1. Introduction
Wildfires are a significant contributor to atmospheric aerosols and trace gases, causing hazardous air quality and adverse health effects. Research has established links between wildfire smoke exposure and all-cause mortality, as well as respiratory health issues (Cascio 2018). The global average mortality attributable to landscape fire smoke exposure was estimated to be 339 000 deaths annually (Johnston et al. 2012).
Air quality forecast during wildfire events is crucial for public health management and emergency response, including early warnings, but it remains a challenging task due to uncertainties in fire emissions (Pan et al. 2020), plume rise calculations (Ye et al. 2021; Li et al. 2023), and other model inputs/processes (Delle Monache and Stull 2003).
Ensemble forecasting techniques have been increasingly used to improve the predictability of extreme air quality episodes. Sessions et al. (2015) and Xian et al. (2019) developed and evaluated the International Cooperative for Aerosol Prediction (ICAP) multimodel ensemble (MME), a global operational aerosol multimodel ensemble for the aerosol optical depth (AOD) forecast. Li et al. (2020) used an ensemble forecast to predict surface PM2.5 during the 2018 California Camp Fire event using the National Oceanic and Atmospheric Administration (NOAA) Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) dispersion model with different emissions, plume heights, and model setups. Makkaroon et al. (2023) successfully demonstrated a multimodel ensemble forecast system that effectively simulated the 2020 western U.S. “Gigafire,” with the ensemble mean outperforming individual models. These studies highlight the potential of ensemble forecasting to improve the predictability of wildfire air quality.
While multimodel ensemble often outperforms single-model forecasts, some challenges remain. The ensemble mean does not work best all the time (Xian et al. 2019; Makkaroon et al. 2023). For instance, insufficient diversity among models in the multimodel ensemble can limit the ability of the ensemble to capture the full uncertainties and variability tied to different inputs and assumptions. Moreover, if individual models in the ensemble are biased, the ensemble itself may exhibit systematic bias (DelSole and Tippett 2016).
This study presents a new Hazardous Air Quality Ensemble System (HAQES) over the contiguous United States (CONUS) by leveraging real-time forecasts from three U.S. federal agencies (NOAA, NASA, and Navy). We applied a weighted ensemble forecast approach to enhance skill and further improved it by incorporating quantile regression, weighted regression methods to enhance extreme air quality forecasts, and ridge regression to address overfitting concerns. We also introduced a combination of random walk and categoric metrics to assess the performance of the ensemble and individual models against AirNow observations for the year 2022.
2. Materials and methods
a. Fires in 2022.
This paper focuses on the year 2022 when 66 255 fires (12th most since 2001) burned 7 534 403 acres (11th most), as reported by the National Interagency Fire Center. Figure 1 displays the annual and monthly total fire radiative energy (FRE) from Global Biomass Burning Emissions Product (GBBEPx; Zhang et al. 2019), which is highly correlated with fire emissions (Wooster et al. 2005), across the 10 U.S. Environmental Protection Agency (EPA) regions for 2022. In the eastern United States, biomass-burning emissions were concentrated in the southeastern states (Region 4). Although the southeast fires affected a large area, the total FRE was not as high as that of the western wildfires. Region 4’s peak fire period was in March, releasing 6281 TJ of fire energy in 1 month. In the central United States, fire emissions were primarily located in Regions 6 and 7. Central U.S. fires peaked in spring (April–May), releasing 41 634 TJ of fire energy within 2 months. In the western United States, fires were primarily located in Regions 9 and 10, with the peak fire period occurring in the summer, especially in September, when 22 804 TJ of fire energy was released in 1 month. Overall, the strongest fire energy occurred in September (33 631 TJ), followed by May (30 330 TJ).
The (a) annual and (c) monthly total FRE across the (b) 10 U.S. EPA regions for 2022.
Citation: Bulletin of the American Meteorological Society 105, 6; 10.1175/BAMS-D-23-0208.1
b. Description of ensemble members.
The air quality forecast ensemble in this study was developed using both regional and global chemical transport models, including the NOAA High-Resolution Rapid Refresh-Smoke (HRRR-Smoke), Global Ensemble Forecast System-Aerosols (GEFS-Aerosols), National Air Quality Forecasting Capability (NAQFC), the NASA Goddard Earth Observing System (GEOS), and the Naval Research Laboratory (NRL) Navy Aerosol Analysis and Prediction System (NAAPS). These models range from simple smoke tracer models to full air quality models with gas/aerosol chemistry, from high-resolution regional to coarse-resolution global. The ensemble exploits the strengths of these widely different models to improve forecasting accuracy. These models encompass a wide range of emission datasets and plume rise schemes. The study utilizes the 12–36-h surface PM2.5 forecasts initialized at 1200 UTC (forecast hour: 0000–2300 UTC the next day) for all five models. Each model is briefly described below.
1) HRRR-Smoke.
HRRR-Smoke (Ahmadov et al. 2017; Dowell et al. 2022) is an operational real-time three-dimensional coupled weather-smoke forecast model operating at a 3-km spatial resolution over the contiguous United States (CONUS) domain, maintained by NOAA National Centers for Environmental Prediction (NCEP). The HRRR data assimilation system provides initial conditions and a background ensemble for meteorological data assimilation. HRRR-Smoke ingests the satellite fire radiative power (FRP) data from the Suomi National Polar-Orbiting Partnership (SNPP), NOAA-20, and Moderate Resolution Imaging Spectroradiometer (MODIS) Terra/Aqua satellites to estimate wildfire smoke emissions. Since HRRR-Smoke is designed to forecast PM2.5 where smoke is a dominant pollution source, it does not include any nonfire emissions (e.g., anthropogenic emissions) and gas/aerosol chemistry.
2) GEFS-Aerosol.
GEFS-Aerosols (Zhang et al. 2022) is a global atmospheric composition model developed by the NCEP in collaboration with the NOAA Global Systems Laboratory, Chemical Sciences Laboratory, and Air Resources Laboratory. It integrates Finite-Volume Cubed-Sphere Dynamical Core (FV3)-based Global Forecast System (GFS) version 15 meteorology and WRF-Chem’s atmospheric aerosol chemistry. The Aerosol module is based on the NASA Goddard Chemistry Aerosol Radiation and Transport model (GOCART) (Chin et al. 2002) with both fire emission and anthropogenic emission. The biomass-burning emission is from GBBEPx. Smoke plume rise is calculated using a one-dimensional time-dependent cloud module from the HRRR-Smoke model (Freitas et al. 2007). This study utilized the GEFS-Aerosols global PM2.5 forecasts at 0.25° × 0.25° resolution.
3) NAQFC.
NOAA’s operational NAQFC uses the Community Multiscale Air Quality (CMAQ) modeling system version 5.3.1 driven by NOAA’s latest operational FV3-GFSv16 meteorology at the horizontal spatial resolution of 12 km with 35 vertical layers (Campbell et al. 2022). The chemical gaseous boundary conditions are based on static, global GEOS-Chem simulations, while aerosol boundary conditions are dynamically updated from NOAA’s operational GEFS-Aerosols model. NAQFC employs GBBEPx for biomass-burning emissions. The model uses the Briggs (1969) plume rise algorithm to compute wildfire smoke plumes. It also includes anthropogenic emissions and biogenic emissions.
4) GEOS.
The GEOS (Gelaro et al. 2017) system was developed by NASA’s Global Modeling and Assimilation Office. This study used the GEOS Forward Processing (GEOS-FP, version 5.27.1) system at a 0.25° × 0.3125° spatial resolution, which generates analyses, assimilation products, and 10-day forecasts in near–real time. GEOS-FP is built around the GEOS Atmospheric General Circulation Model, the GEOS atmospheric data assimilation system (hybrid–4DEnVar ADAS), and aerosol assimilation (Randles et al. 2017). Aerosols are an integral component of the model physics and are simulated with the GOCART (Chin et al. 2002). Fire emissions come from the Quick Fire Emission Dataset (QFED; Darmenov and da Silva 2015) and leverage low-latency MODIS fire locations and FRP (Collection 6) data. Emissions from fires are distributed in the planetary boundary layer (PBL). The model also includes anthropogenic and biogenic emissions.
5) NAAPS.
NAAPS (Lynch et al. 2016) is developed at NRL Marine Meteorology Division and provides an operational forecast of 3D atmospheric anthropogenic fine and biogenic fine aerosols, biomass burning smoke, dust, and sea salt concentrations on a spatial resolution of 0.333° × 0.333°. The current NAAPS is driven by global meteorological fields from the Navy Global Environmental Model (NAVGEM; Hogan et al. 2014). NAAPS uses a biomass burning source from the Fire Locating and Modeling of Burning Emissions (FLAMBE) inventory, which is based on near–real-time MODIS fire hotspot data (Reid et al. 2009). The wildfire smoke at emission is distributed uniformly through the bottom 4 layers within the PBL. The NAAPS analysis is constrained by the assimilation of MODIS AOD (Zhang et al. 2008; Hyer et al. 2011).
c. Description of observations.
The hourly ground PM2.5 observations from the U.S. EPA AirNow network for the year 2022 are used to evaluate the surface air pollution predictions in this study. The real-time AirNow measurements are collected by the state, local, or tribal environmental agencies using federal references or equivalent monitoring methods approved by the EPA. It contains air quality data for more than 500 cities across the United States (total of 1156 sites), as well as for Canada and Mexico.
d. Ensemble design.
We experimented with four regression methods to determine these weights: multilinear regression (MLR), ridge regression (RR), quantile regression (QR), and weighted regression (WR).
1) MLR.
2) RR.
3) QR.
4) WR.
e. Evaluation method
1) Random walk.
2) Categorical metrics.
Standard metrics like fractional bias have limitations in evaluating the model performance of extreme events, such as wildfires. To address this, categorical metrics can be used to measure the model’s ability to predict U.S. EPA National Ambient Air Quality Standards (NAAQS) 24-h PM2.5 exceedance events (>35 µg m−3; U.S. EPA 2020). Here, we used three categorical metrics (0%–100%) described by Kang et al. (2007):
- 1)Area hit rate (aH) indicates match between forecasted and observed poor air quality exceedances. Higher aH implies a more reliable model.
- 2)Area false alarm rate (aFAR) measures incorrect predictions of poor air quality. Lower aFAR implies a more reliable model.
- 3)Weighted success index (WSI) considers hits, false alarms, and missed exceedance forecasts. A higher WSI suggests a more reliable model.
The equations for these metrics are shown in appendix C.
3. Results
This section begins with evaluating the performance of the unweighted ensemble forecast compared to each individual model [referred to as model 1–5 (M1–M5); the purpose of this study is to assess ensemble forecast skills rather than delving into the performance of individual models. We intentionally rearranged the order of these models and renamed them to models 1–5 to avoid focusing on specific model performance.]. Then, we compare the performance of the unweighted ensemble with that from different weighted ensemble methods.
a. Comparison of MMA with individual models.
The annual mean surface PM2.5 concentration (Fig. 2) predicted by models 1–5 and the MMA is compared to the AirNow observations. The results from different models varied substantially, highlighting the significant uncertainty in wildfire air quality forecasts. Models 1, 2, and 4 overestimate PM2.5 in the Southeast and Northwest, where models 3 and 5 underestimate it. The ensemble mean balanced these overestimations and underestimations and is closer to the observations.
Annual mean surface PM2.5 concentration (contour) predicted by models 1–5 and MMA and observed by the AirNow network (colored circles) for the year 2022.
Citation: Bulletin of the American Meteorological Society 105, 6; 10.1175/BAMS-D-23-0208.1
We compared the MMA with each individual model using the random walk method for major fire regions (EPA regions 4, 6, 7, 9, and 10 from Fig. 1c to Fig. 3). Figure 3 shows the relative forecast score of individual models compared to MMA. A negative value on day n indicates that the overall performance of MMA surpasses that of the individual model from day 1 to day n; the negative trend observed from day n1 to n2 signifies that the MMA consistently outperforms the individual model between n1 and n2, and vice versa. In regions 4 and 10, the consistent downward trend of the random walk scores implies that MMA consistently outperforms individual models. In regions 6, 7, and 9, the scores are mostly negative with some transient positive scores in early 2022 as well as some positive tendency at the end of the year (because of the underestimate of the anthropogenic emission), indicating that MMA performs better than each model most of the time, especially in the wildfire season (summer and fall). The MMA improves air quality forecasting because it balances the model bias of the five individual models (Fig. S1 in the online supplemental material). Overall, the MMA outperforms individual models, demonstrating that ensemble forecasts can effectively reduce forecast uncertainty.
Comparison of individual models to MMA using the random walk method for major fire regions.
Citation: Bulletin of the American Meteorological Society 105, 6; 10.1175/BAMS-D-23-0208.1
To evaluate the forecasting ability of extreme events by individual models and MMA, we calculated the area hit rate, area false alarm rate, and weighted success index for the year 2022 (Table 1). The MMA obtains the highest WSI, third highest hit rate, and the second lowest false alarm rate. Model 4 excels in hit rate but has the highest false alarms. Model 3 has not only the lowest false alarm rate but also the lowest hit rate. Overall, the MMA ensemble works better than the individual models in extreme events air quality forecast, consistent with prior research (Li et al. 2020; Makkaroon et al. 2023).
The aH, aFAR, and WSI for models 1–5 and the MMA for the year 2022. The best results are highlighted in bold, while the worst results are underlined.
b. Weighted ensemble.
MMA improved air quality forecasting, but there is still room for further improvement. Therefore, we explored various weighted ensemble approaches to further enhance forecasting performance. As explained in section 2d, the initial 9 months are utilized for weight calculation, while the subsequent 3 months serve as the testing data, which will be assessed in this section. The first weighted ensemble approach we tested is MLR. Compared to the MMA, MLR reduces the fractional bias by 34%, increases the hit rate by 17%, significantly reduces the false alarm rate by 72%, increases the WSI by 5% (Table 2), and is closer to the observations (Fig. 4). These results demonstrate that the weighted ensemble outperforms the unweighted ensemble.
The FB, aH, aFAR, and WSI for the different models (M1–M5), MMA, and four weighted ensemble forecasts (MLR, RR, QR, and WR) for the October–December 2022 testing period (bold represents the best results and underline represents the worst results).
Scatterplots between predicted and observed PM2.5 for MMA (green), MLR (magenta), and RR (black) for five fire-prone EPA regions. The solid black line represents the 1:1 ratio line for the observations and forecasts, while the dashed black lines represent the 1:2 and 2:1 ratio lines.
Citation: Bulletin of the American Meteorological Society 105, 6; 10.1175/BAMS-D-23-0208.1
The performance of RR is generally comparable to that of MLR (Fig. 4). RR has a slightly lower hit rate, lower false alarm rate, and lower weighted success index (Table 2) compared to MLR. Employing RR to mitigate the overfitting concern of MLR does not notably enhance model performance. This could be attributed to the modest number of models, so the data are not too noisy. Previous studies found that RR can produce a more robust and stable model when the number of predictors is large and the data are noisy (DelSole 2007; Pena and van den Dool 2008).
MMA, MLR, and RR all tend to underestimate the PM2.5 exceedance events (Fig. 4), particularly on the West Coast with high wildfire emissions (R9 and R10). Therefore, we applied QR to enhance predictions of extreme cases. QR enables the ensemble model to predict more polluted events than MLR and MMA (Table 2). QR has a much higher hit rate, which is about 55% higher than the MMA and 33% higher than the MLR. However, sometimes QR overestimates the pollution level when the actual pollution level is not high. Its false alarm rate reaches 32.89%. QR has the highest WSI among all models, including individual models and ensemble forecasts. The fractional bias of QR is higher than that of MLR and RR but still 18% lower than that of MMA.
Another approach to improve the ensemble forecast’s ability to predict extreme cases is WR. WR improved the forecast for PM2.5 exceedance by increasing the area hit rate by 7% compared to MLR and 25% compared to MMA, respectively. QR focuses on the top 10% of cases, whereas WR assigns greater weight to the top 20% of cases. Consequently, predictions using QR tend to be higher than WR (Fig. S6). QR exhibits more overestimation and less underestimation compared to WR. As reflected in Table 2, the fractional bias of QR is 15% higher than WR, the area hit rate of QR is 17% higher, and the false alarm rate is also elevated (55%) compared to WR. WR offers a balanced enhancement. WR’s WSI is the second highest which surpasses MMA and all individual models.
4. Conclusions
In this study, we built a new real-time HAQES by leveraging operational and research fire wildfire smoke forecasts from U.S. federal agencies: GEOS from NASA, NAAPS from NRL, and GEFS-Aerosol, HRRR-Smoke, and NAQFC from NOAA. Automated transfer links have been established between these agencies and George Mason University (GMU). Individual model daily forecast results are automatically transmitted to GMU each day to generate the real-time ensemble forecast results. HAQES significantly enhances forecast accuracy compared to single model forecasts, reducing model bias and increasing the weighted success index for PM2.5 exceedances.
To further enhance forecasting performance, we introduced a weighted ensemble forecast using multilinear regression. Compared to the unweighted ensemble mean, the multilinear regression weighted ensemble reduced model bias by 34%, reduced the false alarm rate by 72%, and increased hit rate by 17%. We also used ridge regression to reduce the overfitting issue of multilinear regression; however, the ridge regression weighted ensemble is close to multilinear regression weighted ensemble, indicating that the overfitting was not significant in our ensemble system.
Finally, we improved the weighted ensemble using quantile regression and weighted regression to enhance the forecasting capability during extreme air quality events. The advanced weighted ensemble increased the hit rate by 55% for PM2.5 exceedance compared to that by the ensemble mean. Our findings provide insights into the development of advanced ensemble forecast methods for wildfire air quality, which offers a practical way to enhance decision-making support through leveraging existing forecasting efforts across federal agencies.
Acknowledgments.
This study is financially supported by NASA Health and Air Quality Program and NOAA Weather Program Office. We thank NASA, NOAA, and NRL for providing the model prediction data used for constructing the ensemble forecast. We appreciate the discussions with Dr. Shunjie Tu on ridge regression, quantile regression, and weighted regression. The views expressed are those of the authors and are not necessarily reflective of the federal agencies (NOAA, NASA, NRL, etc.) or institutions.
Data availability statement.
Here are the links for each model: GEFS: https://ftp.ncep.noaa.gov/data/nccf/com/gens/prod; GEOS: https://portal.nccs.nasa.gov/datashare/gmao/geos-fp/forecast; HRRR: https://nomads.ncep.noaa.gov/pub/data/nccf/com/hrrr/prod; NAQFC: https://airquality.weather.gov; and NAAPS: https://usgodae.org/pub/outgoing/fnmoc/models; HAQES: http://air.csiss.gmu.edu/haqes. AirNow data can be downloaded from https://files.airnowtech.org/?prefix=airnow/2022/.
APPENDIX A
Fractional Bias
APPENDIX B
Significance Test Ka for Random Walk
APPENDIX C
Categorical Metrics
References
Ahmadov, R., and Coauthors, 2017: Using VIIRS fire radiative power data to simulate biomass burning emissions, plume rise and smoke transport in a real-time air quality modeling system. 2017 IEEE Int. Geoscience and Remote Sensing Symp., Fort Worth, TX, Institute of Electrical and Electronics Engineers, 2806–2808, https://doi.org/10.1109/IGARSS.2017.8127581.
Briggs, G., 1969: Plume rise: A critical review. National Technical Information Service Tech. Rep., 81 pp.
Campbell, P., and Coauthors, 2022: Development and evaluation of an advanced National Air Quality Forecasting Capability using the NOAA Global Forecast System version 16. Geosci. Model Dev., 15, 3281–3313, https://doi.org/10.5194/gmd-15-3281-2022.
Cascio, W., 2018: Wildland fire smoke and human health. Sci. Total Environ., 624, 586–595, https://doi.org/10.1016/j.scitotenv.2017.12.086.
Chin, M., and Coauthors, 2002: Tropospheric aerosol optical thickness from the GOCART model and comparisons with satellite and sun photometer measurements. J. Atmos. Sci., 59, 461–483, https://doi.org/10.1175/1520-0469(2002)059<0461:TAOTFT>2.0.CO;2.
Darmenov, A., and A. da Silva, 2015: The Quick Fire Emissions Dataset (QFED): Documentation of versions 2.1, 2.2 and 2.4. NASA Global Modeling and Assimilation Office Tech. Rep. Series on Global Modeling and Data Assimilation NASA/TM–2015-104606/Vol. 38, 212 pp., https://ntrs.nasa.gov/api/citations/20180005253/downloads/20180005253.pdf.
Delle Monache, L., and R. Stull, 2003: An ensemble air-quality forecast over western Europe during an ozone episode. Atmos. Environ., 37, 3469–3474, https://doi.org/10.1016/S1352-2310(03)00475-8.
DelSole, T., 2007: A Bayesian framework for multimodel regression. J. Climate, 20, 2810–2826, https://doi.org/10.1175/JCLI4179.1.
DelSole, T., and M. Tippett, 2016: Forecast comparison based on random walks. Mon. Wea. Rev., 144, 615–626, https://doi.org/10.1175/MWR-D-15-0218.1.
Dowell, D., and Coauthors, 2022: The High‐Resolution Rapid Refresh (HRRR): An hourly updating convection‐allowing forecast model. Part I: Motivation and system description. Wea. Forecasting, 37, 1371–1395, https://doi.org/10.1175/WAF-D-21-0151.1.
Freitas, S., and Coauthors, 2007: Including the sub-grid scale plume rise of vegetation fires in low resolution atmospheric transport models. Atmos. Chem. Phys., 7, 3385–3398, https://doi.org/10.5194/acp-7-3385-2007.
Gelaro, R., and Coauthors, 2017: The Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA-2). J. Climate, 30, 5419–5454, https://doi.org/10.1175/JCLI-D-16-0758.1.
Hanna, S. R., 2006: A review of uncertainty and sensitivity analysis of atmospheric transport and dispersion models. Preprints, 28th NATO/CCMS Int. Technical Meeting on Air Pollution Modeling and Its Application, Leipzig, Germany, NATO/CCMS, 225–237.
Hoerl, A., and R. Kennard, 1970: Ridge regression: Biased estimation for nonorthogonal problems. Technometrics, 12, 55–67, https://doi.org/10.2307/1267351.
Hogan, T., and Coauthors, 2014: The Navy Global Environmental Model. Oceanography, 27 (3), 116–125, https://doi.org/10.5670/oceanog.2014.73.
Hyer, E., J. Reid, and J. Zhang, 2011: An over-land aerosol optical depth data set for data assimilation by filtering, correction, and aggregation of MODIS Collection 5 optical depth retrievals. Atmos. Meas. Tech., 4, 379–408, https://doi.org/10.5194/amt-4-379-2011.
Johnston, F., and Coauthors, 2012: Estimated global mortality attributable to smoke from landscape fires. Environ. Health Perspect., 120, 695–701, https://doi.org/10.1289/ehp.1104422.
Kang, D., R. Mathur, K. Schere, S. Yu, and B. Eder, 2007: New categorical metrics for air quality model evaluation. J. Appl. Meteor. Climatol., 46, 549–555, https://doi.org/10.1175/JAM2479.1.
Koenker, R., and G. Bassett, 1978: Regression quantiles. Econometrica, 46, 33–50, https://doi.org/10.2307/1913643.
Li, Y., and Coauthors, 2020: Ensemble PM2.5 forecasting during the 2018 Camp Fire event using the HYSPLIT transport and dispersion model. J. Geophys. Res. Atmos., 125, e2020JD032768, https://doi.org/10.1029/2020JD032768.
Li, Y., and Coauthors, 2023: Impacts of estimated plume rise on PM2.5 exceedance prediction during extreme wildfire events: A comparison of three schemes (Briggs, Freitas, and Sofiev). Atmos. Chem. Phys., 23, 3083–3101, https://doi.org/10.5194/acp-23-3083-2023.
Lynch, P., and Coauthors, 2016: An 11-year global gridded aerosol optical thickness reanalysis (v1.0) for atmospheric and climate sciences. Geosci. Model Dev., 9, 1489–1522, https://doi.org/10.5194/gmd-9-1489-2016.
Makkaroon, P., and Coauthors, 2023: Development and evaluation of a North America ensemble wildfire forecast: Initial application to the 2020 western United States “Gigafire”. J. Geophys. Res. Atmos., 128, e2022JD037298, https://doi.org/10.1029/2022JD037298.
Pan, X., and Coauthors, 2020: Six global biomass burning emission datasets: Intercomparison and application in one global aerosol model. Atmos. Chem. Phys., 20, 969–994, https://doi.org/10.5194/acp-20-969-2020.
Pena, M., and H. van den Dool, 2008: Consolidation of multimodel forecasts by ridge regression: Application to Pacific sea surface temperature. J. Climate, 21, 6521–6538, https://doi.org/10.1175/2008JCLI2226.1.
Randles, C., and Coauthors, 2017: The MERRA-2 aerosol reanalysis, 1980 onward. Part I: System description and data assimilation evaluation. J. Climate, 30, 6823–6850, https://doi.org/10.1175/JCLI-D-16-0609.1.
Reid, J., and Coauthors, 2009: Global monitoring and forecasting of biomass‐burning smoke: Description of and lessons from the Fire Locating and Modeling of Burning Emissions (FLAMBE) program. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 2, 144–162, https://doi.org/10.1109/JSTARS.2009.2027443.
Sessions, W. R., and Coauthors, 2015: Development towards a global operational aerosol consensus: Basic climatological characteristics of the International Cooperative for Aerosol Prediction Multi-Model Ensemble (ICAP-MME). Atmos. Chem. Phys., 15, 335–362, https://doi.org/10.5194/acp-15-335-2015.
U.S. EPA, 2020: Review of the National Ambient Air Quality Standards for Particulate Matter. Federal Register, Vol. 85, No. 244, 82684, https://www.govinfo.gov/content/pkg/FR-2020-12-18/pdf/2020-27125.pdf.
Wooster, M. J., G. Roberts, G. L. W. Perry, and Y. J. Kaufman, 2005: Retrieval of biomass combustion rates and totals from fire radiative power observations: FRP derivation and calibration relationships between biomass consumption and fire radiative energy release. J. Geophys. Res., 110, D24311, https://doi.org/10.1029/2005JD006318.
Xian, P., and Coauthors, 2019: Current state of the global operational aerosol multi-model ensemble: An update from the International Cooperative for Aerosol Prediction (ICAP). Quart. J. Roy. Meteor. Soc., 145, 176–209, https://doi.org/10.1002/qj.3497.
Ye, X., and Coauthors, 2021: Evaluation and intercomparison of wildfire smoke forecasts from multiple modeling systems for the 2019 Williams Flats fire. Atmos. Chem. Phys., 21, 14 427–14 469, https://doi.org/10.5194/acp-21-14427-2021.
Zhang, J., J. S. Reid, D. L. Westphal, N. L. Baker, and E. J. Hyer, 2008: A system for operational aerosol optical depth data assimilation over global oceans. J. Geophys. Res., 113, D10208, https://doi.org/10.1029/2007JD009065.
Zhang, L., and Coauthors, 2022: Development and evaluation of the Aerosol Forecast Member in the National Center for Environment Prediction (NCEP)’s Global Ensemble Forecast System (GEFS-Aerosols v1). Geosci. Model Dev., 15, 5337–5369, https://doi.org/10.5194/gmd-15-5337-2022.
Zhang, X., S. Kondragunta, A. Da Silva, S. Lu, H. Ding, F. Li, and Y. Zhu, 2019: The blended global biomass burning emissions product from MODIS and VIIRS observations (GBBEPx). Algorithm Theoretical Basis Doc., version 3.1, 30 pp., https://www.ospo.noaa.gov/Products/land/gbbepx/docs/GBBEPx_ATBD.pdf.
