Grantee Research Project Results
Final Report: Ensemble Analysis of Global Change Projections for US Air Quality Using a Novel Combination of Lagrangian and Gridded Air Quality Models
EPA Grant Number: R835874Title: Ensemble Analysis of Global Change Projections for US Air Quality Using a Novel Combination of Lagrangian and Gridded Air Quality Models
Investigators: Lamb, Brian , Lee, Yunha , Walden, Von P. , Vaughan, Joseph , Guenther, Alex , Avise, Jeremy C. , Zaveri, Rahul A. , Fast, Jerome D.
Institution: Washington State University , University of California - Irvine , California Air Resources Board , Pacific Northwest National Laboratory
EPA Project Officer: Keating, Terry
Project Period: January 1, 2016 through December 31, 2018 (Extended to December 31, 2020)
Project Amount: $789,547
RFA: Particulate Matter and Related Pollutants in a Changing World (2014) RFA Text | Recipients Lists
Research Category: Air , Climate Change
Objective:
Our overall goal is to improve our understanding of the effects of global change on future particulate matter (PM) levels in the western United States. Our specific objectives are the below.
- Develop a novel application of Lagrangian air quality modeling using an ensemble of high resolution and bias corrected downscaled climate data based on the Multivariate Adaptive Constructed Analogs (MACA) method to provide comprehensive descriptions of PM changes due to meteorological changes.
- Employ this method to examine the effects of the full range of climate projections upon PM levels associated with representative air quality issues in the western US, including wintertime stagnation events, summertime urban to rural transport cases, and wildfire impacts on rural and urban populations.
- Incorporate changes in US anthropogenic emissions, background concentrations, and land use changes within the ensemble of Lagrangian modeling cases to assess the sensitivity of PM to these factors in the western US.
- Integrate the results from these simulations to present the results in forms suitable to inform effective air quality management in the western US and elsewhere.
Conclusions:
The detailed findings from this study can be found in Fan (2020). Below are summaries that relate to the Objectives of Research [which have been adapted from the chapter abstracts of Fan (2020)].
Air quality regulations have reduced emissions of pollutants in the U.S., but many studies suggest that future air quality may still be degraded by global climate change. The simulated climate by various climate models shows a large variation in future decades, so it is important to account for these variations when studying future air quality. A typical approach to forecasting future air quality uses three-dimensional (3D) Eulerian models. However, this approach is too computationally expensive to perform long-term simulations of future air quality under many different climate projections. Therefore, we have developed an efficient Lagrangian air quality modeling framework, called HYSPLIT-MOSAIC (H-M), to study how future air quality at local scales will be influenced by climate change.
In this study, we evaluated H-M simulations against various observations and 3D Eulerian model outputs such as WRF-CMAQ and WRF-Chem and examined future air quality in four urban areas (i.e., Seattle, Sacramento, Salt Lake City, and Boise). Four different model evaluations were performed using: 1) the Carbonaceous Aerosol and Radiative Effects Study (CARES) field campaign over the central valley of California during June 6-19, 2010; 2) the U.S. EPA’s Air Quality System (AQS) measurements in California during June 6-19, 2010; 3) 54 AQS sites over the Pacific Northwest (PNW) from March 2018 to February 2019; and 4) the four urban observation sites from 2005 to 2014. We used 3D Eulerian models as a benchmark for the first three evaluation studies; WRF- Chem simulations for the first two evaluation studies; and WRF-CMAQ simulations for the third evaluation study.
Based on the first three evaluations, we find that overall the H-M simulations are comparable to the observations and the 3D model results. The H-M model captures the diurnal cycle of ozone, although it over predicts ozone concentrations by up to 50% with respect to observations, while the 3D model over predicts by up to 25%. The H-M model also tends to under predict PM2.5 concentrations by more than 40% compared to observations, and the WRF- CMAQ simulations in the PNW under predict by 35%. The H-M model simulates reasonable ozone concentrations for Seattle in July between 2010 and 2019, but it under predicts in Sacramento due to missing wildfire emissions. The PM2.5 predictions are close to the observations at these two cities. The January ozone is over predicted in Salt Lake City and Boise because the temperature is nearly 2.5 degrees higher than the observations. The PM2.5 is largely under predicted because the model poorly predicts air quality in winter stagnation conditions.
Overall, our evaluation results demonstrate that the H-M framework is capable of predicting air quality at local scales. The H-M framework is highly efficient and takes only 10 to 15 minutes to simulate air quality for one day at one site using one processor, while typical 3D Eulerian models use tens or hundreds of processors for multiple hours, depending on domain size and spatial and temporal resolution. Because it is computationally efficient, the H-M model is appropriate for performing numerous ensemble long-term runs with multiple climate scenarios.
We applied this H-M modeling framework to study future air quality in Seattle and Sacramento during July in the 2050s and in Salt Lake City and Boise in January in the 2050s. Because of the higher temperatures projected due to climate change and the associated increase in biogenic emissions in July in the 2050s, ozone concentrations are expected to increase relative to the 2010s in both Seattle and Sacramento. Although biogenic emissions can be neglected in winter, the higher temperatures still cause increased ozone formation in both Salt Lake City and Boise in winter. Due to the reduction of anthropogenic emissions in summer, the PM2.5 concentration is decreased in the 2050s, especially for primary organic aerosol and black carbon, which are reduced by up to 21% and 92%. The organic carbon emission is increased, and other aerosol emissions are reduced in winter, so the change in daily PM2.5 is small and is within 1 μg/m3 in January in the 2050s in Salt Lake City and Boise.
Compared to 3D models, H-M still has some limitations, such as a limited number of trajectories cannot fully represent the whole incoming air pollution, and the simple assumption that the air pollutants are well mixed in the planetary boundary layer. However, the H-M model is a useful tool for providing air quality predictions that require much less computational power than traditional model and yet produce reasonable predictions. The H-M model is especially well- suited for long-term simulations for air quality related to climate change under multiple future scenarios. Because of this, it is a useful tool for predicting future air quality at specific locations under a range of different conditions.
In addition to developing the H-M model, this research project also developed a machine-learning (ML) model framework to demonstrate the forecasting capability for both ozone and PM2.5. Chemical transport models (CTMs) are widely used for air quality forecasting, but these models often suffer from systematic biases that can miss poor air pollution events. In addition, CTMs also require large amounts of computational time. Our initial interest in developing this ML framework was to address difficulty in forecasting unhealthy ozone events during the summer and early fall in Kennewick, WA using a traditional CTM.
We used the 2017 – 2018 historical archives from the Weather Research and Forecasting (WRF) meteorological model forecasts produced daily by the University of Washington, as well as ozone observations from Kennewick to train two ML models, ML1 and ML2: ML1 used the random forest (RF) classifier and multiple linear regression (MLR) models, while ML2 used a two-phase RF regression model with best-fit weighting factors. (To avoid overfitting, we evaluated the ML forecasting system with the 10-time 10-fold and walk- forward cross-validation analysis.) Compared to a traditional CTM (WSU’s AIRPACT5 model), ML1 improves forecasting skill for moderate to high ozone concentration events, while ML2 improves skill for low to moderate ozone concentrations. The ML modeling framework has been used successfully since May 2019 to produce daily 72-hour ozone forecasts for agency and public use. There were no unhealthy ozone events in 2019 or 2020, so the modeling framework only used ML2 forecasts, which showed good performance for days with low ozone concentrations.
After successfully providing operational forecasts at Kennewick, WA, this ML forecast system was applied to other observation sites to predict both ozone and PM2.5. A cross-validation method was used to evaluate the model performance across the PNW. The ML1 captures more high ozone events, but generates more false alarms, while the accuracy of ML2 is better, especially for low ozone events. However, ML2 shows a similar capability to predicting high PM2.5 events as ML1, so there is no need to combine two methods. The ML frameworks provide more accurate results than AIRPACT5 by reducing biases and providing more skill at capturing high ozone events. The PM2.5 predictions from ML2 at specific sites across the PNW are used to interpolate across the region, which allowed direct comparison with the AIRPACT5 model. The ML modeling framework is now providing operational forecasts of ozone and PM2.5 at Air Quality System (AQS) sites across the PNW and the Washington Department of Ecology is actively using these forecasts within their agency.
References:
K Fan, 2020, Computationally Efficient Approaches for Air Quality Modeling, PhD dissertation, Washington State University.
Journal Articles on this Report : 2 Displayed | Download in RIS Format
Other project views: | All 13 publications | 2 publications in selected types | All 2 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Fan K, Dhammapala R, Lamastro R, Lamb B and Lee Y. A machine learning approach for ozone forecasting and its application for Kennewick, WA. EarthArXiv 2020. |
R835874 (Final) |
Exit |
|
Munson J, Vaughan JK, Lamb BK and Lee Y. Decadal Evaluation of the AIRPACTRegional Air Quality Forecast System in the Pacific Northwest from 2009-2018. EarthArXiv 2021. |
R835874 (Final) |
Exit |
Supplemental Keywords:
Air Quality, Climate Change, Lagrangian air quality modeling.Relevant Websites:
Tri-Cities Ozone Forecast Exit
Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.
Project Research Results
- 2019 Progress Report
- 2018 Progress Report
- 2017 Progress Report
- 2016 Progress Report
- Original Abstract
2 journal articles for this project