Abstract
•DT-enhanced Bayesian nowcasting for real-time emission maps with diverse grid geometries.•High-dimensional stochastic emission states enabling scalable inference.•Closed-form steady-state asymptotic solutions for urban emission dynamics.
Urban emissions management requires accurate modeling to support healthy and sustainable urban development, mitigating adverse impacts associated with air pollutants. However, the inherent spatio-temporal variability of emissions, coupled with incomplete or unreliable monitoring data, poses significant challenges for precise emissions accounting. This study proposes a light-weight digital-twin-augmented spatio-temporal Bayesian nowcasting model for the emissions evolution as a high-dimensional discrete-time Markov chain for complex urban grid systems with diverse topologies. Real-time sensor feeds, simulation surrogates, and context variables sourced from urban digital twins enter the model through an augmented observation operator to provide robust and rapid updates of emissions estimates. The observation operator is linearized around the predictive mean, and the Kullback-Leibler-divergence is bounded by the resulting linearization remainder, enabling high computational efficiency. The resulting Gaussian-conjugate structure delivers closed-form expressions for the posterior mean, variance, and Kalman gain. Numerical experiments conducted on multi-city datasets demonstrate the practical value of our method, showing significant improvements in reducing prediction errors and uncertainty compared to existing methods. By integrating high-fidelity digital twins with Bayesian inference, this study delivers urban policymakers and scientists with a scalable and uncertainty-aware toolset for urban emission management, hotspot detection, and data-driven policy design in complex urban systems.