ta forecasts due to inadequate urban representation and limited spatial resolution. While weather observation networks offer promising alternatives for direct modeling with local Ta time-series, an effective framework to leverage these intra-city discrete sensor data remains lacking. Here, we demonstrate that graph neural networks (GNNs) can harness observation network information to refine Ta prediction at individual locations and elucidate underlying mechanisms. Our novel Mix-n-Scale framework with GNNs achieves over 12 % improvement in short-term Ta forecasts compared to conventional local time-series approaches. Further model evaluation disentangles performance variations with local Ta variability in diverse spatiotemporal contexts, indicating distinct patterns of intra-city heterogeneity across seasonal and diurnal scales. Our findings establish graph-based approaches for leveraging proliferating urban sensor data and advancing understanding of Ta spatiotemporal dynamics in complex urban environments.
Wang et al. (2026) studied this question.