Development and Evaluation of North America Ensemble Forecasts of Wildfires and Dust Storms
Date
Authors
Makkaroon, Peewara
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Wildfires and dust storms are two major emission sources of aerosols in the atmosphere, exerting myriad effects on air quality, climate, and human health. Predicting wildfires and dust storms is challenging due to large uncertainties in the inputs and representation of chemical and physical processes in the atmospheric models. Ensemble forecasting has been proposed to improve the predictability of wildfire and dust aerosols. This work presents the development and evaluation of a multi-model ensemble forecast system of wildfire and dust air pollution over North America, leveraging research and operational forecasts operated by George Mason University (GMU) and three U.S. federal agencies: National Oceanic and Atmospheric Administration (NOAA), National Aerospace and Space Agency (NASA), and Naval Research Laboratory (NRL). The ensemble members include three regional models (GMU CMAQ, NOAA NACC-CMAQ, and NOAA HYSPLIT), three global models (NOAA GEFS-Aerosols, NASA GEOS-5, and NRL NAAPS), and one global ensemble (ICAP-MME). Performance of the ensemble forecast was evaluated with aerosol optical depth (AOD) products from MODIS MAIAC, VIIRSSNPP enhanced Dark Target (DT) and Deep Blue (DB), and surface PM2.5 (fine particle) from the AirNow ground network during the 2020 Gigafire events (August-September
2020) in the western United States and the 2021 Spring Dust Season in the Chihuahuan Desert. For the wildfire ensemble, the results showed that, compared to the individual models, the ensemble mean significantly reduced the biases in the wildfire air pollution forecasts and produced more persistently reliable forecasts during extreme fire events. For AOD forecasts, the ensemble mean was able to improve model performance, such as increasing the correlation to 0.57 (0.62) from a range of 0.30-0.53 (0.35-0.56) by individual models when compared to the VIIRS (MAIAC). The ensemble mean also yields the best (second best) overall RANK, a composite indicator representing four statistical metrics (correlation, fractional bias, area hit rate, and false alarm ratio) compared to VIIRS (MAIAC). For the forecast of surface PM2.5 concentration, the ensemble mean demonstrated better performance than any single model with the strongest correlation (0.60 vs 0.43-0.54 by individual models), lowest fractional bias (0.54 vs 0.55-1.32), highest hit rate (87% vs 40%-82%), and highest RANK (2.83 vs 2.40-
2.81), when compared to the AirNow observations. Finally, the ensemble shows the potential to provide a suitable exceedance probability forecast during wildfires with the lowest area false alarm ratio (1.52%) achieved by the ensemble probability of 100%.
For the dust ensemble, the ensemble mean moderately reduced biases in the dust air pollution forecasts and provided fairly reliable AOD and PM2.5 forecasts during extreme dust storms compared to the individual models. For AOD forecasts, the ensemble mean improved forecasting performance less successfully than expected, as demonstrated by slightly decreasing mean bias to 0.01 (0.07) based on VIIRS DT (VIIRS DB), increasing correlation to 0.32 at the low level highest from a range of 0.09-0.31 (VIIRS DB), and yielding the third best overall RANK compared to VIIRS DT and DB. For surface PM2.5 forecasts, the ensemble mean underperformed with a slightly reduced mean bias (3.14), moderately improved low-level correlation (0.40), low area hit rates (15%), and the third best RANK. The ensemble was able to provide only low-medium (20-60%) exceedance probability forecasts during dust events. In addition, the low correlations and large biases of the dust ensemble forecasts during the extreme dust episodes indicate worse performance compared to that of wildfire ensemble forecasts due to larger uncertainties in predicting dust emission, dispersion, and removal. The thesis findings highlight that using the ensemble approach can reduce biases in air pollution forecasts and reasonably improve the model predictability during extreme events such as wildfires and dust storms. The proposed ensemble exceedance probability forecast can be further applied to early warnings of severe air pollution episodes during wildfires and dust storms. However, the reliability of the ensemble forecast is still subject to types of extreme events due to different emission sources as well as initial and boundary meteorological conditions.
Description
Keywords
Dust storms, Wildfires, Multi-model ensemble mean, Ensemble forecast, Air quality forecast