Using Surface Networks to Infer CO2 and PM2.5 Emissions from On-road Vehicles

Cities and regional governments throughout the world are increasingly making commitments to reducing both total greenhouse gas (GHG) emissions and air quality (AQ) inequities within their boundaries. To plan emission reduction strategies, governments need information regarding the sector and subsector breakdown, spatial origin, and temporal variability in emissions, as well as strategies for tracking emissions changes over the policy-relevant time scales of 1-3yrs. While a wide variety of emission calculator tools are available, commonly used inventories disagree with one another by up to 100% and in different ways in different locations. The result is substantial uncertainty in how well we can describe the total quantity and the sectoral, spatial, and temporal distribution of emissions. One approach to reducing the uncertainty is to create inventories that account for spatially resolved processes with explicit sectoral details (e.g., fleet composition, congestion) using directly measured activity data and comparing these emission inventories with emissions inferred from atmospheric measurements. In this dissertation, we use atmospheric observations to describe vehicle emissions of CO2 and aerosol. Vehicles are the largest sector contributing to CO2 emissions in US cities, and a substantial contributor to health inequities caused by exposure to co-emitted pollutants such as aerosol and aerosol precursors. We develop a novel method for using transportation data (vehicle flows, truck fraction) and near-road observations of aerosol and CO to derive Heavy Duty Vehicle (HDV) aerosol emission factors. We demonstrate that HDV primary aerosol emission factors derived using this method are in line with observations by other studies in the San Francisco (SF) Bay Area and elsewhere, that they decreased a by a factor of ~7 in the past decade, and that they are still 2-3 times higher than would be expected if all HDV were in compliance with California HDV regulations. Second, we use the BErkeley Air quality and CO2 Network (BEACO2N), of low-cost sensors, paired with the Stochastic Time-Inverted Lagrangian Transport (STILT) model and a Bayesian inversion framework to estimate the variation of traffic emissions with speed on a stretch of road in the SF Bay Area. We show that the BEACO2N-STILT-derived fuel efficiency estimates are within 3% of those predicted by the state of California’s EMissions FACtor (EMFAC) model and that our network-inversion system should be able to detect changes in fuel efficiency of the fleet in 3 years or less. Finally, we quantify the impacts of error in background concentration and meteorology, measurement density, and measurement duration on the ability of the BEACO2N-STILT system to accurately constrain monthly and annual CO2 emissions from the transportation sector in the SF Bay Area. We find large seasonal biases in posterior emissions and show that these biases may be significantly reduced by correcting for seasonal biases in background concentration and wind speed. We explore a method for determining thresholds for the number of nodes necessary for convergence of emissions estimates. In assessing the ability of the current BEACO2N-STILT inversion framework to measure highway emissions, we find that this threshold is almost never met when less than 10 BEACO2N nodes are operational, but almost always met when greater than 30 nodes are operational.This dissertation illustrates two methods for using networks of sensors, paired with activity data to make sector and subsector-specific inferences about emissions in urban areas. These methods have the potential to observe both emissions at high resolution and changes in emissions over policy relevant time-scales, giving feedback to governments designing and implementing emissions reduction plans