The chemical lateral boundary conditions (CLBCs) are critical to the prediction accuracy of regional chemical transport models, particularly during intrusion events (Tang et al., 2009, 2021). The CLBCs represent the spatiotemporal distribution of chemical species along the lateral boundaries of the domain of a regional model. NACC-CMAQ uses methods described in Tang et al. (2021) and implements dynamic CLBCs (updated every 6 h) for dust and smoke aerosol data that are extracted (and mapped to CMAQ CB6-Aero7 species) from the NOAA operational global atmospheric aerosol model, known as the Global Ensemble Forecast-Aerosols (GEFS-Aerosols) member (Fig. 2). GEFS-Aerosols is also based on the FV3GFS dynamical core, which uses the Goddard Chemistry Aerosol Radiation and Transport (GOCART) model for its sulfate, dust, BC, OC, and sea salt aerosol predictions (Chin et al., 2000, 2002; Ginoux et al., 2001). GEFS-Aerosols uses the same wildfire smoke and windblown dust dataset and algorithms as in NACC-CMAQ. The operational version of GEFS-Aerosols is run by the NWS as a special unperturbed forecast of the Global Ensemble Forecast System version 12 (https://www.ncdc.noaa.gov/data-access/model-data/model-datasets/global-ensemble-forecast-system-gefs, last access: 5 April 2022), which provides an ensemble forecast product four times per day. Dynamic CLBCs capture the signals of aerosol intrusion events such as biomass burning or windblown dust plumes from outside the domain, which can improve the prediction accuracy of downstream O3 and PM2.5 concentrations at the surface (Tang et al., 2021).

Compared to NMMB used in the prior NAQFC, the GFSv16 model has lower actual TEMP2 in the east and southeast and parts of the northwest (Fig. 4a–d) but has higher TEMP2 in the central Great Plains, northern Great Plains, and parts of the western and southwestern US, with higher 10 m wind speeds (WSPD10) in these regions (Fig. 4i–l). GFSv16 is drier with widespread lower 2 m specific humidity (Q2; Fig. 4e–h) and lower cloud fractions (CFRAC) (Fig. 4m–p), higher solar radiation absorbed at the ground (GSW; Fig. 5a–d), lower longwave radiation absorbed at the ground (GLW; Fig. 5e–h), deeper planetary boundary layer height (PBLH; Fig. 5i–l), and generally more regions of increased precipitation (PRECIP; Fig. 5m–p). Differences in the CFRAC are (in part) impacted by differences in the model definition of cloud cover; NMMB uses a binary cloud cover definition at each grid point, while GFSv16 uses fractional cloud cover to calculate CFRAC. For stable conditions, the PBLH in the prior NAQFC is re-diagnosed based on the Troen and Mahrt (1986) incremental calculation of the bulk Richardson number (Rib) from the surface up to a height above the neutral buoyancy level (i.e., approaching the critical Richardson number, Ricrit) in the Asymmetric Convective Model v2 (ACM2) PBL scheme in CMAQ (Pleim 2007a, b). For unstable conditions, the re-diagnosed ACM2 uses a slightly different PBLH formulation based on first finding the convectively unstable mixing layer (zmix) and then defining the point where Rib=Ricrit for the entrainment layer above zmix. For both stable and unstable conditions, however, NACC-CMAQ directly uses the diagnosed PBLH from the turbulent kinetic energy (TKE)-based PBL scheme in GFSv16 (Table 1; Han and Bretherton, 2019), which is also based on the Troen and Mahrt (1986) incremental Rib formulation. Thus, NACC/GFSv16-CMAQ calculation is similar to the re-diagnosed ACM2 PBLH for nighttime-stable conditions (with slight differences in Ricrit values), while there exists some distinct differences in their daytime-unstable PBLH formulations and Ricrit calculations.

Consequently, the GFSv16 (NACC) and re-diagnosed ACM2 (prior NAQFC) diurnal PBLH patterns are similar at night; however, the GFSv16 PBLH is considerably higher than the prior NAQFC during the daytime for all regions in September and January (Figs. S4–S5).

The meteorological differences between GFSv16 and NMMB (Figs. 4–5) influence chemical predictions in CMAQ, which include a deeper daytime PBL and more precipitation that can effectively dilute the gaseous and aerosol concentrations for NACC-CMAQ in some regions across the CONUS. Areas of lower CFRAC and higher TEMP2 in GFSv16, however, will increase photolysis and daytime O3 formation in NACC-CMAQ in certain regions including the southern US and upper Great Plains. We note that although there are differences in the PBLH calculation methodologies between the prior NAQFC and NACC-CMAQ (particularly for the unstable daytime PBLH), the differences in near-surface meteorology (i.e., generally warmer and drier) conditions in the GFSv16 (Tables 2 and S2) also in part affect the differences in PBLH (Fig. 5i–l). These differences affect the pollutant mixing and dilution, and in part the resulting air quality predictions between the prior NAQFC and NACC-CMAQ (see Sect. 4.4 below).

Evaluation of the simulated day 1 (0–24 h) forecasted meteorology against the METAR network shows that GFSv16 generally has a higher positive TEMP2 (warmer) bias (Fig. 6) in the west and a CONUS-wide higher negative Q2 (dry) bias (Fig. 7) compared to prior NMMB (i.e., prior NAQFC) in both September and January.

There are regions of higher RMSE for T2 and Q2, and lower and degraded ACC (Figs. S7–S8) for GFSv16 compared to NMMB, especially in the southern and western CONUS regions during September. The spatial patterns and magnitudes of WSPD10 bias and error are similar between GFSv16 and NMMB (Fig. 8); however, the higher WSPD10 for GFSv16 in the southern and central CONUS leads to a shift from negative to positive biases from Texas northward to North Dakota, especially during September. The WSPD10 RMSE is higher (Fig. 8) and the ACC is also lower/degraded (Fig. S9) for GFSv16 in those regions, but otherwise the GFSv16 and NMMB have similar performance for WSPD10. The day 1 forecast model performance (MB, RMSE, and ACC) for 10 m wind direction (WDIR10) is similar between NMMB and GFSv16 in both September and January (Figs. S6 and S10).

Overall, the GFSv16 results are favorable for driving the advanced NACC-CMAQ system, with some areas of concern in the degraded TEMP2 and Q2 in the warmer and drier regions, particularly in the south and west CONUS during September. This roughly correlates with warmer/drier top-layer soil conditions in GFSv16 in these regions (Fig. S11), and thus land surface and soil data assimilation and model improvement in GFSv16 is an active area of focus at NOAA. The widespread dry bias in GFSv16 appears to be persistent, as an independent evaluation of August 2019 demonstrated a very similar spatial pattern and magnitude of Q2 underpredictions in the eastern half of CONUS compared to the METAR network (not shown).

The GFSv16-driven NACC-CMAQ system extends out to a 72 h forecast. Hence, there is a question of how the day 1 and 2 forecasts perform for NMMB vs. GFSv16 in the eastern (<100∘ W) and western (>100∘ W) US and how a day 3 forecast extension also affects the GFSv16 diurnal and statistical model performance. The GFSv16/NACC diurnal patterns of standard deviation, error, and bias for TEMP2, Q2, and WSPD10 are very similar to each other for days 1–3 (Figs. S12–S14). While there is a slight increase in error and decreased correlation (R), the relevant statistical metrics (e.g., MB, NMB, RMSE, and R) do not change appreciably from day 1 to 3 for both September and January (Tables S3–S4). This lends confidence in the utility of using the updated GFSv16 meteorology to drive a 72 h air quality forecast in NACC-CMAQ.

The day 1 diurnal statistics highlight both similar and contrasting TEMP2 and Q2 patterns for NMMB vs. GFSv16 in the eastern and western CONUS (Figs. S12–S13). In September (Fig. S12a), NMMB has higher error and positive TEMP2 (i.e., warm) bias in eastern CONUS during morning hours and lower error with a slight cool bias in the afternoon and evening, while GFSv16 shows slightly overpredicted TEMP2 during most hours of the day in the east. Over the western CONUS, there are larger diurnal TEMP2 differences that include small oscillating TEMP2 biases (about zero) for NMMB, along with distinctly large warm biases during all daytime hours for GFSv16 in the west. There are larger error and negative Q2 (i.e., drier) biases for GFVSv16 compared to NMMB in eastern and western CONUS (Fig. S13a). In January, the TEMP2 and Q2 diurnal statistical patterns are similar for NMMB and GFSv16 in both the eastern and western CONUS; however, the GFSv16 daytime hours have slightly higher error and warmer and drier biases compared to NMMB (Figs. S12b and S13b).

The total PRECIP is generally higher in GFSv16 compared to NMMB toward the east (Fig. 5), which leads to larger overpredictions on average in the CONUS compared to PRISM (Fig. 9). GFSv16 has a positive PRECIP bias on average in the CONUS, NMMB has a negative bias, and there is a relatively large difference in the spatial patterns between NMMB and GFSv16 for September compared to January. The difference is impacted by higher convective activity during late summer and early fall in September compared to winter in January (not shown). Further analysis indicated that generally heavier PRECIP in GFSv16 reduces the predicted PM2.5 concentrations via wet deposition (not shown) in the east and southeast and in parts of the west and northwest compared to NMMB.

Comparisons of the model vertical profile statistics (i.e., MB, RMSE, and IOA) for TEMP, RH, and WSPD against an average of select RAOB observations across the CONUS indicate that the GFSv16 (NACC) performs consistently with the operational NMMB (NAQFC) column (Fig. 10; IOA nearly identical at ∼0.8–0.9). GFSv16 is warmer and drier than NMMB in the model layers near the surface (>850 mb), especially in September; however, GFSv16 has a moister atmospheric column with higher wind speeds compared to NMMB above the surface and in the free troposphere (<850 mb). Figures S15–S17 show the spatial variability across the different RAOB sites used in the average for Fig. 10. Analysis of the column (1000–250 hPa) average for all CONUS RAOB sites across CONUS indicate that GFSv16 has a predominantly cooler and moister atmospheric column in September, despite being strongly warmer and drier near the surface (Figs. S18–S19).

The updated NEIC2016v1 emissions in NACC-CMAQ are lower compared to the NEI2014v2 emissions used in the operational NAQFC for all major species, except for NH3 (Table 2), as the NEIC2016v1 includes updated data sources and model projections that have generally decreasing emissions compared to the NEI2014v2 (NEIC, 2019).

The spatial emission changes show widespread decreases in the 2-D area and mobile emissions near the major urban cities for CO and NOx and across the major interstates and railways for NOx (Fig. 11a–b).

The spatial variability in NOx emission changes, however, are impacted by changes in a number of on-road inputs including vehicles miles traveled, age distribution, and speeds, which caused some emissions to go up or go down depending on the specific counties. The NOx emissions variability is also impacted by national increases in railway levels and fuel use, while at the same time being impacted by changes to fuel efficiency and cleaner engines for both passenger and commuter trains. There are relatively minor area and mobile changes in SO2 (Fig. 11c), with some exceptions in the east-northeast; however, there are widespread increases in NH3 emissions driven by changes to the livestock counts and updated fertilization methods and inputs found in the NEIC2016v1 (Fig. 11d). Changes in non-point oil and gas production, exploration, and emission factors generation, as well as changes to updated activity and data sources for commercial cooking, residential fuel combustion, and industrial/commercial/institutional (ICI) fuel combustion impact the anthropogenic VOC (AVOC) area emission changes (Fig. 11e). The widespread and spatially consistent decreases in particulate organic carbon (carbon only) ≤ 2.5 μg (POC) and PMC (defined as coarse PM > 2.5 μg and ≤ 10 μg) are due to decreasing fugitive dust sources (Fig. 11f and h), with the exception of the St. Lawrence River valley, that has both increases in POC and AVOC (e.g., formaldehyde; not shown) emissions in the NEIC2016v1. Updated appliance counts and residential wood combustion estimates affect the particulate elemental carbon ≤ 2.5 μg (PEC) area emission decreases (Fig. 11g).

There are also biogenic emissions differences due to the updated inline BEISv3.6.1 and BELD5 in NACC-CMAQ (Table 2) and due to the impacts of NMMB (prior NAQFC) vs. GFSv16 (NACC) meteorology on BEIS calculations (Fig. 12).

The lower GFSv16 temperatures near many of the highly vegetated regions of the CONUS in September (Fig. 4b) decrease the isoprene (ISOP) and terpene (TERP) emissions, with some notable localized ISOP emission increases due to larger relative increases in downward solar radiation at the surface (GSW; Fig. 5b) and resulting photosynthetic active radiation (PAR; not shown). The differences are also impacted by the derivations of leaf temperatures in the updated BEISv3.6.1 and BELD5 in NACC-CMAQ compared to the BEISv3.14 and BELD3 in the prior NAQFC (see discussion in Sect. 2.2). Hence, the differences in spatial variability between ISOP and TERP emission changes stem from both differences in the locations of their relative maxima and from the different algorithms for temperature and light dependencies in BEIS. The GFSv16 (NACC) performs very similarly to NMMB (prior NAQFC) for GSW at the surface compared against BSRN-SURFRAD observations in the CONUS, with a slightly larger overprediction in the late afternoon at some sites (Figs. S21 and S22). The relatively low ISOP and TERP emissions in NACC-CMAQ will effectively lower the ground-level O3 and contribution of secondary organic aerosol (SOA) formation to PM2.5 compared to the prior NAQFC, particularly in the southeast and parts of the western CONUS in the late summer and early fall.

Here we focus on analysis of NACC-CMAQ predictions of gaseous O3 for the late summer and early fall (September 2020) and PM2.5 concentrations during the winter (January 2021) as concentrations are relatively high for the pollutant's respective seasons. During the late US ozone season in September 2020, a large majority of the local NOx concentration increases in NACC-CMAQ (Fig. 13a–b) correlate with areas of NOx emissions increases in the NEIC2016v1 compared to the NEI2014v2 (Fig. 11b). An exception is the large NOx increases in the far west (e.g., California and Oregon) that stem from gaseous NOx emissions from strong wildfires that are captured by the GBBEPx in NACC-CMAQ (Table 1) but are excluded from the prior NAQFC wildfire emissions system (Table S2).

The increases in NOx concentrations and enhanced nighttime O3 titration, widespread decreases in total VOC concentrations due to both anthropogenic and biogenic VOC emission decreases in NACC-CMAQ, GFSv16-meteorology impacts (e.g., higher PBLH), and updated CMAQv5.3.1 chemistry and transport lead to widespread decreases in hourly O3 when averaged over all hours (Fig. 13e–f). Regions of higher NOx emissions, overall drier (i.e., widespread lower Q2) conditions, and stronger mid- to late-afternoon solar radiation reaching the surface (i.e., widespread lower CFRAC) (see Figs. 4–5 and S21–22) lead to enhanced daytime O3 formation, which is shown in the widespread increases in the maximum daily 8 h average (MDA8) O3 for NACC-CMAQ (Fig. 13g–h). This is particularly true for the strongly NOx-limited conditions across much of the western CONUS, where the MDA8 O3 increases are impacted by large increases in wildfire NOx emissions in GBBEPx and VOC decreases (anthropogenic+biogenic but no wildfire VOC emission impacts) in NACC-CMAQ. These effects subsequently impact the ozone NOx-VOC sensitivity regime that enhances the NOx-saturated (i.e., VOC-limited) conditions in this case (Fig. S24). There are exceptions, with MDA8 O3 decreases in the west, including western Oregon, the San Joaquin Valley in California, and regions of the southwestern CONUS, all of which are strongly VOC-limited (Fig. S24). These regions are further impacted by the VOC decreases and further NOx saturation from wildfire emissions in some locations of the west. Although outside the scope of this work, we also found that the NACC/GFSv16-CMAQ system yields reasonable results when comparing fire-enhanced O3 and PM2.5 concentrations to aircraft measurements during the 2019 Fire Influence on Regional to Global Environments and Air Quality (FIREX-AQ) field campaign (https://csl.noaa.gov/projects/firex-aq/, last access: 5 April 2022) (not shown). The widespread decreases in both the hourly and MDA8 O3 over all oceanic regions in the domain are driven by the updated halogen (e.g., bromine and iodine chemistry) mediated O3 loss in NACC-CMAQ, which can reduce annual mean surface ozone over seawater by 25 % (Sarwar et al., 2019).

There are both relatively large increases (north, northeast, and west) and decreases (south, southeast, and parts of the west) for winter (January 2021) total PM2.5 (PM25_TOT) in the CONUS for NACC-CMAQ compared to NAQFC (Fig. 13i–j). The decreases in inorganic PM25_TOT in the east and southeast are dominated by decreases in particulate sulfate (PM25_SO4) and ammonium (PM25_NH4), while the increases in the northern central and eastern CONUS are driven by increases in particulate nitrate (PM25_NO3) and PM25_NH4. Further analysis indicates that the widespread decreases in PM25_SO4 (strongest in the east) are driven strongly by widespread lower CFRAC in GFSv16 (Fig. 4o–p) and lower aqueous-phase oxidation in CMAQ (not shown). There are also contributions from decreased SO2 emissions found in some CONUS regions for NACC-CMAQ (e.g., the northeast; Fig. 11c). Additional consumption of inorganic sulfate as secondary isoprene epoxydiol (IEPOX) organosulfates are formed in the updated AERO7 aerosol mechanism in NACC-CMAQ (Table 1; Pye et al., 2013, 2017), and these further contribute to the PM25_SO4 decreases. The higher total PRECIP for NACC-CMAQ (Fig. 5) also leads to lower PM25_TOT in the eastern and southeastern regions.

The largest PM25_TOT increases in the northern central CONUS are primarily driven by enhanced ammonium nitrate formation, PM25_NO3, and PM25_NH4, which are influenced by increases in NH3 emissions (Fig. 11) and the inclusion of BIDI-NH3 fluxes in NACC-CMAQ (Table 1). BIDI-NH3 in NACC-CMAQ allows for inline calculation of the diurnal pattern of both NH3 evasion (emission) and deposition, while the prior NAQFC only includes deposition. Consequently, BIDI-NH3 in NACC-CMAQ generally increases ambient NH4+ and NO3- aerosol concentrations (Bash et al., 2013; Pleim et al., 2019) compared to the prior NAQFC.

There are also contributions to the increased PM25_TOT from organic carbon sources (Fig. S25; PM25_OC), especially in the northeastern portion of the domain that include the St. Lawrence River valley region. This is in part due to enhanced anthropogenic VOC emissions in NEIC2016v1 (Fig. 11e, e.g., formaldehyde; not shown) and enhanced AERO7 secondary organic aerosol formation in this region for NACC-CMAQ (not shown). There are also small PM25_EC contributions to the PM25_TOT decreases in the east and increases in the west for NACC-CMAQ (Fig. S25), which are mainly due to decreases in anthropogenic PEC emissions in the east (Fig. 11g) but also stem from contributions of relatively small GBBEPx PM emissions in the west (not shown). The prior NAQFC does not include biomass burning smoke emissions during the month of January.

Evaluation of NACC-CMAQ shows overall improvement in the spatial MB of hourly O3 (September) and PM2.5 (January) against the AirNow network across CONUS (Fig. 14). There are clear reductions in the NAQFC overpredictions of O3 and PM2.5 in the east, and overall reduction in NME, and overall improved correlation (R) and IOA for NACC-CMAQ. There are also reduced overpredictions in the west for O3 in September. The shifts to lower concentrations result in larger domain-wide average PM2.5 underpredictions for NACC-CMAQ compared to the prior NAQFC (cf. Fig. 13 above); however, the improvements in R and IOA for NACC-CMAQ are substantial. The MDA8 O3 spatial MB evaluation against AirNow behaves similarly to NAQFC, with slight degradation in the model performance statistics because of areas of higher overpredictions in the eastern US due to reasons discussed above for enhanced daytime O3 formation in NACC-CMAQ (Fig. S26).

The day 2 forecasts have similar spatial model performance and statistics, with improved hourly O3 and PM2.5 model performance (Fig. S27) and slightly higher MDA8 O3 overpredictions in the east for NACC-CMAQ (Fig. S28). The consistent model performance for day 3 also shows utility in extending to 72 h air quality forecasts in the advanced NACC-CMAQ system (Figs. S29–S30). There is, however, a more notable degradation in skill for the day 3 forecast of PM2.5 compared to O3 in NACC-CMAQ (compare Figs. 14 and S29).

There is significant improvement in the average O3 and PM2.5 diurnal patterns for each CONUS region other than higher daytime O3 peaks for NACC-CMAQ compared to prior NAQFC (Fig. 15a–i). This is reflected in the improved R and IOA over the CONUS on average for NACC-CMAQ (Fig. 14a–b). There is improved day-to-night O3 transition, i.e., a sharper slope or cutoff of daytime O3 formation, which leads to lower nighttime O3 mixing ratios in NACC-CMAQ that agree better with AirNow observations for all CONUS regions.

The NACC-CMAQ PM2.5 diurnal pattern is also more consistent with AirNow for most CONUS regions (Fig. 15k–t), which is supported by improved R and IOA (Fig. 14c–d). There are, however, some regions (e.g., the northeast, south, and northwest) that the prior NAQFC shows better diurnal performance in this case.

Overall performance evaluations of hourly O3 in each CONUS region show predominantly improved statistics for NACC-CMAQ, with increased R and IOA for all regions (Table 3). Comparisons of the NMB, NME, and R against statistical benchmark values for photochemical models based on Emery et al. (2017) indicate that both the prior NAQFC and NACC-CMAQ are within specified criteria for hourly O3 in most regions, except for relatively large NMB values in the west and northwest regions. The increased hourly O3 underpredictions in NACC-CMAQ degrades the NMB to fail to meet the benchmark in the west but improves the NMB to fall within criteria in the northwest region.

The higher MDA8 O3 in NACC-CMAQ degrades its regional NMB, NME, and R performance slightly compared to the prior NAQFC (Table 4), but R and IOA illustrate improvements for most regions, in some cases substantially for R (e.g., northeast, southeast, the upper Midwest, and the central Great Plains). The higher daytime O3 overpredictions by NACC-CMAQ in much of CONUS result in higher NMB and NME values that fall outside of the Emery et al. (2017) benchmark criteria. These remain a concern for both the prior NAQFC and NACC-CMAQ, and efforts are underway to address the persistent daytime O3 overprediction in the summer, particularly in the eastern US (see Fig. 14a–b and further discussion in Sect. 5).

There are substantial improvements in the overall statistical PM2.5 performance for NACC-CMAQ, especially for R and IOA in most CONUS regions. In many regions where the prior NAQFC falls outside of photochemical criteria values (Emery et al., 2017), NACC-CMAQ shows significant improvement to fall within the criteria. This demonstrates a substantial improvement in the accuracy of the NACC-CMAQ system for PM2.5 predictions (outside of major wildfires), attributed to the scientific advancements described above.

The day 2 forecast comparisons of the prior NAQFC and NACC-CMAQ regional statistics are similar to day 1, and the day 3 forecast extension for NACC-CMAQ has utility as its O3 and PM2.5 statistics predominantly fall within the benchmark criteria in most regions (Tables S5–S10).