The National Oceanic and Atmospheric Administration (NOAA)'s National Weather Service (NWS) is on its way to deploying various operational prediction applications using the Unified Forecast System (
The operational air quality predictions in the National Oceanic and Atmospheric Administration (NOAA)'s National Weather Service (NWS) contribute to the
protection of lives and health in the US (
It is well known that the role of aerosols in numerical weather prediction (NWP), through interaction with atmospheric radiation and precipitation physics (direct, semidirect, and indirect effects), and their impact on meteorological fields at both weather and climate scales have been widely recognized in many studies (e.g., Fast et al., 2006; Chen et al., 2011; Grell et al., 2011; Forkel et al., 2012; Muhlbauer et al., 2013; Xie et al., 2013; Yang et al., 2014; H. Wang et al., 2014; Q. Wang et al., 2014). Additional studies at operational weather centers indicate the importance of including aerosol feedback in NWP for operational forecasting (Rodwell and Jung, 2008; Reale et al., 2011; Mulcahy et al., 2014; Bozzo et al., 2020). At the National Center for Environmental Prediction (NCEP), the operational RAPid refresh (RAP) and High-Resolution Rapid Refresh (HRRR) storm-scale modeling systems now include the impact of aerosols from biomass burning emissions on radiation. Due to the importance of aerosol feedback in NWP, the performance of predicted aerosols and their optical properties is critical before implementing the aerosol direct and semi-direct effects in NWP.
In the last decade, global aerosol modeling has grown rapidly to provide operational prediction and air quality alerts in NWP. More than 15 years ago, the National Aeronautics and Space Administration (NASA) implemented an aerosol transport module, the Goddard Chemistry Aerosol Radiation and Transport model (GOCART), online within the its Global Modeling and Assimilation Office (GMAO) Goddard Earth Observing System version 4 (GEOS-4) atmospheric general circulation model (AGCM) (Bloom et al., 2005), which is able to run in climate, data assimilation, and replay modes (Colarco et al., 2010). Later on, it switched to the next version of GEOS-5 to provide near-real-time forecast of aerosols and atmospheric compositions (Rienecker et al., 2008; Molod et al., 2015). Since 2008, as part of the Global and regional Earth-system Monitoring using Satellite and in situ data (GEMS) project, the European Centre for Medium-Range Weather Forecasts (ECMWF) began to provide aerosol forecast (Hollingsworth et al., 2008; Morcrette et al., 2009; Benedetti et al., 2009). In 2010, the International Cooperative for Aerosol Prediction (ICAP) was founded, with one of its goals being the development of a global multi-model aerosol forecasting ensemble (ICAP-MME) for basic research and eventual operational use (Benedetti et al., 2011; Reid et al., 2011; Colarco et al., 2014b). In the ICAP, the complete aerosol forecast models are originals from the European Centre for Medium-Range Weather Forecasts Copernicus Atmosphere Monitoring Service (ECMWF-CAMS), the Japan Meteorological Agency Model of Aerosol species in the Global Atmosphere (JMA-MASINGAR), the NASA Goddard Earth Observing System Version 5 (NASA-GEOS-5), and the Naval Research Lab Navy Aerosol Analysis and Prediction System (NRL-NAAPS) modeling systems. There is also the dust-only model from the Barcelona Supercomputer Center Chemical Transport Model (NMMB/BSC-CTM), the United Kingdom Met Office Unified Model (UKMO-UM), and the NOAA NCEP Environmental Modeling System (NEMS) Global Forecast System (GFS) Aerosol Component (NGAC) (Sessions et al., 2015). Xian et al. (2019) summarized and compared the current states and performances of this global operational aerosol model in the ICAP. The aerosol feedback is not included in these operational models, and it is mostly driven by independent operational/quasi-operational meteorological models developed at different NWP/research centers with different vertical and horizontal resolutions. All these models include the major aerosol species of black carbon (BC), organic carbon (OC), sulfate, sea salt and dust, and GEOS-5 as an extra trace of nitrate. The aerosol optical depth (AOD) root mean square error (RMSE) between ICAP-MME and 21 representative sites of the Aerosol Robotic Network (AERONET) from 2012 to 2017 indicates improvements for find-mode AOD, while it shows small signals of potential model improvement over the regions where is impacted by the biomass burning emission and dust (Xian et al., 2019). The NCEP, in collaboration with the NASA/Goddard Space Flight Center (GSFC), developed the NEMS GFS Aerosol Component version 1 (NGACv1) for predicting the distribution of global atmospheric aerosols (Lu et al., 2010). NGAC is an interactive atmospheric aerosol forecast system with the NEMS global spectral model (NEMS GSM) as the atmosphere model and GOCART as the aerosol model (Wang et al., 2018). NGACv1 was implemented in 2012 and provided the first operational global dust aerosol forecasting capability at the NCEP (Lu et al., 2016). In NGACv1 an in-line aerosol module based on the GOCART model from GEOS-5 (Chin et al., 2000) but limited to dust only was used. NGACv1 used the Earth System Modeling Framework (ESMF) to couple the aerosol module with the GFS. Later, NCEP implemented a multispecies aerosol forecast capability NGACv2, based on NGACv1 through collaborations among NCEP, NASA/GSFC, the NESDIS Center for Satellite Applications and Research (STAR), and the State University of New York at Albany (Wang et al., 2018).
In July 2016, the NOAA took a significant step toward developing a state-of-the-art global weather forecasting model by announcing the selection of a new dynamic core developed at the NOAA Geophysical Fluid Dynamics Laboratory (GFDL) to upgrade the GFS. The GFDL Finite-Volume Cubed-Sphere Dynamical Core (FV3) replaced the spectral GFS core in June of 2019 to drive global NWP systems with improved forecasts of severe weather, winter storms, and tropical cyclone intensity and track. The NOAA is now on the way to integrating various operational applications into the Unified Forecast System (UFS), a comprehensive, community-based coupled Earth modeling system, designed as both a research tool and the basis for NOAA operational forecasting applications.
Here we describe a new aerosol model component developed through
collaborative efforts among the Global Systems Laboratory (GSL), the
Chemical Science Laboratory (CSL), the Air Resources Laboratory (ARL), and the Environmental Modeling Center (EMC). This aerosol component was implemented
operationally in September 2020 to provide 5 d global aerosol forecasts with
The current study presents the development of GEFS-Aerosols and evaluations of its performance in real time and retrospective experiments. Section 2 describes the coupling components of the GEFS-Aerosols member, including the atmospheric component of the FV3GFS model, the aerosol component, and the observation, reanalysis, and model data used for evaluation and comparison. The emission inventories of both anthropogenic emission and biomass burning emissions and other chemical input data are presented in Sect. 3. Sections 4 and 5 are the evaluations of Day-1 real-time forecasts since July 2019 and the Day-1 retrospective forecast for the Atmospheric Tomography Mission (ATom-1) periods of the 2016 summer, respectively. The conclusions and future plans are summarized in Sect. 6.
The global FV3 developed by the GFDL was chosen by the NOAA as the non-hydrostatic dynamical core to be the Next Generation Global Prediction System (NGGPS) of the National Weather Service in the US (Black et al., 2021). Currently, the FV3 was successfully
implemented within the physical scheme of GFS version 15 (named FV3GFS v15), which became operational in June 2019. It has the capability to
provide the metrological basis for coupling with the aerosol prediction component. The GEFS is a weather forecast modeling system made up of 31
separate forecasts, or ensemble members, which have the same horizontal
(
In GFS v15, all sub-grid-scale transport and convective deposition related to aerosol are handled inside the atmospheric physics routines of the simplified Arakawa–Schubert (SAS) scheme. It requires consistent implementation of positive definite tracer transport and wet scavenging in the physics parameterizations, which was implemented subsequent to the forecast system of GEFSv12.
The current aerosol component in the GEFS-Aerosols model is based on the simple bulk aerosol modules from the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) (Grell et al., 2005; Powers et al., 2017), and the first to be used in the global model is the Flow-following finite-volume Icosahedral Model (FIM) as FIM-Chem (Zhang et al., 2022), including aerosol modules from GOCART. The metrological fields (such as land use and other climatological surface fields and vegetation type) are imported from the FV3 atmospheric model to the chemical model to drive the aerosol components. They are consistent in the FV3 atmospheric model and chemical model. Other than the aerosol convective wet scavenging, all the chemically related processes of source and sink, such as emission, dry deposition, settling, large-scale wet deposition, and chemical reactions, are handled by the chemical model. The large-scale wet deposition and dry deposition modules are from WRF-Chem for the GOCART aerosol scheme, which are column-model-driven by meteorological input from the atmospheric model. Large-scale wet removal of aerosols includes below-cloud removal (washout) following Easter et al. (2004), and the details of below-cloud wet scavenging via interception and impaction can be found in Slinn (1984). The dry deposition is the same as in Chin et al. (2002). After updating the chemical tracers in the chemical model, they are passed back to the FV3 atmospheric model for transport and advection.
The GOCART aerosol modules use simplified sulfur chemistry for sulfate
simulation and bulk aerosols of BC, OC, and sectional dust and sea salt (Chin et al., 2000). For OC and BC, the hydrophilic and hydrophobic components are
considered, and the chemical reactions for gaseous sulfur oxidations are calculated using prescribed OH, H
The sea-salt scheme was updated to the most recent version with five size bins based on NASA's second-generation GOCART model (Colarco et al., 2010). The model has the capability of handling volcanic eruptions, which need the
estimate of injection height and SO
A new dust emission scheme, referred to as FENGSHA, was implemented in
GEFS-Aerosols. The scheme, which is also used in the NOAA's National Air Quality Forecast Capability, is modified from the original Owen equation (Tong et
al., 2017, Owen, 1964; Shao et al., 1993),
What makes FENGSHA unique is the way in which the threshold values are
determined. Unlike models based on Marticorena and Bergametti (1995) or Shao
et al. (2011), threshold values are based on surface and wind tunnel flux
measurements of saltation (Gillette, 1988). The drag partition in the
FENSGHA scheme is described by the MacKinnon et al. (2004) parametrization
using the model surface roughness (
A new sediment supply map, the Baker–Schepanski map (BSM), which was developed from the ideas of Chappell and Webb (2016), is currently used within the GEFS-Aerosols FENGSHA implementation. Chappell and Webb (2016) created an approach similar to that of the Raupach (1992) model for lateral cover but instead used a top–down view to describe the area of the turbulent wake using an analogous shadow instead of a 2-D view. The shadow approach is sensitive to the configuration of the roughness elements, meaning that it is sensitive to the placement of the roughness elements in relation to each other. The BSM describes the probability of momentum mixing directly to the soil surface through the canopy. For the application to GEFS-Aerosols, a monthly 3-year climatology of the BSM was created which refers to a monthly average over 3 observation years, in this case 2016, 2017, and 2018, as these were the latest full years at the time of model development.
The aerosol component of GEFS-Aerosols couples directly with the FV3-based atmospheric component via the NUOPC layer (Theurich et al., 2016), which is the foundation of the NOAA's modeling framework (Fig. 1). Figure 2a shows the model-coupled structure that the aerosol component imports meteorological fields from the atmospheric model and exchanges aerosol mixing ratios at each coupling time step via standard NUOPC connectors. Each aerosol species is simulated as a prognostic atmospheric tracer, which is advected by the FV3 dynamical core and undergoes convective mixing and PBL diffusion within the atmospheric physics. All aerosol composition and emission-related processes are computed in GEFS-Aerosols after the atmospheric physics has been advanced and passed to the chemical model following the sequences as emission, settling of dust and sea salt, plume rise of fire emission, dry deposition, large-scale wet deposition, chemical reactions, and carbonaceous aerosol updating. Tracer mixing ratios are then updated and exported back to the atmospheric model.
Diagram showing the components within the NEMS infrastructure.
Bundling all aerosol composition processes in a single model component led to the implementation of a sequential coupling scheme with the atmospheric component. At each coupling time step, the atmospheric dynamical core and physics processes (including radiation) are computed first. The aerosol component is then executed to perform all air composition processes and transfer the updated tracers back to the atmospheric component. Finally, control returns to the atmospheric model, which updates the atmospheric state with new meteorology and aerosol concentrations. To minimize overhead associated with data exchange between model components, GEFS-Aerosols is run on the atmospheric grid, which is imported from the atmospheric component through the NUOPC. Additionally, the coupling run sequence assigns to the aerosol component the identical set of persistent execution threads (PETs) used by the atmospheric model's forecast component. This allows the model to leverage the NUOPC's ability to access coupling fields by memory reference, minimizing the memory footprint for the coupled system.
The sequence of steps involved in moving from the beginning to the end of a forecast process is controlled by the workflow. In a retrospective or real-time forecast, the chemical tracers are cycled from the output of a previous forecast as the initial condition. In operation, the computational cost with an aerosol component would take 129 min for a 120 h forecast. Therefore, the efficiency is about 2.53 times the computational cost by including the aerosol component compared to the one without an aerosol component in the forecast. In the operation, there is no execution time by including the aerosol component as one of the ensemble members since this member only performs a 120 h forecast by including the aerosol component, which is shorter than other members without the aerosol component that perform a 384 h forecast.
The workflow shown in Fig. 2b describes the steps including
pre-processing (prepare input data) and post-processing (process output
data) before and after forecast for GEFS-Aerosols in the forecast system. This initial implementation of GEFS-Aerosols does not include
aerosol data assimilation, so the chemical tracers in the restart files are
used as the chemical initial condition for the next forecast. The yellow box
includes the tasks/steps for the atmospheric mode, while the green box includes the tasks/steps for chemical model. The AOD is calculated in the
post-processing part of the workflow, using a look-up table (LUT) of aerosol
optical properties from the NASA GOCART model (Colarco et al., 2010, 2014a), which was implemented in the Unified Post Processor (UPP,
The real-time forecast experiments were evaluated using the following ensemble analysis, reanalysis data, satellite and in situ observational data, aircraft measurements, and model predictions. We compare each day's model forecast hours with the same day's reanalysis or analysis data and compute the AOD statistics (bias, RMSE, correlation, etc.) for each grid for each pair of model and reanalysis or analysis data for that model forecast hour. We then calculate that for the entire 4 months of the study period and average it over the entire 4 months for each grid point. This method gives an overall estimate of the systematic bias of the model at spatial and temporal scales.
Total AOD instantaneous reanalysis dataset from the second Modern-Era
Retrospective analysis for Research and Application (MERRA-2, Gelaro et al.,
2017). The MERRA-2 reanalysis provides various AOD products at 0.625
MODIS provides near-global coverage of aerosol measurements in space and
time. We have used a MODIS Level-3 (daily and monthly at 1 The Visible Infrared Imaging Radiometer Suite (VIIRS) sensor on board the Suomi National Polar Orbiting (S-NPP) satellite provides sets of aerosol
environmental data records (EDRs) based on daily global observations from
space (Jackson et al., 2013; Liu et al., 2013). Beginning in 2012, the VIIRS has provided AOT at 550 nm at a global 0.25 AERONET, which is a global ground-based network of automated sun-photometer measurements, provides AOT, surface
solar flux, and other radiometric products (Holben et al., 1998). It is a well-established network of over 700 stations globally, and its data are
widely used for aerosol-related studies (Zhao et al., 2002). AERONET employs
the CIMEL Sun–sky spectral radiometer, which measures direct Sun radiances at eight spectral channels centered at 340, 380, 440, 500, 675, 870, 940, and 1020 nm. AOT uncertainties in the direct Sun measurements are within The Atmospheric Tomography Mission (ATom) studies the impact of
human-produced air pollution on greenhouse gases and on chemically reactive
gases in the atmosphere (Wofsy et al., 2018). ATom deploys instrumentation
to sample atmospheric composition, profiling the atmosphere in the 0.2 to 12 km altitude range. Flights took place in each of the four seasons over a 22-month period in 2016 through 2018. They originated from the Armstrong Flight
Research Center in Palmdale, California, flew north to the western Arctic,
south to the South Pacific, east to the Atlantic, north to Greenland, and
returned to California across central North America over the Pacific and
Atlantic oceans from
AERONET site information, the correlation coefficients, and the root mean square error (RMSE) of GEFS-Aerosols, ICAP, and NGACv2 AOD with respect to that of AERONET observation for the period 5 July–30 November 2019. Correlation coefficients are at the 95 % confidence interval.
The International Centers for Aerosol Prediction – Multi-Model Ensemble (ICAP-MME) provides daily 6-hourly forecasts of total and dust AOD globally
out to 120 h at 1 The NEMS GFS Aerosol Component Version 2.0 (NGACv2) for global multispecies
aerosol forecast developed by NCEP and collaborators was previously used to
provide operational global multispecies aerosol forecasts at the NCEP (Wang et al., 2018). The anthropogenic emissions are based on EDGAR V4.1
(Janssens-Maenhout, 2010) and AeroCom Phase II (Diehl et al., 2012). The
fire emissions of carbonaceous aerosols and SO NGACv2 uses the same physics package as the 2015 version of the operational
GFS. NGACv2 included additional aerosol species of sea salt, sulfate, organic
carbon, and black carbon from the updated GOCART modules. Both science and
software upgrades in the global forecast system were updated and implemented
in NGACv2 in March 2017 to provide 5 d multispecies aerosol forecast products at the T126 L64 resolution at approximately 100 km. The comparison of
model configurations for GEFS-Aerosols and NGACv2 has been shown in Table 2
based on the model information from Wang et al. (2018).
Comparison of model configurations between GEFS-Aerosols and NGACv2.
The preprocessor, PREP-CHEM-SRC version 1.7 (Freitas et al., 2011), a comprehensive tool that prepares emission fields of trace gases and aerosols
for use in atmospheric chemistry transport models, was used to generate the
anthropogenic emissions, background fields of OH, H
Figure 3 shows the comparisons of anthropogenic emissions between CEDS and
HTAP for SO
Anthropogenic emissions of CEDS (2014) and HTAP (2010) for
SO
We validated the GOCART background fields of OH and H
The operation of GEFS-Aerosols uses the GBBEPx v3 emission with FRP. The GBBEPx v3 system produces daily global biomass burning emissions of PM
A one-dimensional (1-D) time-dependent cloud module from the High-Resolution Rapid Refresh (HRRR)-Smoke model has been implemented in GEFS-Aerosols to calculate injection heights and emission rates online (Freitas et al., 2007). The new scheme in HRRR-Smoke is a modified version of the 1-D plume rise scheme used in WRF-Chem (Freitas et al., 2007). The new plume rise scheme uses the FRP data instead of the look-up table to estimate the fire heat fluxes (Ahmadov et al., 2017). The 1-D cloud module is able to be applied to GBBEPx v3 fire emission datasets to account for plume rise that distributes the fire emissions vertically and to better simulate the fire events and pollution transport of smoke plumes.
To validate model performance when using the GBBEPx v3 fire emissions with a plume rise module based on real-time FRP data, we compare the real-time GEFS-Aerosols AOD with other reanalysis data, satellite observations, and the NGACv2 model for the big fire event in August 2019. Smoke from large fires burning in the Amazon rainforest, primarily in Brazil, Bolivia, Paraguay, and Peru, stretched over northern South America in mid-August. Figure 4 shows the total AOD forecast on 25 August compared against the NGACv2 model, MERRA-2 reanalysis data, and satellite observations of VIIRS and MODIS. For both satellites, daily gridded AOD is used to compare against the model forecast at 18:00. The GEFS-Aerosols AOD is able to reproduce the enhanced AOD due to several fire events over South America near the border of Bolivia, Paraguay, and Brazil, which were also observed by the VIIRS and MODIS satellite instruments and captured by the MERRA-2 analysis. Although there are a lot of missing data downwind from the fires in the satellite observations of VIIRS and MODIS, especially over the South Pacific, GEFS-Aerosols and MERRA-2 results are consistent in showing the transport of fire plumes into the tropical Pacific and South Atlantic. In contrast, the NGACv2 model does not capture these fire events and exhibits only a very slight AOD enhancement. NGACv2 AOD is more than 80 % smaller than the observations over the fire source region and produces little or no transported smoke over the surrounding areas.
Total AOD forecast on 25 August compared to the NGACv2 model, MERRA-2 reanalysis data, and satellite observations of VIIRS and MODIS. The 18:00 forecasts from both models for that day and daily satellite data are used in the figure. Satellite data gaps are in white.
Beyond the fires burning in South America, an even greater number of blazes
on the African continent are observed by the satellite images at almost the
same time in August 2019. Angola experienced almost 3 times more fires than Brazil in mid-August 2019. There were around 6000 fires in Angola, more than 3000 in Congo, and just over 2000 in Brazil, according to NASA satellite imagery
(
Future work will explore the use of diurnal fire profiles based on historic Geostationary Operational Environmental Satellites-R Series (GOES-R) fire products applied to estimate biomass burning emissions to enhance forecast behavior. Additionally, a parameterization based on the fire weather index (FWI) to estimate biomass burning emissions on longer temporal scales may help to improve and extend the forecast of fire impacts.
A real-time forecast was performed starting on 1 July 2019 at
We evaluated the GEFS-Aerosols model performance with the daily AERONET data globally. The locations of the 60 selected AERONET sites where these comparisons were made are listed in Table 1. It also indicates the correlation and RMSE of GEFS-Aerosols, ICAP, and NGACv2 AOD with respect to that of AERONET observation. The GEFS-Aerosols, NGACv2, and ICAP predictions are sampled at the same locations as the AERONET sites for these comparisons. Figure 5a and b show the correlation coefficients between daily total AOD observed by AERONET and the Day-1 forecast of model AOD from GEFS-Aerosols and NGACv2 for the period between 5 July and 30 November 2019. The correlation coefficients range from 0.5 to 0.9 for GEFS-Aerosols at most sites, except for several sites in South America, Africa, and East Asia near fire source regions, which are slightly lower than those of the ICAP. In contrast, the correlation coefficients of daily total AOD between the NGACv2 and AERONET observations are lower than 0.5 globally, even ranging from 0.1 to 0.3 at most sites. A more quantitative display of correlation coefficients for a selection of 60 AERONET sites for GEFS-Aerosols and NGACv2 is presented in Table 1. This comparison strongly indicates the improved performance of total AOD daily variation in GEFS-Aerosols prediction when compared to NGACv2. There are 20 sites (about 30 % of the site total) displaying highly correlated total AOD for the AERONET data and GEFS-Aerosols, with the correlation coefficients exceeding 0.7. In contrast, there is only 1 site with a correlation coefficient larger than 0.7 for NGACv2 model vs. AERONET, and 19 sites have correlation coefficients that are less than 0.2 for AERONET and NGACv2. Figure 5c shows the RMSE of GEFS-Aerosols and NGACv2 with respect to AERONET observation. Most of the RMSE values are below 0.25 in GEFS-Aerosols over North America, Europe, and Africa. However, the RMSE values in a lot of sites over Africa and Asia are above 0.3 in NGACv2. From Table 1, the ICAP results show the best performance in both the correlation and RMSE.
Correlation coefficients and RMSE between AERONET daily total AOD observations and GEFS-Aerosols, ICAP, and NGACv2 for the period 5 July–30 November 2019. Correlation coefficients are at the 95 % confidence interval.
In addition to comparing to the AERONET data, Fig. 6 shows the Day-1 AOD prediction of GEFS-Aerosols and NGACv2 compared to the MERRA-2 reanalysis and MODIS observations averaged from July to November 2019. The GEFS-Aerosols prediction is able to capture the geographical features of AOD as represented by the MERRA-2 reanalysis data and MODIS satellite observations, such as the dust plumes over northern Africa and the Arabian Peninsula, biomass burning plumes in southern Africa, South America, northwestern North America and eastern Europe, polluted air over East and South Asia, and high-latitude sea-salt bands over the Southern Hemisphere. The high AOD over southern Africa and northern India is more comparable to the MODIS observation than that of NGACv2. As pointed out by Bhattacharjee et al. (2018), the NGACv2 predictions exhibit widespread underestimates over most of these high AOD regions, such as East Asia, and fire source regions of southern Africa, eastern Europe, and Southeast Asia.
Day-1 AOD prediction averaged during 5 July–30 November 2019 for GEFS-Aerosols and NGACv2 compared to MERRA-2 reanalysis and MODIS.
Day-1 AOD forecast biases of GEFS-Aerosols and NGACv2 compared to MERRA-2 averaged during 5 July–30 November 2019 for dust, OC, and sulfate.
Figure 7 indicates the Day-1 AOD forecast biases of GEFS-Aerosols and NGACv2 with respect to MERRA-2 reanalysis between 5 July and 30 November 2019 for dust,
OC, and sulfate. The predicted dust AOD in GEFS-Aerosols is quite comparable to that of MERRA-2 results, with only small negative biases of
Differences of GEFS-Aerosols and NGACv2 Day-1 predictions of total AOD compared to MERRA-2 reanalysis averaged during 5 July–30 November 2019.
OC is a major component emitted from wildfires, and OC AOD is a good indicator of the performance of fire impacts. GEFS-Aerosols OC AOD shows smaller biases compared to the MERRA-2 reanalysis than those of NGACv2 (Fig. 7c and d). Positive biases in GEFS-Aerosols OC AOD of less than 0.2 occur mainly over southern Africa, East Asia, South Asia, and the Middle East. The GEFS-Aerosols overprediction of OC AOD compared to MERRA-2 over eastern China may be associated with the overestimate of anthropogenic emissions by using CEDS 2014, since this is not a major fire source region. GEFS-Aerosols shows small negative biases, of less than 0.1, over South America and central and eastern Europe. Overall, the biases of OC AOD in NGACv2 relative to MERRA-2 are dominated by underprediction globally, with the largest biases of more than 0.3 over major fire source regions of southern Africa, the Amazon region of South America, Southeast Asia, and Siberia (Fig. 7d).
For sulfate AOD, the GEFS-Aerosols forecast overpredicts MERRA-2 by
The summary comparison of the GEFS-Aerosols and NGACv2 Day-1 total AOD prediction biases with respect to MERRA-2 reanalysis between 5 July and
30 November 2019 is shown in Fig. 8. Generally, the GEFS-Aerosols model is able to
reproduce the total AOD very well, much better than NGACv2 (see Fig. 8a
and b). The GEFS-Aerosols overpredictions over eastern China and the Southern Hemisphere (
We choose some sites near the major fire source region, which have available observation data for the duration of this study and hold long records based on various previous studies. Figure 9 indicates the total AOD time series of AERONET observations compared against ICAP, NGACv2, and GEFS-Aerosols model predictions at the four AERONET sites near the fire source regions of South America during the period of 1 July–30 November 2019. At the Alta Floresta site, which is in the middle of the Amazon fire source region, the daily AOD variations of both the ICAP and GEFS-Aerosols Day-1 predictions are quite consistent with those of the AERONET data, especially as they are able to reproduce two peaks in AOD enhancements in late August and late September caused by fire plumes (Fig. 9a). The correlation (RMSE) is 0.66 (0.23), 0.9 (0.12), and 0.68 (0.31) for GEFS-Aerosols, ICAP, and NGACv2. Obviously, NGACv2 results underpredict AERONET observations almost throughout the whole period, with a significantly larger bias than GEFS-Aerosols or ICAP, and the two August–September peaks in total AOD enhancements are essentially missed in the NGACv2 prediction.
Day-1 AOD forecasts of GEFS-Aerosols, ICAP, and NGACv2 verified against AERONET sites in South America during 5 July–30 November 2019.
The Itajuba site is located southeast of the Alta Floresta site and in the downwind areas of the Amazon fire source region. The total AOD time series of GEFS-Aerosols prediction match closely those of ICAP and AERONET during most of the time period, though there are some discrepancies from the end of August to mid-September, when GEFS-Aerosols underpredicts the high AOD episode (Fig. 9b). GEFS-Aerosols is able to predict the two AOD enhancements in mid-October and early November, which is quite comparable as ICAP. The correlation (RMSE) is 0.856 (0.15) and 0.936 (0.09) for GEFS-Aerosols and ICAP with respect to AERONET at the site of Itajuba and only 0.451 (0.22) for NGACv2. The NGACv2 prediction also generally underestimates the observations at this site too (Fig. 9b). The NGACv2 results are closer to ICAP, GEFS-Aerosols, and AERONET before August, and NGACv2 shows a slight increase in total AOD in early September, but the NGACv2 AOD magnitude is much lower than the AERONET magnitude by about a factor of 5–7 from mid-August onward.
Located in the southern part of the Amazon fire region, the site of Santa Cruz Utepsa is south of the Alta Floresta site. The correlation (RMSE) values of GEFSA-Aerosols and ICAP with respect to AERONET are 0.8 (0.18) and 0.88 (0.13), respectively, which shows better performance than those of NGACv2, with 0.3 (0.39) at this site in predicting the total AOD through the 5 months from July to November (Fig. 9c). The model not only reproduces the total AOD temporal variation of the AERONET results, but also captures several fluctuations of high AOD in August and September caused by Amazon fire events. Again, some of the fluctuations in total AOD were captured by the NGACv2 prediction, but the modeled AOD magnitudes are 2–4 times lower than the observations.
The last site of Rio Branco is also located in the Amazon fire source region, but to the west of the Alta Floresta site. There are some missing data at this site for the AERONET total AOD from mid-July to mid-September (Fig. 9d). During this period, the GEFS-Aerosols prediction is slightly lower than ICAP, by about 5 %–10 %. Both ICAP and GEFS-Aerosols total AOD matches the AERONET variations well when the AERONET data are available again from mid-September. Several peaks of total AOD are also captured by GEFS-Aerosols in mid-September and early November. The NGACv2 prediction shows enhanced total AOD in mid-August, with low biases by more than 2–3 times compared to ICAP and GEFS-Aerosols. For other enhancements of total AOD after October, the NGACv2 results completely miss the fire events and do not show any fluctuations. The correlation (RMSE) is 0.80 (0.24) and 0.90 (0.17) for GEFS-Aerosols and ICAP with respect to AERONET at the site of Rio Branco and only 0.51 (0.44) for NGACv2.
We also evaluate the total AOD time series of AERONET against ICAP, NGACv2, and GEFS-Aerosols for fire regions of central and southern Africa. The comparisons at seven AERONET sites from July to November are shown in Fig. 10. Generally, the GEFS-Aerosols predictions are able to capture the daily total AOD variation measured by AERONET. At the site of Misampfu, the GEFS-Aerosols mode is somewhat better than that of ICAP in predicting the peaks of high AOD. The correlation coefficients at the sites of Ascension Island and Lubango are much higher than those of ICAP (see Table 1). While both GEFS-Aerosols and ICAP overpredicted the total AOD most of the time throughout the 3 months at the Bamenda station located north of the major African fire source region, the NGACv2 total AOD forecast shows underprediction at most of the AERONET sites in this region. Meanwhile, NGACv2 and ICAP predictions are not consistent with AERONET either, especially for several observed high peaks which are not reproduced by the model results (e.g., Gabon). At the remote site of Ascension Island located west of the African fire source region, GEFS-Aerosols and ICAP are able to capture the AOD enhancements in mid-August and show the best performances of the three different models (see Table 1). For other sites that are located in the fire source region, such as Monguinn, Misamptu, Maun Tower, and Lubango, the prediction of the GEFS-Aerosols model shows higher correlations of 0.68, 0.79, 0.84, and 0.71 than those of NGACv2 at 0.51, 0.26, 0.29, and 0.35. The RMSE values of GEFS-Aerosols are 0.18, 0.15, 0.10, and 0.15, which are much lower than the NGACv2 values of 0.32, 0.34, 0.23, and 0.32. This suggests that GEFS-Aerosols better matches the observed temporal variation of total AOD than NGACv2. One peak in early August at the Monguinn site, one peak in mid-September at the Misampfu site, two peaks in early August and early September at the Maun Tower site, and one enhancement in August at Lubango are all predicted by the GEFS-Aerosols model. The ICAP forecasts show lower biases against the AERONET total AOD for predicting these peaks, while none of these peaks is captured by NGACv2. GEFS-Aerosols shows slight overpredictions in mid-July and late August for Gabon and early August for Lubango.
Day-1 AOD forecasts of GEFS-Aerosols, ICAP, and NGACv2 verified against AERONET sites in Africa during 5 July–30 November 2019.
Thirteen AERONET sites inside the major dust source regions of western northern Africa, Asia, and the Middle East and surrounding areas have available data from July to November 2019. The total AOD time series of GEFS-Aerosols, ICAP, and NGACv2 at six of these sites are shown in Fig. 11. Overall, the GEFS-Aerosols model is able to closely predict the observed total AOD variation, especially at the sites of Banizoumbu, Tenerife, Saada, Ben Salem, Granada, and Sede Boker, with much better performance than those of NGACv2 according to the correlation (RMSE) values in Table 1 of GEFS-Aerosols at 0.74 (0.15), 0.77 (0.07), 0.76 (0.14), 0.82 (0.07), 0.85 (0.09), and 0.73 (0.08) vs. NGACv2 at 0.33 (0.24), 0.25 (0.18), 0.25 (0.26), 0.25 (0.23), 0.21 (0.16), and 0.19 (0.14). In addition to NGACv2's overprediction at the sites of Ben Salem and Granada, it does not accurately capture observed temporal variations of total AOD at these sites.
Day-1 AOD forecasts of GEFS-Aerosols, ICAP, and NGACv2 verified against AERONET sites in dust source regions and surrounding downwind areas during 5 July–30 November 2019.
We compare the daily AERONET total AOD to the 1 d forecasts of total AOD from GEFS-Aerosols and NGACv2 at the AERONET sites of Cape Verde, Tamanrassett, and Tenerife located in the dust source region over northern Africa in Fig. 12. The slope of the linear regression of AERONET total AOD
vs. GEFS-Aerosols is quite different from that of NGACv2 at the site of
Tamanrassett, which is located in southern Algeria and in the middle of the
Saharan dust source region. The GEFS-Aerosols linear regression slope is
much closer to 1 than that of NGACv2, and the
Daily AERONET total AOD vs. modeled total AOD from GEFS-Aerosols (blue) and NGACv2 (orange) at the AERONET sites of
Figure 13 shows Day-1 predictions of total AOD time series by GEFS-Aerosols and NGACv2 compared against the MERRA-2 reanalysis averaged over nine major global regions from August 2019 to March 2020. The comparison clearly shows the consistency between GEFS-Aerosols and the MERRA-2 reanalysis over most of these nine regions, especially northern Africa, the North Atlantic, southern Africa, and the South Atlantic, with only minor discrepancies during these 8 months. The total AOD is dominated by dust in northern Africa and fire emissions in southern Africa. The aerosols emitted from dust and fire regions and their long-range transport play important roles in impacting the total AOD over the North and South Atlantic oceans. The good agreement with MERRA-2 shows that GEFS-Aerosols captures the emissions and transport of dust and fire emissions in these regions.
GEFS-Aerosols and NGACv2 Day-1 total AOD forecast time series against MERRA-2 reanalysis data averaged over major global regions of northern Africa (0–35
Total AOD variation in South America is mainly related to biomass burning emissions. GEFS-Aerosols has some slight low biases relative to MERRA-2 from mid-September to early October 2019 that are associated with the Amazon fire event. GEFS-Aerosols underpredicts MERRA-2 in this region from mid-November 2019 to March 2020, outside the main biomass burning season, which suggests that the GEFS-Aerosols AOD low biases in this region are mostly associated with sources other than fires.
The European region has the largest differences between GEFS-Aerosols and
MERRA-2 reanalysis total AOD among the nine regions. Although their temporal variations are similar, GEFS-Aerosols underpredicts the MERRA-2 total AOD throughout the whole period by a factor of 0.5. The large absolute low
biases from August to early October 2019 and March 2020 in Europe are associated with GEFS-Aerosols underestimates of sulfate AOD (Fig. 8).
From August to early December 2019, the GEFS-Aerosols total AOD looks quite
consistent with the MERRA-2 reanalysis on average across East Asia.
GEFS-Aerosols high biases starting in mid-December 2019 and increasing from January to March 2020 may be associated with the lockdown in China
during the Coronavirus disease 2019 (COVID-19) pandemic. Anthropogenic
emissions of NO
Both the eastern and western US regions exhibit GEFS-Aerosols low biases of about 5 %–30 %, with the largest differences in the eastern US occurring in August 2019. However, the trends of total AOD temporal variations, low in summer and high in winter, in the GEFS-Aerosols prediction and the MERRA-2 reanalysis are quite consistent over the eastern and western US. The minor underpredictions by GEFS-Aerosols need further investigation.
In comparison, the NGACv2 predictions show significant underprediction of MERRA-2 total AOD for almost all of these nine regions throughout this 8-month period. The one exception is northern Africa, where the NGACv2 results are close to the MERRA-2 reanalysis, with overprediction in August 2019 and low biases from December 2019 to March 2020. In addition to its general underprediction of MERRA-2 total AOD, NGACv2 is not able to capture the temporal variations of total AOD in some regions, such as the enhanced AOD due to fire emissions in southern Africa, the South Atlantic, and South America. Though NGACv2 shows similar temporal variations to MERRA-2 total AOD in Europe, East Asia, and the US, the magnitudes of NGACv2 predictions are too low, by a factor of 1 to 3. This analysis is consistent with a 1-year evaluation of GEFS-Aerosols AOD that shows improvements over NGACv2 (Bhattacharjee et al., 2022).
Retrospective simulations of GEFS-Aerosols and NGACv2 were performed for the summer of 2016 and evaluated using aircraft measurements from the first deployment of ATom-1 in July and August 2021. During ATom-1, plumes from dust storms and large biomass burning events and low-level sea-salt aerosols were observed over the South and central Atlantic, and anthropogenic pollution was observed over the United States on the last flight from Minnesota to southern California.
In this section, we evaluate the 24 h forecast skill of GEFS-Aerosols and
NGACv2 by comparing to ATom-1 observations. The GEFS-Aerosols and NGACv2 model results are sampled at the same latitude, longitude, and altitude as the ATom-1 measurements. The model output is hourly with
Figure 14 shows the tropospheric column sums of OC along the flight tracks
of the NASA DC-8 for the ATom-1 observations and GEFS-Aerosols model
experiments. The OC column sums using GBBEPx v3 fire emissions at
Tropospheric column sums of OC (
Results of the model–measurement comparisons for dust are shown in Fig. 15. GEFS-Aerosols simulations show good agreement with ATom-1 observations over the tropical North Atlantic and downwind of the western Africa dust source
region. However, the model underestimates the dust columns over the tropical South Atlantic, Greenland, and southeastern Canada while underestimating dust over the US, Alaska, and broad areas of the Pacific Ocean. The GEFS-Aerosols
model shows a clear enhancement of the dust event sampled on 17 August 2016 east of
the African coastline near 22
Tropospheric column sums of dust (
Table 3 gives median bias and correlation statistics for column sums of all
GEFS-Aerosols model cases as well as the NGACv2 dust forecasts for
ATom-1 and GEFS-Aerosols column sum statistics of mean bias and correlation for sulfate, OC, BC, and dust.
Dust, on the other hand, shows a slight underprediction in column amount in the model results. Dust sources depend critically on surface wind speed, have very little overlap with the anthropogenic and biomass burning sources of the other species, and are associated with areas of weather and surface conditions, all which may contribute to the different responses of dust emissions. Correlations of dust are also much lower than for other species, and there is a very obvious difference between GEFS-Aerosols and NGACv2 model forecast statistics, as discussed further below. We note that sea-salt columns are not calculated or compared to ATom-1 observations, due to the large number of observations below the detection limit, especially above 2 km altitude.
ATom-1 flight tracks are separated into two sections and labeled the “Pacific” side for 29 July to 8 August 2016 flights and the “Atlantic” side for 15 to 23 August 2016 flights. For this analysis the 1 s
model and observed data are binned into 10 equally spaced vertical intervals
(
Vertically resolved statistical comparisons of median values
(black line is ATom-1 observation, red line is GEFS-Aerosols),
Vertically resolved statistical comparisons of median values
(black line is ATom-1 observation, red line is GEFS-Aerosols, green line is
NGACv2),
Vertically resolved statistics of naturally occurring dust and sea salt are
shown in Fig. 17. For dust over the Pacific, median values of
GEFS-Aerosols are too low, while the NGACv2 results are too high compared to the observations and the correlations are almost less than 0.5. The performance of GEFS-Aerosols improves over the Atlantic, with median values
comparable to observations above
The ATom-1 flight profiles allow a more detailed comparison of aerosol
spatial patterns from different aerosol sources with the model. High values
of OC and BC from fires were observed on 15 and 17 August 2016 over the
Atlantic, as were high values of dust and sea salt. The flight track of height–latitude profiles of OC, BC, and sulfate for these combined days are
shown in Fig. 18 for the ATom-1 measurements and the model results. The
model results show similar patterns to ATom-1 in reproducing the profiles of OC even using a log scale, especially the biomass burning plumes near the
tropics, though the model results show slightly low biases. However, they also show some bias for OC at levels above 4–5 km over the North Atlantic, where model results show high biases. Overall, predicted BC
(middle column of Fig. 18) is able to capture the decreasing trend with
increasing altitude in the latitude–height profiles; however, they are underpredicted in the biomass burning plumes near the tropics from the
surface to 5 km height in both model experiments, which has been seen in other models due to insufficient wet scavenging (Wang et al., 2014; Choi et
al., 2020). Similar to the OC profiles, the model results overpredict above
the 4–5 km height levels. It appears that the model does not reproduce the enhancements of BC at 1–4 km height very well over this area. This may be
possibly due to relative weak convection or a low modeled injection height
where the fire emission has not been lifted enough to this altitude, which needs further studies. For sulfate (right column), the model experiments show
high concentrations at low altitude, similar to the observations, though
there are still some differences for the plume location at 2–4 km height
that shift the plume from near the Equator to near 20
Height–latitude profiles of OC, BC, and sulfate over the Atlantic on 15 and 17 August 2016 for
Height–latitude profiles of dust (
Figure 19 shows the comparisons of the naturally occurring dust and sea-salt aerosols for the same time period. In the left column of dust, we also
include the NGACv2 results. For more consistent comparisons, here the
modeled dust results are summed up by the first two bins to match the
observation particle size range (less than 3
Since the dynamical core of FV3 developed by the GFDL has been selected by the NOAA to be the dynamical core for the Next Generation Global Prediction System (NGGPS), development of a coupled weather and atmospheric chemical composition model for chemical weather and air quality forecasting based on the FV3 framework began a couple of years ago. The development as a single ensemble member of the Global Ensemble Forecast System (GEFS) has been completed. This new model, referred to as GEFS-Aerosols, was implemented as one member of the GEFS in operations as part of the NOAA's first coupled UFS model in September 2020 and replaced the previous operational global aerosol prediction system (NGACv2) at the NCEP.
The chemical component of atmospheric composition in GEFS-Aerosols is based
on WRF-Chem, which is a community modeling system used by thousands of users
worldwide. The aerosol modules are based on modules from the GOCART model.
Features of the new model include (1) the biomass burning plume rise module added from WRF-Chem, (2) the FENGSHA dust scheme implemented and developed by the NOAA Air Resources Laboratory (ARL), (3) all sub-grid-scale tracer transport and deposition handled inside the physics routines requiring consistent implementation of positive definite tracer transport and wet scavenging in
the SAS scheme, (4) the updated background fields of OH, H
The updates in anthropogenic and fire emission indicate that GEFS-Aerosols shows much better performance in matching the AOD observations when configured to use the CEDS anthropogenic emission and GBBEPx v3 fire emissions with plume rise module compared to NGACv2, especially over the fire source regions. For more extensive evaluation, we performed 9 months of Day-1 real-time forecast of GEFS-Aerosols starting in July 2019, and the predicted AOD was used to compare to the satellite observations from MODIS and VIIRS, reanalysis data of ICAP-MME and MERRA-2, AERONET observations, and the model predictions from MERRA-2 and NGACv2. Overall, GEFS-Aerosols indicates substantial improvement for both composition and variability of aerosol distributions over those from the currently operational global aerosol prediction system of NGACv2. Globally, GEFS-Aerosols-predicted biases with respect to MERRA-2 forecast for dust, OC, and sulfate AOD were improved compared to those from NGACv2. Substantial improvements were seen for the total AOD prediction when compared to MERRA-2 reanalysis during the period of July to November 2019. Though there are still some high biases over the southern African fire region and East Asia and low biases over South America and dust source regions, GEFS-Aerosols reproduces the prominent temporal and geographical features of AOD as represented by satellite observations and reanalysis data, like dust plumes over northern Africa and the Arabian Peninsula, biomass burning plumes in the Southern Hemisphere, South America, northwestern America and eastern Europe, polluted air over East and South Asia, and high-altitude sea-salt bands. We also sampled the forecast total AOD of GEFS-Aerosols and NGACv2 at the same location as 60 AERONET sites, which are spread globally and represent different aerosol regimes, and compared their variations for 5 July–30 November 2019. Much higher correlation coefficients against AERONET data are indicated for GEFS-Aerosols than those for NGACv2 globally and are quite comparable to those of the ICAP-MME.
During the biomass burning events, GEFS-Aerosols captured major fires over southern Africa, Siberia, the central Amazon, and central South America much better than NGACv2. Part of the improvement may be due to the vertical transport by the plume rise module. Generally, the total AOD time series of GEFS-Aerosols predictions matches closely to those of ICAP and AERONET during most of the time from July to November 2019 at the AERONET sites over South America, except that there are some minor underpredictions of several of the highest AOD episodes. In contrast, NGACv2 substantially underpredicted almost throughout the whole period and almost entirely missed many high AOD events. For the southern African event, the GEFS-Aerosols predictions are able to capture the daily total AOD variations seen in the AERONET observations, even better than that of the ICAP total AOD at the sites near the fire source regions, though there are overpredictions at the sites in downwind areas, which may be due to the lack of removal processes or uncertainties in fire emission in central and southern Africa. In contrast, the NGACv2 results show underprediction in total AOD forecast at most of the AERONET sites in this region.
Overall, the model-predicted total AOD variation by GEFS-Aerosols indicates much better performance than that of NGACv2 over western northern Africa. Although GEFS-Aerosols shows reductions in dust emissions over the Saharan dust source, the correlations with observations from downwind AERONET sites in western Africa are improved over those for NGACv2. The largest biases and discrepancies of GEFS-Aerosols and NGACv2 are both indicated in the sites in Tajikistan, which may be associated with a missing dust source near this site for both models. Obviously, other than the updates in anthropogenic and fire emissions, the implementation of the FENGSHA dust scheme in GEFS-Aerosols also shows great improvements in the dust concentration and AOD predictions over the dust regions compared to that of NGACv2, which used the original GOCART dust scheme.
We also evaluated predicted aerosol concentrations with different resolutions against the ATom-1 aircraft measurements from July to August 2016. Overall, predicted aerosol concentrations are quite comparable to the
ATom-1 measurements along the flight tracks globally with
This paper provides an overview of advances and challenges in model development for operational atmospheric aerosol predictions at the NOAA. This implementation advanced the global aerosol forecast capability for NOAA and made a step forward toward developing a global aerosol data assimilation system. Currently, the assimilation of AOD based on satellite observations is under development to constrain aerosol distributions in the GEFS-Aerosols model. Initial testing shows promise for improvement of predictions as well as limitations, indicating the need for refinements in quality control, data assimilation impacts on aerosol composition and vertical distribution, as well as bias correction of satellite observations, with bias and other errors substantially reduced in GEFS-Aerosols, especially when it is equipped with an aerosol data assimilation system. Currently, though the aerosol feedback from the aerosol components has not yet been included in the atmospheric model for direct and indirect radiative feedback, the model provides a good starting point from which to investigate at the impact on weather predictions out to sub-seasonal and seasonal scales when including the aerosol feedbacks in the atmospheric system in the future plan.
The GEFS-Aerosols v1 code and model configuration for aerosol forecast here
are available at
LZ and RM were the major developers of the GEFS-Aerosols model, including implementing and coupling the aerosol components to the FV3GFSv15 meteorological model. SAM helped to process the anthropogenic emission and background input data, provided suggestions during the development of GEFS-Aerosols, and evaluated the model performance with ATom-1 observations. SAM retired at the end of 2021. BB developed and implemented the FENGSHA dust scheme in GEFS-Aerosols. PSB helped to evaluate the GEFS-Aerosols real-time and operational predictions. GAG provided oversight of the model development. LZ and JH developed the workflow for GEFS-Aerosols prediction and worked with LP to perform and manage the real-time and retrospective forecasts. RA provided guidance on the implementation of the fire plume rise scheme. SK, XZ, and FL provided the GBBEPx v3 data. The other co-authors provided help, suggestions, and project management throughout the GEFS-Aerosols modeling system development. LZ prepared the manuscript with contributions from all the co-authors.
The contact author has declared that none of the authors has any competing interests.
The scientific results and conclusions as well as any views or opinions expressed herein are those of the authors and do not necessarily reflect the views of the NOAA or the Department of Commerce.Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Li Zhang, Raffaele Montuoro, Haiqin Li, and Ravan Ahmadov were supported by funding from NOAA GSL award no. NA17OAR4320101. This work was also supported by the UFS Research to Operations Medium Range Weather/Seasonal to Subseasonal Atmospheric Composition sub-project.
This research has been supported by the National Oceanic and Atmospheric Administration (grant no. NA17OAR4320101).
This paper was edited by Samuel Remy and reviewed by two anonymous referees.