Evaluation of Aerosol Optical Depth Forecasts From NOAA'S Global Aerosol Forecast Model (gefs-aerosols)

Abstract

Abstract The NWS/NCEP recently implemented a new global deterministic aerosol forecast model named the Global Ensemble Forecast Systems Aerosols (GEFS-Aerosols), which is based on the Finite Volume version 3 GFS (FV3GFS). It replaced the operational NOAA Environmental Modeling System (NEMS) GFS Aerosol Component version 2 (NGACv2), which was based on a global spectral model (GSM). GEFS-Aerosols uses aerosol modules from the GOCART previously integrated in the WRF Model with Chemistry (WRF-Chem), FENGSHA dust scheme, and several other updates. In this study, we have extensively evaluated aerosol optical depth (AOD) forecasts from GEFS-Aerosols against various observations over a timespan longer than one year (2019–20). The total AOD improvement (in terms of seasonal mean) in GEFS-Aerosols is about 40% compared to NGACv2 in the fall and winter season of 2019. In terms of aerosol species, the biggest improvement came from the enhanced representation of biomass burning aerosol species as GEFS-Aerosols is able to capture more fire events in southern Africa, South America, and Asia than its predecessor. Dust AODs reproduce the seasonal variation over Africa and the Middle East. We have found that correlation of total AOD over large regions of the globe remains consistent for forecast days 3–5. However, we have found that GEFS-Aerosols generates some systematic positive biases for organic carbon AOD near biomass burning regions and sulfate AOD over prediction over East Asia. The addition of a data assimilation capability to GEFS-Aerosols in the near future is expected to address these biases and provide a positive impact to aerosol forecasts by the model. Significance Statement The purpose of this study is to quantify improvements associated with the newly implemented global aerosol forecast model at NWS/NCEP. The monthly and seasonal variations of AOD forecasts of various aerosol regimes are overall consistent with the observations. Our results provide a guide to downstream regional air quality models like CMAQ that will use GEFS-Aerosols to provide lateral boundary conditions. Introduction Interactions of atmospheric aerosols with clouds and radiation are the largest source of uncertainty in modeling efforts to quantify current climate and predict climate change (IPCC 2021). Their influence depends in part on the concentration, composition, sizes, optical properties, and vertical distribution of the aerosol, which are influenced by emission, deposition, chemical reactions, and transport. Tropospheric aerosols arise from natural sources, such as wind erosion dust, biomass burning aerosols, sea spray, and volcanoes; and from anthropogenic activities, such as combustion of fuels, power plants, and residencies. Anthropogenic emissions leading to atmospheric aerosol have increased dramatically over the past century and have been implicated in human health effects (Kelly and Fussell 2015), in visibility reduction in urban areas, in acid deposition, and perturbing Earth’s radiation balance (Seinfeld and Pandis 1998). Previous studies have unveiled the importance of aerosol research and demonstrated the direct and indirect contributions of aerosols on climate change from regional to global scale (Twomey 1974; Chylek and Wong 1995; Menon et al. 2002; Huang et al. 2014). Despite many aerosol studies, estimates of aerosol radiative forcing are still one of the largest uncertainties in model simulations for global climate projection (Boucher et al. 2013; Bellouin et al. 2020). There is large temporal and spatial variability in global aerosol composition and aerosol sources, which is one of the key factors resulting in such large uncertainty in model radiative forcing and its climatic effect (Myhre et al. 2013). In parallel to aerosol simulation in climate models, the weather effects of aerosols have been relatively less considered in weather prediction models because weather forecasting prioritizes issues like cloud and moisture for fast and accurate predictions, whereas simulating prognostic aerosols requires considerable computational resources and challenging data assimilation of aerosols fields. Improvements in numerical weather prediction (NWP) were reported by Tompkins et al. (2005) who found that an updated dust climatology led to a northward shift of the southern African easterly jets (AEJ-S) in the European Centre for Medium-Range Weather Forecasts (ECMWF) NWP model. Results from NASA’s Goddard Earth Observing System (GEOS-5) forecasting system showed that the net impact of the interactive aerosol associated with a strong Saharan dust outbreak resulted in a temperature enhancement at the lofted dust level and a reduction near the surface levels, which improved forecasts of the AEJ (Reale et al. 2011). Grell et al. (2011) showed that coupling aerosols to radiation and microphysics schemes in high-resolution weather forecasting models may improve forecasts of temperature and wind during a significant wildfire event in Alaska. Toll et al. (2016) showed considerable improvement in forecasts of near-surface conditions during Russian wildfires in the summer of 2010 by including the direct radiative effect of realistic aerosol distributions. Considering the significant influence of the aerosol–radiation interaction on weather meteorological forecasts as illustrated in above studies and other studies (Zhang et al. 2010; Mulcahy et al. 2014). Several weather forecast centers have started to facilitate the inclusion of aerosol predictions in operational NWP models before including the aerosol feedback. NASA GEOS-5 started to provide near-real-time forecasts of aerosols and atmospheric compositions (Rienecker et al. 2008; Molod et al. 2015) and the ECMWF began to include global aerosol forecasts since 2008 (Hollingsworth et al. 2008; Morcrette et al. 2008; Benedetti et al. 2011).Three different gas–aerosol chemistry schemes were implemented in a two-way fully inline coupled global weather-chemistry prediction model (FIM-Chem) developed at NOAA Global Systems Laboratory (GSL), which showed good performance in forecasting the chemical composition for both aerosol and gas-phase species when compared with the observations without turning on the aerosols feedback (Zhang et al. 2022b). The Global Forecast System (GFS) is the cornerstone of the operational production suite used for numerical guidance at National Centers for Environmental Prediction (NCEP). In collaboration with NASA/Goddard Space Flight Centre (GSFC), NCEP developed NOAA Environmental Modeling System (NEMS) GFS Aerosol Component (NGAC) for predicting the distribution of global atmospheric aerosols. The model became operational in 2016 with dust-only forecasting capability (NGACv1) (Lu et al. 2016) and all species in 2018 (NGACv2) (Wang et al. 2018). NGAC was built upon Earth System Modeling Framework (ESMF) and used the NEMS Global Spectral Model (GSM) as the atmospheric model. Since the 2019 implementation, the atmospheric forecast model used in the GFS consists of the Geophysical Fluid Dynamics Laboratory (GFDL) Finite Volume Cubed-Sphere dynamical core (FV3) and several physics updates associated with it. An aerosol model component was coupled online with the FV3 Global Forecast System (FV3GFS) for global aerosol prediction at the NCEP since September 2020 as one of the ensemble members of the Global Ensemble Forecast System (GEFS), named as GEFS-Aerosols v1 (Zhang et al. 2022a). However, in GFS, the aerosol attenuation coefficients are still determined from prescribed aerosol distributions based on a global climatological aerosol database (Hess et al. 1998) and aerosol indirect effects on clouds and precipitation formation are not accounted for. NOAA’s National Weather Service (NWS) is on its way to deploy various operational prediction applications using the Unified Forecast System (UFS, https://ufscommunity.org/), a community-based coupled, comprehensive Earth modeling system. Including the prognostic aerosols from online coupled chemical model to represent real-time predicted direct and semi-direct aerosol radiative effects is the target of the next-generation of global modeling systems developing at EMC and NCEP. This requires more realistic forecasting of aerosol optical properties to generate a more realistic forecast of aerosol feedback. Better understanding of the model performance in the current operational global aerosol forecast system, GEFS-Aerosols, would help to provide more information for potential improvements. A broader and more detailed evaluation from different aspects can help to improve the model performance significantly before including the aerosol feedback in the next step. This in turn will also help us to understand the possible impact of aerosol on weather in the future. The paper is organized as follows. Section 2 provides a general description of GEFS-Aerosols, its available aerosol products, and data that are used in this study. Section 3 presents satellite, ground station, and analysis datasets that are used to validate the model forecasts. The evaluation of AOD against observational datasets followed by some case studies are described in section 4. Finally, section 5 includes a discussion and the conclusions. 2. Model description NCEP has partnered with NOAA/Earth System Research Laboratories (ESRL) Global Systems Laboratory (GSL), Chemical Sciences Laboratory (CSL), NOAA/Oceanic and Atmospheric Research (OAR) Air Resources Laboratory (ARL), the NOAA National Environmental Satellite, Data, and Information Service (NESDIS) Center for Satellite Applications and Research (STAR) and NASA GSFC to develop a global aerosol model that replaced the operational NEMS GFS Aerosol Component version 2 (NGACv2) (Wang et al. 2018). The major difference between GEFS-Aerosols model and NGACv2 is not only in the chemical model part. The dynamical core, model physics, resolution, microphysics scheme, land surface model are completely different between these two models. In Zhang et al. (2022a), Table 2 summarizes the comparison of model configurations between GEFS-Aerosols and NGACv2. The new model was included as a single member named GEFS-Aerosols in Version 12 of the Global Ensemble Forecast System (GEFS) (Zhou et al. 2022). The meteorology of this new model is based on version 15 of the Global Forecast System (GFS v15) with 64 vertical levels while the aerosol modules are based on the Weather Research and Forecasting model with Chemistry (WRF-Chem) version of the Goddard Chemistry Aerosol Radiation and Transport (GOCART) model (Chin et al. 2002; Colarco et al. 2010) GEFS-Aerosols treats the sources, sinks, and chemistry of 15 externally mixed aerosol species: dust (five non-interacting size bins), sea salt (five non-interacting size bins), hydrophobic and hydrophilic black and organic carbon (BC and OC, respectively; four tracers), and sulfate (SO4). Dust and sea salt have emissions dependent on frictional velocity and wind speed, respectively. Primary sulfate and carbonaceous aerosol species incorporate emissions principally from anthropogenic and biomass burning sources, with additional biogenic sources of organic carbon. Secondary sources of sulfate include chemical oxidation of sulfur dioxide gas (SO2) and dimethyl sulfide (DMS). Loss processes for all aerosols include dry deposition (with gravitational settling), large-scale wet removal, and convective wet scavenging. Recent updates and additions include a biomass burning plume rise module; tracer convective transport and wet scavenging implemented in the simplified Arakawa–Schubert (SAS) convection scheme (Han and Pan 2011); the FENGSHA dust scheme implemented and developed at ARL (Dong et al. 2016). The FENGSHA dust scheme in GEFS-Aerosols uses three land-use types (barren land, shrub–grassland, and cropland) as potential erodible dust sources instead of dust source maps, and the dust vertical flux was calculated according to a modified Owen’s equation (Owen 1964) when the friction velocity ( u∗) exceeded the threshold friction velocity (u*t) (which was set as a constant value for each potential erodible land-use type and soil texture based on literature and field measurements) (Dong et al. 2016). Biomass-burning emission is based on the real-time Blended Global Biomass Burning Emissions Product (GBBEPx V3) emission, and fire radiative power (FRP) data are provided by NESDIS (Zhang et al. 2012). There is an unavoidable time delay in the input of fire emissions to the operational run since they are from the real-time satellite observation. The 2-day temporal lag fire emissions are used in the operational configuration and we used the same setting in the retrospective runs in this study. We have used the Community Emissions Database System (CEDS) based on 2014 inventory for anthropogenic emissions (Hoesly et al. 2018). It should be noted that CEDS2014 anthropogenic emissions data may not be able to catch up the date of real-time forecast, considering the decreasing emission trend over China, the CEDS 2014 anthropogenic would result in some overprediction after 2014 (Zhang et al. 2022a). The model uses biomass burning plume rise module from WRF-Chem (Freitas et al. 2007; Freitas et al. 2010) with updates as in the High-Resolution Rapid Refresh (HRRR)-Smoke model (Ahmadov et al. 2017). Subgrid scale tracer transport and deposition is handled inside the physics routines requiring consistent implementation of positive definite tracer transport and wet scavenging in the SAS scheme, the updated background fields of OH, H2O2, and NO3 from Global Modeling Initiative (GMI) model and global anthropogenic emission inventories derived from CEDS 2014 inventory. No diurnal variation of fire emissions are provided as input. More detailed description of GEFS-Aerosols, its configuration, AOD calculation can be found in Zhang et al. (2022a). GEFS-Aerosols provides a 5-day forecast of total aerosol as well as dust, OC, BC, sea salt, and sulfate aerosol at C384 grid, which is converted to ∼0.25° × 0.25° by the Unified Post Processor (UPP) 4 times per day (at 0000, 0600, 1200, and 1800 UTC). The GFS data are used as the meteorological initialization in each cycle (24 h). 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. It provides AOD at seven different wavelengths such as 340, 440, 550, 650, 860, 1120, and 1640 nm. It also provides particulate matter (PM2.5 and PM10) forecasts along with three-dimensional mixing ratios of aerosol species at 64 model vertical levels. Fire emissions are updated every 24 h for each day of the run, while monthly CEDS emissions are included from the 2014 inventory. A 13-month retrospective run (only 0000 UTC cycle) has been conducted using GEFS-Aerosols to provide multispecies forecasts of aerosol optical depth (AOD) and other aerosol properties (such as single scattering albedo, angstrom exponent at different wavelengths). In this study we have used daily AOD forecast (from August 2019 to August 2020) at 550 nm (AOD550) from GEFS-Aerosols and evaluated model results against a set of observational datasets. Moreover, we have also evaluated 6-hourly forecasts of GEFS-Aerosols total and individual species (dust, OC, BC, sulfate, and sea salt). 3. Observation, reanalysis, and other model data We have used the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA2) (Gelaro et al. 2017) aerosol dataset to evaluate GEFS-Aerosols total AOD forecast. The dataset was produced using version 5.12.4 of the Goddard Earth Observing System Model (GEOS) Data Assimilation System (DAS). Gridded data are released at a 0.625° longitude × 0.5° latitude resolution on 72 sigma–pressure hybrid layers between the surface and 0.01 hPa (Buchard et al. 2017). MERRA-2 uses the Goddard Earth Observing System, version 5 (GEOS-5) Earth system model (Rienecker et al. 2008; Molod et al. 2015) and the three-dimensional variational data assimilation (3DVar) Gridpoint Statistical Interpolation analysis system (GSI) (Wu et al. 2002; Kleist et al. 2009). The GEOS-5 model is radiatively coupled to the GOCART aerosol module. GEOS-5 is driven by daily biomass burning emissions derived from Moderate Resolution Imaging Spectroradiometer (MODIS) FRP retrievals using the Quick Fire Emission Dataset (QFED) emissions (Darmenov and da Silva 2015). In MERRA-2, aerosol and meteorological observations are jointly assimilated within GEOS-5. The assimilation of AOD involves very careful cloud screening and homogenization of the observing system by means of a neural network based scheme that translates MODIS and Advanced Very High Resolution Radiometer (AVHRR) radiances into Aerosol Robotic Network (AERONET)-calibrated AOD (550 nm). The system also assimilates (non-bias-corrected) Multiangle Imaging SpectroRadiometer (MISR) 550-nm AOD over bright surfaces (albedo > 0.15) and surface-based AERONET AOD observations at 550 nm (Buchard et al. 2017). Although MERRA2 assimilates AOD from AERONET and other satellite sources, it still underestimates AOD over some regions (Frey et al. 2021). The underestimation is likely caused by cloud contamination, using older anthropogenic emission database, no treatment for nitrate particles in GOCART (Buchard et al. 2017). We have used MERRA2 AOD from NASA Global Modeling and Assimilation Office (GMAO) generated product “M2T1NXAER v5.12.4” (GMAO 2015; Bosilovich et al. 2016). We have obtained daily hourly MERRA2 analyzed AOD550 fields. AERONET (http://aeronet.gsfc.nasa.gov) is a ground-based and global-scale sun photometer network, which has been providing high-accuracy measurements of aerosol properties since 1990 (Holben et al. 1998). It is often used as the primary standard for validating satellite products and model simulations (e.g., Colarco et al. 2010; Levy et al. 2013). For this study, we use the quality-assured AERONET Version 3 level 1.5 product, which has better cloud-screening and better preservation of high AOD values that were often discarded in previous versions. The complete set of Version 3 cloud-screening and quality assurance algorithms and comparisons of the Version 3 product to Version 2 are provided in Sinyuk et al. (2020). For this analysis, we selected a number of AERONET sites based on the availability of contiguous data records from August 2019 to August 2020 and dominated by dust influence. Since AERONET instruments do not measure AOD550 directly, we have used AOD at 440 and 675 nm that are linearly interpolated on a log–log scale to provide 550-nm AOD. All AERONET data are sampled temporally at ±1 h of the daily GEFS-Aerosols forecasts (e.g., at any given location, AERONET measurements between 1100 and 1300 UTC are averaged to compare them to the 1200 UTC model forecast). The 2-h time window is created to ensure comparable temporal and spatial representativeness of the AEROENT data and model data. The Visible Infrared Imaging Radiometer Suite (VIIRS) sensor onboard the Suomi National Polar-Orbiting Partnership (SNPP and NOAA-20) satellite provides AOD550 at 0.25° × 0.25° horizontal resolution. In this study we have used daily gridded enterprise processing system (EPS) VIIRS AOD (Liu et al. 2014). We have also used daily 3-hourly VIIRS gridded data (at 0.25°) from NESDIS that are at GEFS-Aerosols model forecast lead times (e.g., 0, 3, 6, 9, …, 21 h) as well for case studies. We have used MODIS Collection 6.1 Level-3 (daily) AOD dataset (Levy et al. 2013) from the Aqua satellite to verify model forecast for case studies. MODIS data include AOD data based on refined retrieval algorithms, in particular, the expanded Deep Blue (DB) algorithm (Hsu et al. 2013; Sayer et al. 2013). We have used a merged AOD product, which combines retrievals from the Dark Target (DT) and DB algorithms to produce a consistent dataset covering a multitude of surface types ranging from oceans to bright deserts (Sayer et al. 2014). We have included NOAA’s previous operational global aerosol forecast model NGACv2 (Wang et al. 2018) AOD forecast in all our evaluations for comparisons against the new operational model. NGACv2 output AOD at T126 grid, which roughly corresponds to ∼1° × 1° resolution and became operational in May 2017 using the spectral version of GFS (v14). It uses an earlier version of the GOCART aerosol module, the relaxed Arakawa–Schubert (RAS) scheme, the original GOCART dust scheme and puts biomass burning emissions at the lowest model layer (compared to use of plume rise in GEFS-Aerosols). For quantitative comparison, model performances against observation dataset are evaluated using the Development Testbed Center (DTC) Model Evaluation Tools (MET) software package (https://www.dtcenter.org/community-code/model-evaluation-tools-met) (Brown et al. 2021). We use this tool to compute statistics for both point observation and gridded observations (reanalysis and satellite data). For gridded statistics (called “grid-stat” in MET) it accumulates AOD statistics separately for each horizontal grid location over time on matched forecast and observation pairs and computes both categorical and continuous type statistics. Continuous statistics are computed over the raw matched pair AOD values whereas categorical statistics are generally calculated by applying an AOD threshold to the forecast and observation values (Murphy and Winkler 1987). In this study, we have used three continuous statistics metrics: mean error or bias (Bias), root-mean-square-error (RMSE), and correlation coefficient (R) and three categorical statistics metrics: critical success index (CSI), false alarm ratio (FAR), and probability of detection (POD). Definitions of these terms are provided in appendix A. 4. Results a. Seasonal mean and bias of AOD We have divided the 13 months (August 2019–August 2020) of daily model forecasts used in this study into four seasons to understand aerosol characteristics in terms of its sources and transport from comparisons with observational data. Both GEFS-Aerosols and NGACv2 daily model forecasts of total AOD valid at (0000, 0600, 1200, and 1800 UTC) are averaged to compute daily means for each of the season’s analyses. We also averaged MERRA2 data at 6-h intervals to compute the daily mean for each season. VIIRS AOD data are daily gridded data with no retrievals of AOD outside of 65°S–65°N. Figure 1 shows day-1 forecasts of GEFS-Aerosols total AOD averaged over August–September–October (ASO) 2019 against MERRA2, NGACv2 and VIIRS. The model shows significant improvements compared to NGACv2 which underpredicts AOD from forest fire burning regions (mostly contributed by OC AOD in southern Africa, South America, Southeast Asia), anthropogenic emissions from East Asia (mostly contributed by sulfate and OC AOD) (Fig. 1b). Also, NGACv2 largely overpredicts dust from the North Africa source region compared to both reanalysis and satellite observation. GEFS-Aerosols is able to capture all the major aerosol events during this time period (Fig. 1a) that also include a large forest fire event in Siberia and dust transport from Taklamakan region. GEFS-Aerosols also simulates high AOD over northern India (AOD ∼ 0.5), which is largely contributed from post-monsoon agricultural fires in the northwestern part of the Indian subcontinent and transports further east (Sahu et al. 2021). Both MERRA2 and VIIRS show similar high AOD over northern India as well. However, we have found large overprediction biases of AOD over southern Africa in GEFS-Aerosols, which may be due to convection related transport of species and subsequent slow removal process.