Monitoring Urban CO₂ Emissions at Street Scale: Building-Resolving Dispersion Modeling Coupled with an MCMC–Gaussian Process Inversion

Inventories constrained by street-scale atmospheric observations provide a powerful tool for evaluating urban emission-reduction strategies. Kilometer-scale transport models cannot resolve rooftops and street canyons where affordable sensors operate, while conventional inversions typically rely on prescribed error statistics. Here, we present, to our knowledge, the first full-year urban CO₂ inversion driven by building-resolving computational fluid dynamics (CFD) transport, constrained by observations from a mixed mid- and low-cost sensor network, and implemented using Markov chain Monte Carlo (MCMC) sampling with Gaussian process (GP). For Munich, we simulated hourly 2024 CO₂ dispersion at 10 m resolution by sector and source region. Heating and traffic accounted for 60–70% of simulated CO₂ enhancements at sensor sites and were better constrained than power plants. The inversion resolves emissions spatiotemporally while using observations to refine prior assumptions, including emission uncertainties and spatiotemporal correlation scales. Munich’s posterior anthropogenic total was 5690.4 ± 457.2 kt CO₂ yr⁻¹, 4.8% above the prior inventory. In observing system simulation experiments, it recovered the annual total within 1% from priors biased by up to 6.9%. Together, street-scale transport and data-adaptive inversion provide a robust and adaptable framework for translating observations from affordable sensors into refined and accurate information on urban emission.