UC Berkeley

Cities are dominant sources of air pollutants and greenhouse gases, with transportation and residential combustion representing two major sectors. Yet urban emissions are rarely observed at the spatial resolution needed to resolve source heterogeneity or evaluate the efficacy of local emission reduction efforts. As a result, persistent discrepancies remain between activity-based emission inventories and estimates derived from atmospheric observations, with unknown uncertainties across sectors. Seeking to address that gap, I developed and applied an analytical and calibration framework that leverages dense networks of miniaturized sensors to infer sectoral emission factors in cities and diagnose biases in current emission inventories.Dense urban sensor networks, such as the Berkeley Environmental Air Quality and CO2 Network (BEACO2N) used in this study, have unveiled fine-scale spatial variability in atmospheric composition that was previously unresolved due to the limited spatial density of traditional observing systems. This new observational capability also raises a new challenge: how to translate dense concentration measurements into quantitative constraints on emission sources. In the first part of this work, I developed an analytical approach to inferring transportation and residential heating emission factors in the San Francisco (SF) Bay Area. The framework models their decay away from major highways using a simplified transport scheme and interprets spatial patterns in enhancement ratios (ERs), where enhancements are defined as the difference between measured concentrations and a network-derived background (i.e., ∆NOx/∆CO2 and ∆CO/∆CO2). Comparing the inferred emission factors with those assumed in existing inventories revealed systematic differences, particularly for residential NOx emissions. Ensuring accurate and consistent sensor measurements across diverse urban settings presents a second challenge, as co-located reference instruments are often sparse or absent depending on deployment location. To address this limitation, the second part of the dissertation developed a physically grounded calibration strategy designed for dense sensor networks. The method combines multi-variable linear regression with constraints from atmospheric chemistry, specifically the NO-O3 titration events that frequently occur in urban environments. This calibration framework enables robust measurement of multiple pollutants while minimizing dependence on co-located reference instruments, thereby providing a scalable pathway for maintaining measurement accuracy in large urban sensor networks at manageable cost.Equipped with the analytical and calibration frameworks developed above, I applied them in the third part of this dissertation to multiple U.S. cities where BEACO2N observations were established—the SF Bay Area, CA, Los Angeles, CA, and Providence, RI—to track and assess how urban emissions vary across space and time under differing infrastructure, energy use patterns, and regulatory implementations, an area that still requires further attention from the research community. Transportation emissions were broadly consistent with inventory expectations, while residential heating emissions showed larger discrepancies. Inventories overestimated residential NOx emission factors by a factor of 2–3, and residential CO emission factors exhibited inter-city variability. These results identify distributed residential combustion as a major remaining uncertainty in urban emission inventories and highlight the value of dense sensor networks for diagnosing sector-specific emissions across diverse cities.Together, this dissertation demonstrates that dense urban sensor networks, combined with scalable and physically grounded calibration, provide the observational resolution needed to refine emission inventories and inform city air quality and climate actions.