Enhancing air quality forecasts across the contiguous United States (CONUS) during wildfires using an Analog-based post-processing methods

With our growing understanding of the risks of air pollution to human health, air quality forecasting has become a very important tool to enable decision makers to take preventive and corrective measures for current and future policies. In addition, accurate predictions of air quality can help predict the impacts of wildfires on human health, which have an increased risk due to anthropogenic climate change, and mitigate their impacts.  However, errors in air quality forecasts limit their value in long-term decision-making processes. Thus, increasing the accuracy of forecasts is of significant importance.

In this study, we have utilized the Community Multiscale Air Quality (CMAQ) modeling system with a 12 km horizontal grid resolution to generate air quality forecasts for the CONUS domain for June 1^st through September 29^th. Our study spans the seven years from 2015 to 2021, and covers the months when there is a high risk of wildfires. CMAQ is an open-source Cartesian modeling system that simulates the concentrations of atmospheric pollutants at regional scales using emission data and meteorological inputs. We generate these meteorological inputs using the Unified Forecast System (UFS) numerical weather prediction model. We create daily 48-hr forecasts of fine particulate matter (PM2.5), ozone, and related species. We have also included a Carbon Monoxide-FIRE (CO-FIRE) tracer in CMAQ, which tracks CO emitted by wildfires.

Our study consists of three parts. First, we analyze the performance of the CMAQ air quality and UFS meteorological forecasts over seven years of simulation for every EPA defined region using the Air Quality System (AQS) ambient air pollution data from over a thousand monitoring sites across the CONUS. We have found that on average, the CMAQ model performs the best in the east of the CONUS with the lowest RMSE (2 µg/m3) while in the west, where there is a high risk of wildfires, the model has the highest RMSE of up to 8 µg/m3. Temporally, the model under/over-estimates the PM2.5 concentrations during the day/night time, respectively. Next, we quantify the uncertainties in the model’s prediction, and we explore the reasons behind the model biases. Finally, we employ the state-of-the-art Analog Ensemble (AnEn) method to improve the accuracy of the forecasts and quantify the forecast improvements by AnEn. To achieve this, AnEn relies on the current deterministic forecasts, here generated from the CMAQ model, and the past archive of analogous predictions with relative prior observations. By considering the history of predictions along with the current forecast, AnEn has previously shown a significant increase in the accuracy of probabilistic forecasts by requiring significantly less computation resources compared to model-based ensembles. Despite the challenges of using AnEn for wildfires, we hypothesize that it will significantly improve the CMAQ model forecast in the proposed scenarios.