A response surface model of the air quality impacts of aviation

Abstract

Aviation demand is expected to double in the coming decades, and there are growing concerns about its impacts on the environment. Governments seek to mitigate the impacts of aviation on climate, air quality, and noise by setting various emissions and noise regulations. However, there are complex interactions among these three impact pathways which must be carefully considered. The FAA is developing an integrated suite of software tools to allow policy makers to explore the tradeoffs among these environmental impacts for various regulatory options, and to weigh them against the costs to the aviation industry of those regulations. One component of this tools suite is the Aviation Environmental Portfolio Management Tool (APMT) which is designed to analyze industry economics and environmental impacts. Within APMT, there is a desire for faster models that can analyze multiple policy scenarios for decades into the future in order to inform policy decisions on a reasonable time scale. One particular need is that for a fast surrogate air quality model that relates changes in aviation activity to changes in ambient pollutant concentrations. In this thesis, a response surface model (RSM) is developed for the high-fidelity, but time-consuming, Community Multiscale Air Quality (CMAQ) simulation system. The RSM relates changes in aviation emissions in the United States to changes in ambient concentrations of particulate matter, the main driver of the air quality impacts on public health. Specifically, the surrogate model takes in yearly inventories of landing-taxi-take-off cycle fuel burn, sulfur oxides, nitrogen oxides, and non-volatile primary particulate matter, and returns the resulting changes in ground level annual average ambient particulate matter concentrations. The RSM design space is set to capture likely emissions scenarios over the next 20 years. A low discrepancy sequence is used to generate the 27 CMAQ sample points in order to allow the flexibility of adding more CMAQ simulations as necessary without disrupting the coverage of the design space. Three formulations are then explored for the particulate matter RSM, two kriging models and one regression model. A leave-k-out cross-validation is performed to select the final RSM formulation and analyze its error behavior with the addition of successive CMAQ training points. Finally, the RSM is compared to a previous surrogate model based on the intake fraction method. The ordinary least-squares regression model is found to perform better than the two kriging formulations, yielding a root-mean-square prediction error of around 1%. The error decays at a rate of just over 0.01% with the addition of each of the last 5 CMAQ runs. Running the RSM with a baseline emissions inventory then yields an estimate of the air quality and subsequent public health impacts of current aviation emissions. A Monte Carlo simulation provides uncertainty distributions on the RSM outputs. The net increase in risk of adult premature mortality due to aviation-reported as the number of new incidences across the modeling domain-is estimated at 210, with a 95% confidence interval between 140 and 290. The total cost of the increased risk of adult premature mortality, as well as of several other health endpoints, is estimated at $1.21 billion with a 95% confidence interval between about $370 million and $2.15 billion (year 2000 US dollars). These estimates are roughly half of those given by the prior intake fraction model. Of these total impacts, 30% are found to stem from emissions of volatile organic compounds and volatile particulate matter from organics, another 30% from emissions of sulfur dioxide and volatile particulate matter from sulfur, 28% from nitrogen oxide emissions, and about 11% from non-volatile particulate matter emissions.