Author Description

Three dimensional (3-D) Eulerian models, such as CMAQ and WRF-Chem, are widely used for air quality simulations. However, they are computationally expensive and may not provide reliable forecasts at some locations. In this dissertation, two computationally efficient approaches are developed, evaluated, and applied for air quality modeling. The first chapter gives a brief introduction to air quality models and the goals and objectives of this dissertation. In the second chapter, a Lagrangian modeling framework, HYSPLIT-MOSAIC (H-M) is presented. H-M is developed and evaluated to explore the effects of climate change on local air quality in the western U.S. Comparing H-M results with observations and 3-D model output for several short-term and long-term cases shows that H-M can provide reasonable air quality predictions with low computational costs. It predicts reasonable mixing ratios of gas species. The normalized mean bias (NMB) of O3 predictions is generally within ℗ł30%. H-M under-predicts the total aerosol concentrations at most sites, which is comparable to the 3-D model simulations. For the future long-term simulations, statistically downscaled meteorological data from multiple global climate models are used to drive H-M to simulate future air quality under various climate scenarios. The results show the O3 level will be up to 20 ppb higher and the PM2.5 level will be up to 1.5 Îơg/m3 lower in the 2050s than the 2010s. The third chapter introduces a machine learning (ML) modeling framework. ML is firstly used to provide O3 forecasts at Kennewick, WA, where good performance is demonstrated with both accuracy and capability to capture high-O3 events. Next, the ML method is adapted to forecast O3 and PM2.5 concentrations at AQS observation sites in the Pacific Northwest (PNW), which captures up to 79% more high-O3 events and 70% more high-PM2.5 events than the forecasts from the regional chemical transport model (CTM) AIRPACT. With the capability of high-pollutio