NASA GEOS Composition Forecast Modeling System GEOS‐CF v1.0: Stratospheric Composition

Abstract

The NASA Goddard Earth Observing System (GEOS) Composition Forecast (GEOS‐CF) provides recent estimates and 5‐day forecasts of atmospheric composition to the public in near‐real time. To do this, the GEOS Earth system model is coupled with the GEOS‐Chem tropospheric‐stratospheric unified chemistry extension (UCX) to represent composition from the surface to the top of the GEOS atmosphere (0.01 hPa). The GEOS‐CF system is described, including updates made to the GEOS‐Chem UCX mechanism within GEOS‐CF for improved representation of stratospheric chemistry. Comparisons are made against balloon, lidar, and satellite observations for stratospheric composition, including measurements of ozone (O₃) and important nitrogen and chlorine species related to stratospheric O₃ recovery. The GEOS‐CF nudges the stratospheric O₃ toward the GEOS Forward Processing (GEOS FP) assimilated O₃ product; as a result the stratospheric O₃ in the GEOS‐CF historical estimate agrees well with observations. During abnormal dynamical and chemical environments such as the 2020 polar vortexes, the GEOS‐CF O₃ forecasts are more realistic than GEOS FP O₃ forecasts because of the inclusion of the complex GEOS‐Chem UCX stratospheric chemistry. Overall, the spatial patterns of the GEOS‐CF simulated concentrations of stratospheric composition agree well with satellite observations. However, there are notable biases—such as low NO _x and HNO₃ in the polar regions and generally low HCl throughout the stratosphere—and future improvements to the chemistry mechanism and emissions are discussed. GEOS‐CF is a new tool for the research community and instrument teams observing trace gases in the stratosphere and troposphere, providing near‐real‐time three‐dimensional gridded information on atmospheric composition.

Key Points

• Demonstrate the GEOS‐CF system is capable of supporting NASA science missions and applications which observe stratospheric composition

• The GEOS‐CF model produces realistic stratospheric ozone forecasts, a new capability during anomalous polar vortex conditions

• Spatial patterns of the GEOS‐CF simulated concentrations of stratospheric composition agree well with independent observations

1. Introduction

NASA's Global Modeling and Assimilation Office (GMAO) provides a suite of Goddard Earth Observing System (GEOS) model products to the public in near‐real time (analyses and forecasts) and with a month to two‐month latency (reanalysis; https://gmao.gsfc.nasa.gov/GMAO_products/). These products support NASA field missions and assess the impacts of NASA observations on environmental prediction. In order to support near‐real‐time NASA applications focused on atmospheric composition, it is essential GEOS has a realistic representation of stratospheric composition and chemistry (Nielsen et al., 2017). Ozone (O₃) is an important trace gas in the stratosphere where the total O₃ column acts to shield the Earth's surface from harmful ultra‐violet (UV) radiation, while at the surface it is harmful to human health and vegetation (Krzyzanowski & Cohen, 2008; Schlink et al., 2006). Since the total column O₃ (TCO) varies from day‐to‐day depending on stratospheric conditions, forecasting TCO is an important input for accurate surface UV forecasts (Turner et al., 2017). The discovery of the Antarctic “Ozone hole” nearly 40 years ago by ground‐based, sonde and satellite measurements (e.g., Farman et al., 1985; Solomon et al., 1986; Stolarski et al., 1986) indicated decreases in the ozone layer were greater than the 1% per decade that early models were predicting (Bhartia & McPeters, 2018). Tracking the recovery of the Antarctic ozone hole requires the sustained combination of high quality observations and models.

The GMAO has a mature data assimilation system (DAS) within GEOS to provide a realistic global three‐dimentional (3D) stratospheric O₃ product for the “satellite era” (since 1980) (Wargan et al., 2015, 2017; Wargan, Kramarova, et al., 2020; Wargan, Weir, et al., 2020) which can be used in analysis of stratospheric O₃ trends (Wargan et al., 2018). For 5–10 day TCO forecasting, GMAO's state‐of‐the‐science numerical weather prediction GEOS Forward Processing (GEOS FP; Lucchesi, 2018) system first assimilates near‐real time O₃ observations (Figure S1a in Supporting Information S1). However, the GEOS FP forecasts rely on simple parameterized chemistry based on fixed, pre‐calculated, monthly, latitude/altitude production and loss values as described in Nielsen et al. (2017). The GEOS FP forecasts fail when TCO is far from climatological values.

To further support the research community and NASA missions with atmospheric composition simulations, GMAO expanded the GEOS capabilities with the state‐of‐the‐science GEOS‐Chem chemistry transport model (CTM; Bey et al., 2001) fully integrated into GEOS (Hu et al., 2018; Keller et al., 2014; Long et al., 2015). Owing to the complexity of the chemistry, this coupled configuration (Figure S1b in Supporting Information S1) is run once daily and provides near‐real time estimates of recent atmospheric composition and detailed composition forecasts (“GEOS‐CF”) of the 3D state of the atmosphere on the same spatial (0.25°) resolution as the meteorology (Keller et al., 2021). For GEOS‐CF, the combination of the sophisticated GEOS‐Chem chemistry within a GEOS forecasting system allows for improved forecasting of TCO when far from climatological values and, for the first time, provides near‐real time 3D estimates of chemical species that are critical for understanding stratospheric O₃ recovery and loss, such as nitrogen oxides (NO _x ) and hydrogen chloride (HCl). This current study evaluating the GEOS‐CF stratospheric composition (100–1 hPa) is a companion paper to the GEOS‐CF description paper by Keller et al. (2021) which evaluated tropospheric composition and surface air quality forecast skill against independent observations.

The GEOS‐Chem CTM is a global 3D CTM driven by assimilated GEOS meteorological fields (Bey et al., 2001). It has an extensive community of developers and users worldwide (http://acmg.seas.harvard.edu/geos/). As the tropospheric chemistry became increasingly more sophisticated in GEOS‐Chem, the stratospheric chemical boundary condition became a limiting factor for stratosphere‐troposphere coupling analysis (Eastham et al., 2014). Over a similar time frame, the NASA Global Modeling Initiative (GMI) chemistry mechanism was developed at NASA Goddard Space Flight Center and is maintained to be state‐of‐the‐science for stratospheric chemistry modeling to support policy relevant assessments on stratospheric composition and O₃ recovery (e.g., Bucsela et al., 2013; Douglass et al., 1999, 2004; Duncan et al., 2007; Kinnison et al., 2001; Nielsen et al., 2017; Rotman et al., 2001; Strahan & Douglass, 2018). Using a version of the NASA GMI stratospheric chemistry mechanism, updated with the Jet Propulsion Laboratory (JPL)'s stratospheric recommendations for kinetic and photochemical data (JPL Publication 10‐06; Sander et al., 2011), Eastham et al. (2014) extended GEOS‐Chem to have the capability to run with a unified tropospheric and stratospheric chemistry mechanism, “UCX.” The GEOS‐Chem CTM has continued to evolve since the version 9 evaluated in Eastham et al. (2014), with updates which could impact stratospheric composition such as the treatment of halogen species (Chen et al., 2017; Schmidt et al., 2016; Sherwen, Evans, et al., 2016; Sherwen, Schmidt, et al., 2016). This present study benchmarks the stratospheric composition using a more recent version of GEOS‐Chem (version 12.0.1) run in an online high‐resolution global GEOS simulation (GEOS‐CF) to assess the readiness of GEOS‐CF output to support the research community and to prioritize needed improvements.

The paper follows with an overview of the GEOS‐CF version 1.0 (“v1”) system as described by Keller et al. (2021) in Section 2, followed by the description of the independent observations—those which do not constrain the GEOS‐CF constituent concentrations—in Section 3 that are used for validation of stratospheric composition in Sections 4, 5, 6. After initial evaluation of the GEOS‐CF v1 stratospheric composition, several updates were made to the UCX code mid‐production, which are outlined in Section 4. The evaluation of the updated system against ozonesondes, lidar and satellite observations for the year 2020 is presented in Section 5, with case studies of forecast skill in Section 6. Final summary and future developments are discussed in Section 7.

2. GEOS Composition Forecast (GEOS‐CF) System Description

The NASA GEOS‐CF system (Keller et al., 2021) is a near‐real time global 3D coupled chemistry and meteorology modeling system with the offline GEOS‐Chem CTM code fully integrated as a chemistry module in GEOS (Hu et al., 2018; Long et al., 2015). The GEOS‐Chem chemistry components are therefore the same in GEOS‐CF as in the offline CTM except the dynamics and turbulence schemes use the online GEOS meteorology instead of the offline transport scheme within the CTM. Briefly, the GEOS‐CF configuration has the GEOS atmospheric general circulation model (AGCM; Molod et al., 2015) one‐way coupled to the GEOS‐Chem chemistry module (Figure S1b in Supporting Information S1), run on a cube‐sphere horizontal grid at c360 resolution and on 72 GEOS hybrid‐eta model layers from the surface to 0.01 hPa, with output at the global resolution of 0.25° latitude × 0.25° longitude (GEOS‐CF v1; Keller et al., 2021).

The addition of stratospheric chemistry in GEOS‐Chem through UCX (Eastham et al., 2014) includes: stratospheric chemical reactions present in GMI, methane changed from a fixed background value to an advected tracer, and shorter wavelengths in the Fast‐JX photolysis solver. Also, surface boundary conditions for nitrous oxide (N₂O) and several ozone‐depleting substances (ODS; e.g., halogen bromine (Br) and chlorine (Cl) species) were defined following the World Meteorological Organization (WMO) 2006 ozone assessment (Daniel et al., 2006), and a surface boundary condition of 500 ppt was set for carbonyl sulphide (OCS), a precursor for sulfate aerosols (Weisenstein et al., 1997). Lastly, UCX includes treatments of stratospheric sulfate aerosols (Carslaw et al., 1997) and polar stratospheric clouds (PSCs; Kirner et al., 2011) on which heterogeneous reactions can occur and lead to seasonal O₃ depletion in the stratosphere. Additional details regarding the representation of stratospheric chemistry and physics are provided in Eastham et al. (2014). Modifications to UCX and boundary conditions used in this work are described in Section 4.

Since the GEOS‐CF configuration is computationally expensive due to the complexity of the chemistry, instead of running a full DAS, GEOS‐CF relies on GMAO's meteorological “replay” technique (Orbe et al., 2017), where the AGCM computes the increments for pressure, temperature, wind (U, V), specific humidity, aerosol optical depth and O₃ based on pre‐computed analysis fields from a previously run assimilation system (Figures S1a and S1b in Supporting Information S1). Every day, prior to the launch of the forecast, GEOS‐CF replays to the past 24‐hr of GEOS FP for Instrument Teams (GEOS FP‐IT; Lucchesi, 2015) assimilated meteorology, aerosols and O₃ in order to ensure consistent model physics within the AGCM (Figures S1a and S1b in Supporting Information S1; see Figure 1, Keller et al., 2021). Unlike GEOS FP, GEOS FP‐IT is a static model system, designed to have minimal updates to the system in order to support near‐real time retrievals by satellite instrument teams. For similar reasons, a “frozen” model was preferred as the driving meteorology for GEOS‐CF v1. It is important to note that in GEOS‐CF the GEOS‐Chem aerosols and O₃ are run passively, that is with no radiation interaction with the meteorology, and therefore do not directly impact the dynamics nor are the increments applied to the GEOS‐Chem aerosols and O₃.

In the GEOS‐CF v1, there is no direct data assimilation of chemical species; however, near‐real time satellite observations of (a) fire radiative power and (b) stratospheric O₃ are incorporated into GEOS‐CF during the replay segments. Specifically: (a) the Quick Fire Emissions Data set (QFED; Darmenov & da Silva, 2015) informs the model of recent fires, which is then persisted forward for each 5‐day forecast and (b) the GEOS‐CF stratospheric O₃ (pressures less than approximately 56 hPa) is nudged toward the GEOS FP assimilated O₃ 3‐hourly average product (Figure S1b in Supporting Information S1). The GEOS FP O₃ observing system includes the limb‐sounding profiles from the near‐real time Microwave Limb Sounder (MLS; Waters et al., 2006) product, column‐based measurements from Ozone Monitoring Instrument (OMI; Levelt et al., 2006, 2018) and, after March 2019, the O₃ observing system was updated to include TCO from Ozone Mapping and Profiler Suite Nadir Mapper (OMPS‐NM; Bak et al., 2017) instrument aboard Suomi National Polar‐Orbiting Partnership. The nudging method is intended to keep stratospheric O₃ in line with observations on a seasonal time scale while still allowing GEOS‐Chem to simulate complex chemical interactions in the troposphere and stratosphere. The nudging technique in GEOS‐CF v1 is as follows: from the top of the atmosphere (GEOS level 1) down to lower stratosphere (GEOS level 33, approximately 40 hPa), the O₃ is nudged 20% toward the GEOS FP O₃ every 3 hr. There is not a hard cut off in the nudging, but instead from levels 33 to 35 (approximately 56 hPa, well above the tropopause), there is a smooth transition, and then from GEOS level 35 to 72 (model's lowest layer), the O₃ is not constrained.

This replay set‐up provides the best initial conditions for the 5‐day forecast initialized at 12:00 UTC (See Figure 1 of Keller et al. [2021]). Since the end of each replay segment is used to start the next day's replay simulation, these 24‐hr segments can be considered as a continuous model best estimate of the 3D composition of the atmosphere, starting 1 January 2018 for GEOS‐CF v1. The GEOS‐CF 5‐day forecasts remain available to the public for a 2‐week period, and are archived at the NASA Center for Climate Simulation (NCCS) for posterity. Full details of the GEOS‐CF model set‐up, including emission data sets, and available model output can be found in Keller et al. (2021) and Knowland et al. (2022), respectively.

3. Data

In this section, the remote‐sensing and balloon‐based observation data sets used (Table 1) for evaluation of the updates to GEOS‐Chem UCX within GEOS‐CF (Section 4) and the GEOS‐CF stratospheric constituents for the year 2020 (Sections 5, 6) following the applied updates are described. In this study, the GEOS‐CF replay estimates of stratospheric composition for O₃, HCl, chlorine monoxide (ClO), nitric acid (HNO₃), N₂O, NO _y (sum of nitric oxide (NO), nitrogen dioxide (NO₂), HNO₃, chlorine nitrate (ClONO₂), and two dinitrogen pentoxide (N₂O₅)) and water vapor (H₂O) will be the main focus of the evaluation. While there are about 40 chemical species available on stratospheric levels from GEOS‐CF, the analysis is limited to O₃ and key observable species related to O₃ chemistry. Within the GEOS‐CF v1 forecast segment, the model output is reduced to only 2D diagnostics (e.g., TCO) and limited 3D output focusing on tropospheric and lower stratospheric composition. In Section 6, the forecast skill for TCO will be presented for two case study periods and an example of forecasting the impact of stratospheric O₃ on tropospheric composition is reported.

Table 1.

Overview of Observation Data Sets Used for GEOS‐CF Model Validation

Description	Species	Reference
Satellite
ACE‐FTS v4.1	O3, H2O, HCl, HNO3, N2O, NO, NO2, N2O5, ClONO2	Boone et al. (2020)
MLS v5	O3, H2O, HCl, HNO3, N2O	Livesey et al. (2020)
SAGE III/ISS v5.1	O3	McCormick et al. (2020) and H. J. R. Wang et al. (2020)
Ozone Watch	O3	https://ozonewatch.gsfc.nasa.gov/
OMI “TOMS‐like” v3 level 3 product	O3	McPeters et al. (2008) and Bhartia (2012)
SBUV Merged Ozone product v8.6	O3	Frith et al. (2014)
Balloon
Ozonesondes	O3	http://www.woudc.org, ftp://aftp.cmdl.noaa.gov/data/ozwv/ozonesonde/
Ground‐based
TOLNet Lidar	O3	https://www-air.larc.nasa.gov/missions/TOLNet

This manuscript focuses on the satellite observations and the global distribution of ozonesondes that can be used to make general conclusions about the global state of the stratospheric composition in GEOS‐CF. Comparisons against regional networks such as the Pandora network or the Tropospheric Ozone Lidar Network (TOLNet) are active areas of research (e.g., Dacic et al., 2020; Gronoff et al., 2021; Johnson et al., 2021; Robinson et al., 2020) as demonstrated with a case study using TOLNet vertically resolved O₃ measurements (Section 6.3).

3.1. Satellite

In addition to limb‐sounding O₃ profiles, MLS observes other constituents which are useful for monitoring ODSs and atmospheric circulation to a sufficient quality on a profile‐by‐profile bases (Livesey et al., 2020). In this study, MLS level 2, version 5 (Livesey et al., 2020) profiles of O₃, H₂O, HCl, ClO, HNO₃, and N₂O for 2020 are used (Table 1). For MLS O₃, data within half an hour of the synoptic times (00:00, 6:00, 12:00, 18:00 UTC) are co‐located with the GEOS‐CF instantaneous O₃ for January–December 2020 (Figure 5). For zonal and global comparisons of MLS HCl, HNO₃, N₂O, and H₂O, only the MLS data within half an hour of 12:00 UTC are co‐located to the 12:00 UTC instantaneous GEOS‐CF 3D concentrations (Figures 7 and 11; Figures S4 and S6 in Supporting Information S1). The GEOS‐CF outputs MLS observed species on 39 approximate MLS pressure levels from 316 to 0.0215 hPa (Knowland et al., 2022) to support comparisons to MLS profiles. For the polar vortex comparison (Figure 8), all available MLS data in that region were used and the profiles interpolated to three isentropic surfaces.

Figure 5.

(a) PDF of the differences of GEOS‐CF (“CF”) O₃ minus MLS O₃ at 27 MLS pressure levels, with mean difference (cross), median difference (open circle), 1σ standard deviation (long dash), and approximate instrument 1σ combined random and systematic uncertainty from the MLS quality document tables (short dash). (b) The mean concentrations for GEOS‐CF (red) and MLS (black) at 27 MLS pressure levels from 146.8 to 1.0 hPa. For (a and b), only MLS data within half an hour of the synoptic times (00:00, 6:00, 12:00, 18:00 UTC) are used for January–December 2020. (c and f) Zonal 2020 annual mean O₃ for GEOS‐CF co‐located to the satellite overpasses, (d and g) the zonal 2020 annual mean O₃ for the satellite and (e and h) the difference of the model minus the satellite for (c–e) MLS and (f–h) ACE‐FTS.

Figure 7.

Similar to Figure 5 but for HCl and only MLS data within half an hour of 12:00 UTC for January–December 2020. Negative values from ACE‐FTS are colored white (g).

Figure 11.

Similar to Figure 7 but for HNO₃.

Figure 8.

Scatter plots of GEOS‐CF (y‐axis) versus MLS (x‐axis) for HCl (a–b, e–f) and ClO (c–d, g–h) for NH February 2020 polar vortex (a–d) and SH August 2020 polar vortex (e–h). “Outside vortex” is defined as from 30°N or °S to the vortex edge. The vortex edge is defined as in Wargan, Weir, et al. (2020).

Other independent observations for model evaluation include measurements from two solar occultation instruments: the Stratospheric Aerosol and Gas Experiment (SAGE) III instrument aboard the International Space Station (ISS) and the Atmospheric Chemistry Experiment‐Fourier Transform Spectrometer (ACE‐FTS) on the Canadian SCISAT satellite. The solar occultation measurements from SAGE III/ISS (June 2017–present; Cisewski et al., 2014) and ACE‐FTS (February 2004–present; Bernath, 2017; Bernath et al., 2005) provide high vertical resolution profiles of O₃, H₂O and other species but there are far fewer observations per day (15–30) compared to MLS profiles (3500). SAGE III/ISS has a measurement range from about 70°S to 70°N (H. J. R. Wang et al., 2020) while ACE‐FTS covers further into the polar regions because of its high orbital inclination (74° compared to 52° for the ISS). The measurements are mainly in the stratosphere, however the retrieved profiles can be extended into the troposphere (generally limited to the cloud top height; Boone et al., 2020; Mauldin et al., 1998) and into the mesosphere (SAGE III/ISS; Mauldin et al., 1998; McCormick & Chu, 2004) and lower thermosphere (ACE‐FTS; Boone et al., 2020). Here, SAGE III/ISS version 5.1 and ACE‐FTS version 4.1 profiles are used. The O₃ profiles from both instruments were interpolated to the GEOS‐CF “MLS pressure levels” and profiles ±30 min of each hour were selected for comparisons to the GEOS‐CF hourly instantaneous O₃ on MLS pressure levels. For the additional chemical species which are not reported by MLS but are reported by ACE‐FTS, the GEOS‐CF 3D 3‐hourly, instantaneous output on 35 isentropic surfaces (from 270 to 3000 K; Knowland et al., 2022) are compared to ACE‐FTS measurements. The ACE‐FTS observations were interpolated to isentropic surfaces from 330 to 1600 K for the comparison and the GEOS‐CF isentropic output within 1.5 hr of the ACE‐FTS measurements are selected.

Along with the satellite level 2 products for the instruments detailed above, publicly available O₃ values from the NASA “Ozone Watch” website (https://ozonewatch.gsfc.nasa.gov/) are used for verification of O₃ forecasts. Ozone Watch daily values of the Northern Hemisphere (NH) polar cap total O₃ and the Southern Hemisphere (SH) ozone hole area are historically based on a wide range of satellite observations; since July 2016 it is based on the OMPS‐NM. If OMPS‐NM data is missing, the Ozone Watch product relies on the near‐real time GEOS FP assimilated TCO product. Merged, homogenized satellite products are useful for evaluation of long‐term simulations since biases across multiple instruments are removed relative to a reference data set. We use two merged products—version 8.6 of the SBUV Merged Ozone Data set (Frith et al., 2014) and version 3 of the OMI “TOMS‐like” level 3 gridded product (Bhartia, 2012; McPeters et al., 2008)—for evaluation of the GEOS Chemistry Climate Model (GEOS CCM; Nielsen et al., 2017) simulated TCO and GEOS‐CF TCO (see Section 4).

3.2. Ozonesonde Observations

Ozonesondes provide profile measurements of tropospheric and stratospheric O₃, up to about 30–35 km altitude (Stauffer et al., 2020; Sterling et al., 2018; Thompson et al., 2017). Data was selected from 20 of the 24 sites in Keller et al. (2021), distributed globally (Table 2, Figure S2 in Supporting Information S1), and accessed through the World Ozone and Ultraviolet Data Center (WOUDC, http://www.woudc.org) and from Global Monitoring Laboratory, National Oceanic and Atmospheric Administration (NOAA) network (ftp://aftp.cmdl.noaa.gov/data/ozwv/ozonesonde/). Keller et al. (2021) reported on the tropospheric portion of the profiles (1,000–200 hPa) for 2018–2019; this study focuses on stratospheric composition and will evaluate the profiles from 400 to 10 hPa. De Bilt, Pohang, Paramaribo, and Marambio were excluded from this study, as sites were selected using the criteria that each location has at least one observation reported in each month, similar to Steinbrecht et al. (2021). The number of ozonesonde launches in 2020 compared to the number of launches in previous years was reduced at many stations because of COVID‐19 restrictions; nonetheless, there were still enough measurements for scientific study at the selected 20 stations (Table 2). At these sites, the frequency of ozonesonde launches is generally once or twice per week, and covers a range of launch times (Table 2).

Table 2.

Ozonesonde Launch Locations, Listed From North to South, Grouped Into Five Latitude Bands (See Also Figure S2 in Supporting Information S1): NH Polar (>60°), NH Mid‐Latitudes (30°–60°), Subtropics/Tropics (−30° to 30°), SH Mid‐Latitudes (−30 to −60°) and SH Polar (<−60°)

Station name	Latitude (°N)	Longitude (°E)	Launch hour (UTC)	N (2020 only)
NH polar
Alert	82.5	−62.3	18 or 23	17
Eureka	80.0	−85.9	11, 18 or 23	65
NH mid‐latitudes
Legionowo	52.4	21.0	11	41
Valentia	51.9	−10.2	11	30
Uccle	50.8	4.3	11–12	144
Praha	50.0	14.4	11	46
Payerne	46.5	6.6	10–12	111
Trinidad Head	41.1	−124.2	16–21	45
Madrid	40.5	−3.6	10–11	54
Boulder	40.0	−105.2	16–21	59
Tateno	36.1	140.1	14–15	37
Subtropics/tropics
King's Park	22.3	114.2	5	48
Hilo	19.7	−155.1	18–19	50
Pago Pago	−14.3	−170.7	14–24	38
Suva	−18.1	178.4	21–23	15
SH mid‐latitudes
Broadmeadows	−37.7	144.9	0–3	51
Lauder	−45.0	169.7	19–8	54
Macquarie Island	−54.4	158.9	5 or 23	51
SH polar
Syowa	−69.0	39.6	2, 8 or 14	46
South Pole	−90.0	169.0	8–11, 20–22	51

Note. Number of launches (N) for January–December 2020 are provided.

The vertical resolution of the ozonesonde profiles (often >2,000 pressure levels) is reduced by interpolating the ozonesonde data onto 200 constant pressure levels from 1,000 to 10 hPa.

For comparisons, the model data are selected for the closest hour to the launch hour and then the closest grid‐box to the ozonesonde station location. Furthermore, the model output is interpolated from the native resolution to the 200 constant pressure levels to match the sonde resolution, as was done in Keller et al. (2021).

3.3. TOLNet Ozone Lidars

In addition to comparisons against sounding data, the capability of the NASA GEOS‐CF model to simulate and forecast the impact of stratospheric O₃ on tropospheric atmospheric composition can be assessed by comparing the GEOS‐CF model output to observations from TOLNet. TOLNet is a network of 8 tropospheric O₃ lidars distributed throughout North America supported by NASA and NOAA (https://www-air.larc.nasa.gov/missions/TOLNet). These ground‐based lidars provide Differential Absorption Lidar (DIAL)‐derived, high vertical and temporal resolution, observations of tropospheric O₃ with high accuracy and precision continuously for many hours or even days (Leblanc et al., 2018; L. Wang et al., 2017). While Keller et al. (2021) found on average the NH free tropospheric O₃ was biased low compared to ozonesondes for 2018–2019, there is demonstrable synergy between the data from these lidar systems and the vertical structure of O₃ concentrations simulated by GEOS‐CF (Dacic et al., 2020; Johnson et al., 2021), including episodic events when stratospheric O₃ descends to lower altitudes into the troposphere (Gronoff et al., 2021).

For this study, observations from the NASA JPL Table Mountain Facility (TMF) tropospheric O₃ lidar (TMTOL; McDermid et al., 2002), located in the San Gabriel Mountains near Los Angeles, California (34.38°N, 117.68°W) at an elevation of 2,285 m above sea level (asl) are used. This system has the capability to conduct continuous observations for multiple hours or days (Chouza et al., 2019) providing O₃ measurements from 100 m above ground level (agl) to the tropopause. For a qualitative comparison to GEOS‐CF for the case study in Section 6.3, the lidar data is averaged hourly with 30 m vertical resolution.

4. Model Updates to GEOS‐Chem UCX for GEOS‐CF

Early evaluation of GEOS‐CF v1 in 2018 against MLS observations indicated that GEOS‐CF had significant biases in the stratosphere (not shown), caused by inaccurate initial conditions of ODSs as well as erroneous stratospheric removal of NO _x . Though the irregular stratospheric concentrations and distribution of some of the species had limited impact on the main observable tropospheric pollutants (Keller et al., 2021), it was critical that the state of the GEOS‐CF stratosphere be addressed in order to be a suitable product for supporting NASA campaigns and remote‐sensing instruments which require realistic stratospheric composition. To do so, parallel long‐term free‐running (i.e., no meteorological replay) GEOS CCM simulations using the two troposphere‐stratosphere chemistry mechanisms—GMI and GEOS‐Chem—were performed (Figure S1c in Supporting Information S1) to assess the GEOS‐Chem UCX stratospheric chemistry against the established GMI chemistry. This analysis confirmed that a well spun up GEOS‐Chem stratosphere does lie within the observable TCO range (Figure 1a).

Figure 1.

Near‐global average (60°S–60°N) TCO (a) monthly mean for the GEOS‐Chem GEOS CCM free‐running simulation (1999–2018; blue), GMI GEOS CCM free‐running simulation (2000–2016; red), and the SBUV Merged O₃ Data Set (1998–2018, black). Vertical lines and gray shaded region represent the standard deviation about the monthly mean for the GEOS CCM simulations and observations, respectively, and (b) daily mean from OMI “TOMS‐like” level 3 gridded product (Bhartia, 2012; McPeters et al., 2008, 7‐day running mean, black line; standard deviation, gray shading) and GEOS‐CF (7‐day running mean, magenta line; standard deviation, magenta vertical lines) for the region from 60°S to 60°N.

From the comparison of these two long‐term free‐running GEOS CCM simulations, four major updates were made to the GEOS‐Chem UCX code base in GEOS‐CF to be more in line with the GMI mechanism since Eastham et al. (2014). In addition, two more changes were made to improve the O₃ nudging technique and the run‐time performance. Finally, the GEOS‐CF stratospheric concentration fields were updated using the well‐spun up (20‐year) GEOS‐Chem GEOS CCM simulation (blue line, Figure 1a). The updates and new initial conditions were implemented in the GEOS‐CF near‐real time system on 31 July 2019. The four major updates to the GEOS‐Chem UCX code are:

First, the stratospheric photolysis and reaction rate constants were updated from JPL Publication 10–16 (Sander et al., 2011) to follow recommendations provided by a more recent release of the JPL kinetic evaluation (JPL Publication 15‐10; Burkholder et al., 2015). The surface mixing ratio boundary conditions for ODSs were updated from the WMO 2006 ozone assessment to follow the newer baseline emission scenario from the WMO 2018 ozone assessment (Carpenter & Daniel, 2018). This update includes changing the methyl bromide (CH₃Br) boundary conditions to follow the WMO 2018 scenario rather than fixed zonal mean values (Parrella et al., 2012). Surface mixing ratio boundary conditions for N₂O in GEOS‐CF are taken from the Representative Concentration Pathway 6.0 scenario for the fifth assessment report of the Intergovernmental Panel on Climate Change (Collins et al., 2013). The original evaluation of the UCX stratosphere was conducted using the GEOS‐Chem version 9.01.03 mechanism that had a detailed tropospheric bromine mechanism (Parrella et al., 2012). In addition the updates to the ODS boundary conditions outlined above, the GEOS‐Chem version (v12.0.1) used by GEOS‐CF v1 includes updates to the tropospheric halogen mechanism that impact the quantity of Cl and Br species reaching the stratosphere (Chen et al., 2017; Schmidt et al., 2016; Sherwen, Schmidt, et al., 2016). In GEOS‐CF v1, the mean annual stratospheric total Cl and Br content for 2020 are 3.0 ppb and 19 ppt, respectively, in general agreement with the stratospheric supply estimated by Engel and Rigby (2018). The amount of Cl supplied to the stratosphere by tropospheric total inorganic Cl (Cl _y ) and VSLS is minor, less than 2%. Based on simulated mixing ratios at the tropical tropopause pressure, 5.6 ± 0.2 ppt of Br is supplied to the stratosphere by tropospheric Br _y and VSLS, in agreement with the previous modeling studies and aircraft observations summarized by the WMO 2018 Ozone Assessment (Engel & Rigby, 2018).

Second, more bromine was activated in GEOS‐Chem than in the GMI simulations, contributing to greater O₃ loss in the lower stratosphere than observed, especially at low and mid‐latitudes (see Figure 1b). Two heterogeneous reactions on PSCs (reactions 1 and 2) and three reactions on stratospheric sulfate aerosols (reactions 1–3) were identified as not included in GMI and subsequently turned off in GEOS‐CF. These reactions are:

$${\mathrm{C}\mathrm{l}\mathrm{O}\mathrm{N}\mathrm{O}}_2(g)+\mathrm{H}\mathrm{B}\mathrm{r}(l,s)\to \mathrm{B}\mathrm{r}\mathrm{C}\mathrm{l}+{\mathrm{H}\mathrm{N}\mathrm{O}}_3$$ (1)

$$\mathrm{H}\mathrm{O}\mathrm{C}\mathrm{l}(g)+\mathrm{H}\mathrm{B}\mathrm{r}(l,s)\to \mathrm{B}\mathrm{r}\mathrm{C}\mathrm{l}+{\mathrm{H}}_2\mathrm{O}$$ (2)

$${\mathrm{B}\mathrm{r}\mathrm{O}\mathrm{N}\mathrm{O}}_2(g)+\mathrm{H}\mathrm{C}\mathrm{l}(l,s)\to \mathrm{B}\mathrm{r}\mathrm{C}\mathrm{l}+{\mathrm{H}\mathrm{N}\mathrm{O}}_3$$ (3)

The heterogeneous reaction 1 between ClONO₂ and hydrogen bromide (HBr) on PSC surfaces was investigated by Hanson and Ravishankara (1992), but this reaction is disabled in the GEOS‐CF system to be consistent with the GMI mechanism. Additionally, Burkholder et al. (2015) recommends that additional studies are needed to properly represent reaction 2, and laboratory analysis suggests that bromine nitrate (BrONO₂) and HCl do not directly react via reaction 3 (Hanson & Ravishankara, 1995).

Third, the family transport of Cl _y and Br _y species is implemented in GEOS‐CF as described by Douglass et al. (2004) for GMI. When halogen species are transported individually, Douglass et al. (2004) identified errors in the advection scheme along sharp gradients between sunlit and nighttime mixing ratios. These advection errors resulted in nonphysical maxima in mixing ratios of Cl _y and Br _y that were detected in earlier versions of the GEOS‐CF stratosphere. Since the total quantities of Cl _y and Br _y do not have sharp day to night gradients, implementing family transport removes occurrences of nonphysical maxima in halogen families in GEOS‐CF v1.

Fourth, the solar zenith angle (SZA) in the photolysis calculations was updated to go beyond 90°, thereby accounting for twilight conditions important for chemistry simulations in the stratosphere and mesosphere. GEOS‐Chem version 12.0.1 and GEOS‐CF now truncate the SZA at 98° as done in GMI and allowed for in the Fast‐Jx photolysis calculations. Previous versions of GEOS‐Chem truncated the SZA at 90°, which resulted in longer nighttime conditions and sharpened the day‐night constituent gradients across the terminator. This contributed to the non‐physical advection errors in the Br _y and Cl _y species described above.

In addition to the new initial conditions for GEOS‐CF stratospheric concentration fields using the well‐spun up (20‐year) GEOS‐Chem GEOS CCM simulation, two more adjustments were made to the GEOS‐CF v1 system: (a) the start of the transition layer for the O₃ nudging was raised from GEOS level 38 (approximately 90 hPa) to GEOS level 35 (approximately 56 hPa as described in Section 2) in order to make sure no GEOS FP O₃ was mistakenly added to the upper‐troposphere since the nudging method does not differentiate between the stratosphere and troposphere; and (b) in the original version of GEOS‐CF, GEOS‐Chem UCX does explicit chemistry up to the stratopause and mesospheric chemistry is parameterized based on pre‐defined production and loss rates. To speed up the run time of the GEOS‐CF system, the mesospheric parameterization was disabled and stratospheric chemistry now extends up through the top of the GEOS atmosphere, thus avoiding the need to repeatedly read in production and loss rates. Note, this study is only evaluating stratospheric composition, considering concentrations up to 1 hPa.

For the evaluation of the GEOS‐CF stratospheric composition in the following sections the focus is on only the 12‐month period in 2020, after allowing several months for the stratosphere to stabilize. One can see an improved agreement in the (non‐polar) TCO between GEOS‐CF and OMI from late 2019 onwards in Figure 1b. Prior to the inclusion of the above outlined updates on 31 July 2019, GEOS‐CF mean non‐polar TCO is biased‐low, and any analysis of the total column diagnostics or 3D stratospheric output from GEOS‐CF v1 for this earlier period of the record should consider the potential biases from the stratospheric portion of the column.

It is unlikely that changes to atmospheric composition in 2020 from the COVID‐19 pandemic restrictions impacted stratospheric composition significantly. For this reason, it is suitable to focus on the year 2020 for this study. Numerous studies investigated how the global COVID‐19 pandemic restrictions impacted surface air quality through a reduction in anthropogenic emissions (an extensive collated list available at https://amigo.aeronomie.be/index.php/covid-19-publications/peer-reviewed); however, there are relatively few which explore the impact on free tropospheric (FT) composition—for example, Steinbrecht et al. (2021) and Clark et al. (2021) report moderate decreases of 7% NH FT O₃ for April–August 2020 and up to 12% in FT O₃ over Frankfurt during March–July 2020, respectively—and no studies to our knowledge with a focus on the stratosphere. While the reduction in air traffic from the grounding of a substantial portion of passenger aircraft (Clark et al., 2021; Le Quéré et al., 2020) likely led to a decrease in O₃‐precursor emissions at cruising altitudes in the upper troposphere and lower stratosphere (UTLS), it is likely the anomalously low NH springtime O₃ in the stratospheric polar vortex is the greater driver in UTLS composition anomalies than the pandemic‐related emission reductions (see Figure 3, Steinbrecht et al., 2021). The anomalous polar vortex circulation and chemistry in the NH (January–May 2020) and the SH (May–September 2020), both of interest to stratospheric chemists, will be discussed in detail throughout Sections 5, 6.

5. Evaluation of GEOS‐CF Stratospheric Composition in Replay Mode

In this section, the spatial distribution and variations for stratospheric O₃ (Section 5.1) and several species important for O₃ chemistry (Section 5.2) are evaluated against independent observations and related to the complexity of chemistry and emissions. Once the state of the GEOS‐CF stratospheric composition with analyzed meteorology is established, applications of the GEOS‐CF forecasts are presented (Section 6).

5.1. Ozone

Since the GEOS‐CF stratospheric O₃ is constrained during the replay segment by the GEOS FP O₃ product which assimilates MLS, OMI and OMPS‐NM O₃ observations, independent profile observations from ozonesondes, ACE‐FTS, and SAGE III/ISS are used for validation with a comparison to MLS included.

In general, the median stratospheric O₃ simulated in GEOS‐CF for the period between January through December 2020 agrees well with the median ozonesonde profiles (Figure 2) with median percent bias within ±20% through most of the stratosphere (Figure 3). While Alert and Eureka are located close to each other in northern Canada (see Figure S2 in Supporting Information S1), the median profiles between 150 and 30 hPa are very different for these two stations. This is attributed to the reduced number of profiles in 2020 for Alert compared to Eureka (17 and 65, respectively, Table 2), since this difference is not present when all profiles from 2018 to 2020 are considered (not shown). In addition, while Suva has the fewest profiles (15; Table 2) and exhibits a similar profile to its closest neighboring site Pago Pago (Figure 2) it has the largest median percent bias of all the profiles (>80% at 100 hPa; Figure 3). Furthermore, at the SH locations (King's Park to South Pole), there is a high bias in GEOS‐CF median O₃, most notably between about 200 and 50 hPa (Figures 2 and 3). This is consistent with Stauffer et al. (2019), who assessed the “MERRA2‐GMI” product (GEOS with GMI chemistry replayed to MERRA‐2 meteorology; Strode et al., 2015) against ozonesondes for the period 1980–2016 and found the subtropical and tropical sonde locations had median percent bias over 20% between 15 and 20 km, and as they note, the median percent biases are large but the O₃ concentrations at these altitudes are low. Stauffer et al. (2019) also present a high bias for the MERRA2‐GMI at SH high latitude sites between 10 and 15 km. Here, the differences between GEOS‐CF and the SH polar observations at Syowa and South Pole in 2020 are driven by the model not capturing the low O₃ values in this layer of the atmosphere (between about 200 and 50 hPa, 25th percentile, dashed pink line, Figure 2) during austral winter and spring (individual months not shown). Possible reasons for biases in the SH polar regions in 2020 as it relates to polar chemistry are explored later in Sections 5.2 and 7.

Figure 2.

Median ozonesonde profiles (O₃, mPa) restricted to pressure levels between 400 and 10 hPa at 20 global stations for launches in January–December 2020 (median, black line; interquartile range, gray shading) compared to median GEOS‐CF O₃ profiles (median, magenta solid line; interquartile range, magenta dashed lines). GEOS‐CF profiles selected for the grid‐box and time closest to the ozonesonde measurements. Launch locations displayed in order from North to South, as listed in Table 2.

Figure 3.

Similar to Figure 2, except median percent bias (GEOS‐CF minus ozonesonde divided by ozonesonde). Note, x‐axis range is generally from −40% to 40% except at Pago Pago, Suva, and South Pole.

Stratospheric O₃ in GEOS‐CF also agrees well with SAGE III/ISS solar occultation profiles between 100 and 4.6 hPa for January through December 2020 with correlation coefficients (r) greater than or equal to 0.92 (Figure 4 inset). At higher altitudes, near the stratopause at 1 hPa, the correlation is reduced, r = 0.61, with SAGE III/ISS reporting higher concentrations of O₃ than simulated by GEOS‐CF (Figure 4). Despite weakly nudging the stratospheric O₃ in GEOS‐CF toward the GEOS FP O₃ product which assimilates MLS, the annual mean GEOS‐CF O₃ is also biased low to MLS O₃ between 5 and 1 hPa globally (Figures 5a–5e). Since H. J. R. Wang et al. (2020) demonstrated very good agreement (<5%) between SAGE III/ISS and MLS in the stratosphere (20–55 km altitude), the difference in upper stratospheric O₃ between GEOS‐CF and SAGE III/ISS is likely a deficiency in the model at these altitudes. Previous chemical transport modeling studies have reported negative biases with respect to upper stratospheric/lower mesospheric ozone (e.g., Errera et al., 2019; Eyring et al., 2010; Siskind et al., 2013), and due to the relatively short chemical lifetime of ozone in this region, assimilation was found to not correct for the low bias in the BASCOE system (Skachko et al., 2016). It is very likely that in the upper stratosphere, chemical timescales are shorter than the 3‐hr frequency of the ozone nudging in GEOS‐CF. The annual zonal mean O₃ distribution for ACE‐FTS is greater than GEOS‐CF throughout most of the stratosphere, with the maximum difference located near the stratospheric O₃ concentration peak (Figures 5f–5h); the negative bias is expected as ACE‐FTS has a known positive bias to coincident MLS profiles (Dupuy et al., 2009; Errera et al., 2019; Sheese et al., 2017, 2022).

Figure 4.

SAGE III/ISS solar occultation O₃ profiles for January–December 2020 interpolated to 5 MLS pressure levels—100, 46, 10, 4.6, 1 hPa—and compared to GEOS‐CF O₃.

As demonstrated by this evaluation against independent observations, GEOS‐CF realistically simulates stratospheric O₃ distributions between about 100 and 5 hPa. In the upper stratosphere (5–1 hPa), the disagreement between GEOS‐CF simulated O₃ and satellite observations (SAGE III/ISS and MLS) will require further investigation but is likely associated with the extension of stratospheric chemistry up to the mesosphere. The positive bias in GEOS‐CF O₃ in the SH polar region between about 200 and 50 hPa present in the comparisons against ozonesondes (Figure 2) and satellite observations by both MLS and ACE‐FTS (Figures 5e and 5h) will also be monitored closely.

5.2. Chemical Species Important to Stratospheric O₃ Chemistry

Next, comparisons of the model against satellite observations are presented for stratospheric species that are relevant to polar vortex chemistry and observed by both MLS and ACE‐FTS, including two inorganic chlorine species (HCl and ClO), two nitrogen species (HNO₃ and N₂O), and additional nitrogen species only observed by ACE‐FTS.

5.2.1. Inorganic Chlorine

Inorganic chlorine in the stratosphere is the result of transport of tropospheric long‐lived chlorine compounds, most notably chlorofluorocarbons (e.g., CFC‐11 (CCl₃F) and CFC‐12 (CCl₂F₂)), chlorinated solvents (e.g., carbon tetrachloride (CCl₄)) and methyl chloride (CH₃Cl). Once in the stratosphere, the long‐lived compounds photolyze and react with other chemical species (in the presence of UV) to form reactive chlorine, which through catalytic cycles can lead to loss of stratospheric O₃ (Molina & Rowland, 1974a, 1974b). The CFCs and CCl₄ are the result of industrial activities and other man‐made products which have been phased out following the Montreal Protocol and subsequent amendments (Reimann et al., 2018). CH₃Cl originates mainly from natural sources such as biomass burning emissions, the ocean, and fungi (Keene et al., 1999).

It is critical for the GEOS‐CF forecast capabilities of stratospheric O₃ that species such as these are simulated correctly. Several other Cl _y species are observable from space, however, the focus is limited to (a) HCl, a non‐ozone‐destroying chlorine reservoir and (b) ClO, an active, ozone‐depleting chlorine radical (Stolarski & Cicerone, 1974). HCl is abundant in the stratosphere, especially at high altitudes, and as a reservoir species it is relatively inert. Because of the global distribution of these chlorine species, O₃ loss through catalytic cycles can occur throughout the stratosphere; however, this is usually at a slower rate compared to O₃ loss following the conversion of HCl and ClONO₂ (another chlorine reservoir) to ClO on PSCs (Solomon et al., 1986) within a sunlit winter‐time polar vortex. When polar stratospheric temperatures begin to drop as the vortex forms, the environment becomes favorable for the formation of PSCs. While the maintenance of extremely cold temperatures is more common in the austral winter and spring polar vortex, during the 2020 boreal winter and spring a stable polar vortex led to PSCs which were observed by the OMPS Limb Profiler (LP; DeLand et al., 2020). Within the polar vortex, the heterogeneous chemistry can lead to substantial destruction of stratospheric O₃. This is demonstrated in the snapshot of the NH polar vortex on 29 February 2020 at 22:00 UTC, comparing GEOS‐CF simulated concentrations to measurements from a single MLS overpass (Figure 6). As stated in Section 5.1, it is no surprise that the NH O₃ agrees well to MLS in Figure 6a since GEOS‐CF at 45 hPa is nudged toward the GEOS FP assimilated product. Presented here is how GEOS‐CF simulates the location and chemistry of the vortex; although, GEOS‐CF underestimates the observed high values of HCl outside the vortex (Figure 6b) and the highest ClO values within the sunlit portion of the vortex (Figure 6c) as seen by MLS.

Figure 6.

Snapshot of 29 February 2020 at 22:00 UTC for GEOS‐CF (map) and a single over‐pass of MLS (colored circles; measurements from 21:43 UTC to 22:14 UTC) at 45 hPa for (a) O₃, (b) HCl, and (c) ClO, emphasizing the NH polar vortex chemistry.

Figure 6 is only an example on one pressure level (45 hPa), but it is an accurate representation of the global distribution further investigated in Figures 7 and 8. First, the annual global distribution of HCl from the model is compared against MLS and ACE‐FTS profiles of HCl in Figure 7. Throughout the stratosphere, GEOS‐CF simulates the vertical gradient of increasing HCl concentrations from the lower stratosphere to upper stratosphere as seen by the satellite measurements. However, the model is biased low compared to the 2020 observations. This holds true at all latitudes except in SH polar region in the lower stratosphere when compared against ACE‐FTS measurements where there is a positive difference (100–50 hPa; Figure 7h). The positive bias in ACE‐FTS, which is not seen in the annual zonal difference between GEOS‐CF and MLS, is likely due to a sampling bias by ACE‐FTS. The SCISAT orbit is such that ACE‐FTS has sunrise measurements south of 60°S only during a few months a year (March, April, July, early August, and November; https://ace.uwaterloo.ca/mission_orbit.php). During July and August, there are positive biases between GEOS‐CF and MLS in the SH lower stratosphere (top, Figure S3 in Supporting Information S1), however, there is a large negative bias in late 2020 between GEOS‐CF and MLS (Figure S3 in Supporting Information S1) that likely cancels out the mid‐year positive biases seen in the SH high latitudes. There is also a bias between ACE‐FTS and GEOS‐CF ClONO₂ (Figure S5 in Supporting Information S1), which may indicate that the Cl _y loading is low in the model.

Second, to look at the vortex chemistry in more detail, the polar distributions of HCl and ClO during February 2020 (NH only; Figures 8a–8d) and August 2020 (SH only; Figures 8e–8h) for three isentropic surfaces (400, 500, and 600 K) are compared for GEOS‐CF against MLS. The vortex edge is defined as in Wargan, Weir, et al. (2020) using threshold values of PV scaled by standard static stability (“sPV”) selected on isentropic surfaces based on the winter mean sPV gradients in equivalent latitude (Lawrence et al., 2018; Manney et al., 1994). For the model to correctly simulate the O₃ destruction within the vortex, there needs to be an accurate representation of the heterogeneous processes. Within the polar vortexes (NH and SH), concentrations of HCl both observed by MLS and simulated by GEOS‐CF decreased compared to outside the vortex (Figure 8); however, GEOS‐CF simulated HCl is biased high (low) within the SH (NH) vortex for August 2020 (February 2020) compared to MLS. It is on the PSCs that the chlorine reservoir species are converted to ClO through heterogeneous processes in the presence of sunlight (Figure 8). Within the polar vortexes of 2020, GEOS‐CF simulates the increase in ClO abundance within the sunlit portion, although GEOS‐CF is biased high with respect to MLS at higher altitudes where there is also a low bias in simulated HCl (600 K, Figures 8d and 8h), likely indicating too much chlorine was activated. Since global distributions of ClO are very low outside of the sunlit portion of the vortex, a comparison on the global scale, similar to Figure 7, was not performed.

5.2.2. Nitrogen Family

Another catalytic cycle for stratospheric O₃ loss is with nitrogen oxides (NO _x  = NO + NO₂). In the stratosphere, N₂O is the main source for NO and subsequently other nitrogen species collectively referred to as NO _y . We define NO _y as the sum of major reactive nitrogen species: NO + NO₂ + HNO₃ + ClONO₂ + 2 × N₂O₅. A long‐lived greenhouse gas, N₂O has natural and anthropogenic sources in the troposphere with no significant sinks until reaching the stratosphere. Once in the stratosphere, N₂O dissociates through photolysis and reaction with excited oxygen atoms to produce NO and is thus a major source of stratospheric NO _y (Crutzen, 1970). During the night time, some NO₂ is converted to N₂O₅, which acts as a reservoir species for NO _x until the sunlight returns. The reaction of ClO with NO₂ forms ClONO₂ (Rowland et al., 1976), and ClONO₂ is a reservoir species for both reactive chlorine and nitrogen. HNO₃, another nitrogen reservoir, is formed by the reactions of NO₂ with the hydroxyl radical (OH) and through heterogeneous reactions with N₂O₅ and HNO₃ later photolyzes to return OH and NO₂ to the system (Brasseur & Solomon, 2005).

The annual zonal mean distributions of N₂O, NO _x and NO _y in GEOS‐CF are compared against measurements from ACE‐FTS in Figure 9. While N₂O measurements are available from both MLS and ACE‐FTS, profile measurements of NO _x are only available from ACE‐FTS and there is a known bias in MLS N₂O measurements in the lower stratosphere (Livesey et al., 2021). The expected N₂O distribution based on the known sources and sinks can be clearly seen in Figure 9 (see also Figure S4 in Supporting Information S1 for MLS and ACE‐FTS on pressure levels), with the largest concentrations in both the model and the satellite at lower altitudes (closer to tropospheric sources) as well as reaching higher altitudes near the equator because of strong upwelling into the stratosphere over the tropics. At concurrent sampling of GEOS‐CF to ACE‐FTS measurements, the N₂O spatial patterns for the model and satellite in the stratosphere are consistent, although the model is biased low through much of the stratosphere (isentropic levels up to 1100 K) and biased high in the upper stratosphere (1200–1600 K), particularly in the tropical region (Figure 9; see also from 50 to 5 hPa and 5 to 1 hPa in Figures S4f–S4h of Supporting Information S1 for similar difference patterns in comparison to MLS N₂O).

Figure 9.

Zonal annual means for GEOS‐CF (a, d, and g) and ACE‐FTS measurements (b, e, and h) and the difference of GEOS‐CF minus ACE‐FTS (c, f, and i) for N₂O (a–c), NO + NO₂ + 2 × N₂O₅ (“${{\mathrm{N}\mathrm{O}}_x}^{\ast }$”; d–f), and NO _y (NO + NO₂ + HNO₃ + ClONO₂ + 2 × N₂O₅; g–i) for isentropic levels from 330 to 1600 K. Note, ACE‐FTS does not measure ClONO₂ at high altitudes so above the black line ClONO₂ missing values were set to zero for the calculation of NO _y (g–i).

To reduce the potential errors because of mismatches around twilight between the GEOS‐CF gridpoint and the ACE‐FTS measurements, we included N₂O₅ with NO _x as “${{\mathrm{N}\mathrm{O}}_x}^{\ast }$” to estimate the full diurnal cycle of NO _x in Figures 9d–9f. For both the satellite and GEOS‐CF, there is a maximum in ${{\mathrm{N}\mathrm{O}}_x}^{\ast }$ (15 and 18 ppbv, respectively) in the tropical upper stratosphere (around 1200–1400 K) and concentrations decrease toward the higher latitudes. Regions of high NO _y extend from the tropics toward lower isentropes and higher latitudes as the mixing ratios of ${{\mathrm{N}\mathrm{O}}_x}^{\ast }$ decrease due to the conversion of NO _x into HNO₃ and ClONO₂ (Figures 9g and 9h; see also Figure 11 for HNO₃ only distributions and Figure S5 in Supporting Information S1 for NO _y partitioning for ACE‐FTS and GEOS‐CF).

Since the production of N₂O in the stratosphere is insignificant, it is an ideal tracer for evaluation of model transport that has an anti‐correlation with stratospheric age (e.g., Jin et al., 2009; Manney et al., 2009; Ruiz et al., 2021; Strahan et al., 2007). Stratospheric age is the mean time an air parcel has spent in the stratosphere after crossing the tropopause from the troposphere (Waugh, 2009). While the individual nitrogen species in NO _y are not long‐lived, together they can be considered as a long‐lived tracer. Generally, NO _y mixing ratios increase and N₂O decrease as air “ages” in the stratosphere (see Figure 9); thus, compact relationships form between NO _y and N₂O due to transport and isentropic mixing (e.g., Chang et al., 1996; Koike et al., 2002; Plumb, 2007; Wetzel et al., 2002). Since in the stratosphere air parcels generally move adiabatically, it is useful to explore these relationships using isentropic surfaces (i.e., constant potential temperature). In Figure 10, values of stratospheric NO _y are shown relative to N₂O with colors representing the potential temperature of the individual non‐polar points and black for all polar points. Concentrations of N₂O are the highest near the tropospheric sources, seen in both ACE‐FTS and GEOS‐CF at low potential temperature levels. The relationship between N₂O and NO _y is comparable between the satellite and model as air enters the lower stratosphere from the troposphere and ages as it moves upward (to higher potential temperature levels), evidence that GEOS‐CF has realistic transport in the lower to middle stratosphere (N₂O > 100 ppbv).

Figure 10.

NO _y (NO + NO₂ + 2N₂O₅ + HNO₃ + ClONO₂) versus N₂O for ACE‐FTS (left) and co‐located GEOS‐CF (within 1.5 hr, as in Figure 9; right) colored by potential temperature from 380 to 1600 K for latitudes from ±60°, polar observations are black.

However, GEOS‐CF does not capture the spread of high values of NO _y (>15 ppbv) observed by ACE‐FTS in the stratospherically aged air (i.e., mixing ratios of N₂O < 100 ppbv). When the air reaches the upper stratosphere (warm colors in Figure 10, N₂O < 100 ppbv) and polar regions (black dots in Figure 10 indicate >|60°|), the tracer‐tracer relationship is no longer linear. In the upper stratosphere, chemical processing of NO _y takes place faster than the timescales of the stratospheric transport, as evidenced by the drop off in NO _y as N₂O mixing ratios decrease below 100 ppbv. Similarly, the observed and simulated low values of NO _y and N₂O below the main tracer‐tracer curve (black points in Figure 10) suggest that GEOS‐CF properly represents the polar vortex mechanisms that remove NO _y from the system until NO _y ‐rich air from the mid‐latitudes replenishes the polar regions after the break‐up of the vortex. The NO _y depleting mechanisms that take place within the polar vortex include reversible “denoxification” (removing NO _x from the gas phase) and irreversible “denitrification” (sedimentation of HNO₃‐containing PSCs; Salawitch et al., 1989; Toon et al., 1990).

Isolating HNO₃ from NO _y is portrayed in Figure 11. In GEOS‐CF, the mid‐ to high latitude maxima of HNO₃ are simulated correctly in the lower stratosphere between 100 and 10 hPa, where the photochemical lifetime of HNO₃ is long, however the concentrations are not as large as observed by MLS or ACE‐FTS (Figures 11c–11h). The variability of the differences, as indicated by the standard deviation (long dash lines) in Figure 11a, peaks between 60 and 15 hPa, which aligns with the increase in the instrument's combined random and systematic uncertainty (short dash lines) and the peak in maximum global HNO₃ concentrations (Figure 11b). Because of the systematic low bias of the model compared to MLS HNO₃, the standard deviation about the mean differences and the uncertainty are offset but otherwise have similar magnitudes. Near the poles, concentrations of HNO₃ decrease (Figures 11c, 11d, 11f, and 11g) through denitrification. The spread of the N₂O:NO _y polar points below the majority of the points in Figure 10 indicates that the model is simulating denitrification similar to ACE‐FTS measurements.

In order to inform future model development, we hypothesize some possible reasons for the biases in nitrogen species related to chemistry and emissions that should be considered in future versions of GEOS‐CF.

In the polar regions, there are negative differences between ACE‐FTS and GEOS‐CF ${{\mathrm{N}\mathrm{O}}_x}^{\ast }$ in the upper stratosphere and throughout the polar stratosphere for NO _y . This may be linked to missing sources of mesospheric NO _x . One such source is in the thermosphere whereby energetic electrons from galactic cosmic rays react with molecular nitrogen (N₂) to produce atomic nitrogen (N) in either excited N(²D) or ground N(⁴S) state that can then react with molecular oxygen (O₂) or OH to produce NO (e.g., Siskind et al., 1997; Solomon et al., 1982). There is evidence that some of this NO can be transported down into the mesosphere and stratosphere, especially in the polar regions where there is downwelling in the mesosphere, and concentrations should be higher as it is not photochemically destroyed during polar night (Funke et al., 2005; Randall et al., 2005, 2007). This missing source from galactic cosmic rays has been identified in another modeling study to explain some of the discrepancies in chlorine and nitrogen species associated with the SH winter and spring polar vortex when compared against satellite observations (Grooß et al., 2018). Sources of mesospheric NO _x are not represented in the GEOS‐CF system and may be further confounded by the extension of stratospheric chemistry into the mesosphere in GEOS‐CF (see Section 2). Furthermore, when each month is assessed individually, from April 2020 to August 2020, the SH stratospheric low HNO₃ bias decreases in the same region as the high bias in HCl (Figure S3 in Supporting Information S1) while the biases in both HCl and HNO₃ increase along the vortex edge instead of in the vortex center during winter time in keeping with the findings of Grooß et al. (2018) for HCl. A future version of GEOS‐CF may benefit from upper‐boundary emission sources representing the solar and galactic high energy particles as diagnosed by Grooß et al. (2018).

In the equatorial stratosphere there is a positive bias in ${{\mathrm{N}\mathrm{O}}_x}^{\ast }$ and NO _y between the model and ACE‐FTS (Figures 9f and 9i). As stated in Keller et al. (2021), GEOS‐CF uses the unadjusted lightning parameterization (described in Murray et al. [2012]) which leads to higher lightning NO _x in the tropics. However, this is likely a very small contribution to the positive difference seen in the equatorial region between observations and model in Figure 9. Another theory is the positive bias of N₂O above the large negative bias in the tropics (Figure 9c) may indicate that the vertical transport is too fast, however the model generally captures the observed distributions of nitrogen‐containing species (Figures 9a and 9b) and the N₂O to NO _y relationship in the lower to mid‐stratosphere (Figure 10), both indicating that GEOS‐CF correctly captures the large‐scale transport pathways (Holton, 1986; Mahlman et al., 1986). Instead of transport, the biases may be due to chemistry. Higher in the equatorial stratosphere, the GEOS‐CF maximum in ${{\mathrm{N}\mathrm{O}}_x}^{\ast }$ is larger in magnitude and extends to higher potential temperature surfaces than observed by ACE‐FTS (Figures 9d and 9e). The positive bias in GEOS‐CF ${{\mathrm{N}\mathrm{O}}_x}^{\ast }$ is in a similar location as the positive bias in N₂O. With the increased available N₂O, production of ${{\mathrm{N}\mathrm{O}}_x}^{\ast }$ may be greater in the upper stratosphere than is observed. Also, the conversion to other nitrogen species, such as HNO₃, may be too slow, as indicated by NO _y partitioning (Figure S5 in Supporting Information S1).

6. Evaluation of Stratospheric O₃ Forecast Capability

In Section 5, the state of the stratospheric composition for GEOS‐CF when the model is constrained by observed meteorology was characterized. In this section, a few case studies explore the skill of the GEOS‐CF model during the 5‐day forecasts when the meteorology and chemistry are free‐running. First, the evaluation of 5‐day forecasts for the NH and SH anomalous polar events using GEOS‐CF and GEOS FP against the NASA Ozone Watch merged satellite product is presented. The year 2020 highlighted some aspects of stratospheric O₃ interannual variability which occur because of both atmospheric dynamics and chemistry. In particular, during the boreal winter to spring, the relatively undisturbed stratosphere allowed the NH polar vortex and associated anomalously low polar O₃ to persist (e.g., Dameris et al., 2021; Inness et al., 2020; Lawrence et al., 2020; Manney et al., 2020; Wohltmann et al., 2020). A similar situation existed in the 2020 austral late winter and spring, where, as will be shown below the strongly zonal stratospheric winds allowed the SH ozone hole to extend longer than normal (Lecouffe et al., 2022). Thus, in both time periods, polar O₃ column values were generally far below their climatological values, highlighting the need during these times for O₃ chemistry forecasts based on full stratospheric O₃ chemistry (e.g., GEOS‐CF) rather than parameterized chemistry (e.g., GEOS FP) which can be based on average production and loss rates or an O₃ climatology. GEOS‐CF forecasts are first described for the 2020 NH anomalous event for the total O₃ column in the 63°–90°N polar cap (Section 6.1), followed by forecasts for the area of the 2020 SH ozone hole size as measured by the total O₃ column less than 220 DU (Section 6.2).

Another application of the GEOS‐CF forecasts is the ability to provide the air quality community with realistic 5‐day forecasts of stratospheric intrusion events, when stratospheric O₃‐rich air is irreversibly mixed into the troposphere, which can lead to O₃ air quality exceedances events especially at high altitude locations. This new capability is highlighted in Section 6.3 (see also Duncan et al. [2021]).

6.1. NH Spring 2020 Polar Ozone Anomaly

Figure 12 depicts the record low NH polar cap O₃ during January–April 2020 (blue curve compared to black curve, Figure 12a; e.g., Inness et al., 2020; Lawrence et al., 2020; Manney et al., 2020) with the average value for March being approximately 75 DU (20%) below climatology. During this time, the 5‐day GEOS FP forecast trajectories (gray curves) tended toward the higher climatological values (e.g., for March is on the order of 400–450 DU; Feng et al., 2021), as expected with simplified chemistry. On the other hand, the corresponding GEOS‐CF trajectories (red curves) remained consistent with the future GEOS‐CF initial values, as the sophisticated GEOS‐Chem chemistry is able to simulate a more realistic atmosphere. The smaller GEOS‐CF mean 5‐day bias (with respect to the concurrent GEOS‐CF replay) for February and March period, −1.8 DU, compared to the GEOS FP bias (with respect to concurrent GEOS FP analyses), 11 DU, reflects this tendency (see inset, Figure 12a). The GEOS‐CF mean behavior consistently tracked closely to the independently analyzed Ozone Watch values (blue contour). The closeness of the GEOS‐CF and GEOS FP 5‐day forecast's standard deviation of the error, 3.1 and 3.3 DU respectively (see inset, Figure 12a), indicate that both systems realistically captured the day‐to‐day dynamically induced variations of polar cap O₃.

Figure 12.

Total column O₃ (DU) for the NH Polar Cap region (63°–90°N) from GEOS‐CF 5‐day forecast trajectories (red), GEOS FP 5‐day forecast trajectories (gray), Ozone Watch analysis (blue), and Ozone Watch 1979–2020 climatology (black) (a) from January to April 2020 and (b) from 5 March to 10 March 2020 for forecast initialized 12:00 UTC 5 March 2020. (a) The thick vertical lines denote the first day of each month, while the light vertical lines denote 5‐day intervals starting from 1 January 2020. The yellow box indicates the period of the case study in (b). (b) Date labels correspond to mid‐point in the day (12:00 UTC).

As a specific example, the GEOS‐CF and GEOS FP forecast trajectories, initialized on 5 March 2020, evolved in different directions (Figure 12b). Since GEOS‐CF is nudged toward the GEOS FP O₃, the forecasts start at a similar place; however, GEOS‐CF and GEOS FP forecasted changes of −7.5 and 5.4 DU, respectively, over the 5‐days. In this example, the GEOS‐CF predicted polar cap O₃ decrease agreed well with the Ozone Watch analyzed change of −7.9 DU. Furthermore, a hemispheric view of the 5‐day forecast error for the 10 March 2020 (Figure 13) reveals a substantial increase over most of the polar cap in GEOS FP compared to the more random error pattern found in GEOS‐CF. In addition to the errors in the polar cap, GEOS FP NH middle latitude O₃ column errors often peak higher than the corresponding GEOS‐CF errors (Figure 13, red values).

Figure 13.

Total O₃ forecast error (DU) for 10 March 2020 calculated by the 5‐day forecast initialized on 5 March 2020 at 12:00 UTC minus the analysis (date the forecast is valid) from (a) GEOS‐CF and (b) GEOS FP. Solid black circle indicates the 63°N latitude for the polar cap region of interest.

6.2. SH 2020 Ozone Hole Area

As demonstrated in Figure 14, the distinctive, long duration, 2020 ozone hole (defined where TCO < 220 DU; blue line) kept its area larger than the climatological average (black line) from August until after November. The anomalous polar vortex conditions again push past the limits of the GEOS FP O₃ forecasts with simple chemistry, as in the 2020 NH spring. Forecasting the ozone hole area during the development of the SH ozone hole in August proved difficult for both GEOS‐CF and GEOS FP; since weak gradients near the 220 DU value exist at this time, it makes exact determination of the area difficult, which may influence the analysis uncertainty at this stage. In addition, the sunless August polar region limits coverage of solar backscatter satellite O₃ observations and therefore less observational constraints on the models' analyzed O₃. However, by the middle of September, the ozone hole area determined from the GEOS FP initial conditions (GEOS FP analysis, corresponding to the start of each gray line) and the GEOS‐CF 5‐day forecast trajectories (start of each red line) agreed well with the O₃ Watch 2020 values.

Figure 14.

Similar to Figure 12a except for ozone hole area (10⁶ km²) from August to November 2020, with the 5‐day intervals starting from 1 August 2020.

As expected, the GEOS FP 5‐day forecasts tended toward a smaller ozone hole area, more characteristic of the climatological ozone hole area (black line, Figure 14; see also Figure 8 of Nielsen et al. [2017]). Over the 4‐month period and using self‐validation (in units of 10⁶ km²), the GEOS FP fifth‐day forecast bias (−2.45; see inset Figure 14) greatly exceeded in magnitude the GEOS‐CF forecast bias (−0.13), and the GEOS FP error standard deviation (2.13) also exceeded that of GEOS‐CF (0.64). Thus, despite not simulating the ozone hole area consistent with Ozone Watch at the onset in August, during the SH ozone hole of 2020, GEOS‐CF successfully forecasted changes in the ozone hole area out to 5‐days.

6.3. Forecast Capability for Stratospheric Intrusions

Stratospheric intrusions occur when the tropopause—the boundary between the stratosphere and troposphere—wraps around the jet core, bringing stratospheric air down toward the surface. This folding of the tropopause is generally associated with upper‐tropospheric level troughs and cut‐off lows. These synoptic weather patterns occur year round, however the tropopause folding events are of interest to air quality managers especially in the spring and early summer (March through June). During this time of year there is a maximum in O₃ in the lower stratosphere which is drawn down within a fold and the photochemical production of O₃ at the surface is not yet the dominant source of O₃ leading to air quality standard exceedances. Ott et al. (2016) and Knowland et al. (2017) both demonstrated that the GEOS model run at horizontal resolutions of 50 km or less with O₃ data assimilation can represent stratospheric intrusions which are linked with ground‐level O₃ enhancements, however the tropospheric O₃ is biased from the simplified chemistry used in the GEOS forecast and reanalysis products prior to the inclusion of GEOS‐Chem in the GEOS‐CF system.

Tropospheric O₃ lidars have a demonstrated record of successfully measuring stratospheric intrusions (e.g., Kuang et al., 2012, 2017; Langford et al., 2009). Here one example of a large stratospheric intrusion event forecasted in near‐real time by GEOS‐CF to pass over NASA JPL's TMF (Figure 15) on 13 June 2020 as captured by TMTOL (Figure 16a) is examined. Five days in advance, the GEOS‐CF indicated a potential O₃ enhancement above TMF that is likely of stratospheric origin; above 6 km O₃ concentrations exceeded 150 ppbv, the value commonly used to define the chemical tropopause (Prather et al., 2001) (Figure 15a). This feature was then present in each of the 5‐day forecasts at decreasing lag times (Figures 15b–15e), indicating a high likelihood that it is a dynamic event and will be realized. At the location of TMF, GEOS‐CF simulates both the high levels of stratospheric O₃ and the photochemically produced O₃ enhancement transported from Los Angeles basin up to TMF (high levels of O₃ near the 2,000 m altitude; Figure 15).

Figure 15.

GEOS‐CF 5‐day (120 hr) O₃ forecasts for grid box closest to Table Mountain Facility (TMF) (34.25°N, 117.75°W) initialized at 12:00 UTC on (a) 8 June 2020, (b) 9 June 2020, (c) 10 June 2020, (d) 11 June 2020, and (e) 12 June 2020. The GEOS‐CF O₃ on 23 pressure levels from 1,000 to 10 hPa are interpolated to altitude in meters asl for comparison to TMF observations (see Figure 16). Vertical pink dashed lines indicate the 24‐hr period of 13 June 2020 in each of the forecasts.

Figure 16.

O₃ curtains on 13 June 2020 from (a) the TMTOL measurements (30 m vertical resolution) and (b) similar to Figure 15 except GEOS‐CF replay O₃ for the hours of TMTOL operation. The white areas are where high quality lidar data was unavailable. For comparison, the co‐located model data are also removed and indicated as white space.

On 13 June 2020, the TMTOL operated throughout the day (Figure 16a). The GEOS‐CF replay output (Figure 16b; originally on pressure levels and converted to altitude) simulates the two O₃ tongues around the time of enhancements seen by the TMTOL. While there are differences in the extent and timing of the O₃‐rich air descending into the troposphere, this example highlights the strengths of the GEOS‐CF's coupled stratosphere‐troposphere chemistry in its ability to forecast the impact of stratospheric composition on tropospheric air quality.

7. Conclusions

NASA's GEOS Composition Forecast system (GEOS‐CF; Keller et al., 2021) provides near real‐time estimates of recent atmospheric composition with daily 5‐day forecasts at high spatial resolution (0.25° latitude × 0.25° longitude up to the lower mesosphere) and high temporal frequency (3D at hourly and 3‐hourly intervals). GEOS‐CF products are used to support ground‐based, balloon, and satellite‐based instrument teams, as well as field and aircraft campaigns that measure trace gases in the troposphere and the stratosphere. Specifically, for surface air quality, it is important that GEOS‐CF simulates the stratosphere to troposphere transport as stratospheric O₃ can be transported to the surface and impact surface air quality. Based on this new capability from the GEOS forecast models, GEOS‐CF is used in a daily tailored email alert system for the TOLNet operators. Furthermore, with the meteorology and composition on an identical grid, this makes it ideal to support satellite observations that need a priori information from a model for the trace gas retrievals or to diagnose stratospheric from tropospheric air masses. Instrument teams, such as for TEMPO (Tropospheric Emissions: Monitoring of Pollution; Zoogman et al., 2017), will benefit from near‐real time prior information provided by GEOS‐CF for their satellite retrievals.

This study focused on concentrations of stratospheric O₃ and chemical species which play a role directly or indirectly in stratospheric O₃ chemistry. Not all chemical species simulated by GEOS‐CF have observations available for validation, however an extensive list of chemical species on 3D model output are made available to the public for research purposes (Knowland et al., 2022). Comparisons against independent observations focused on the year 2020, allowing several months for the stratosphere to stabilize after updates were made to the GEOS‐Chem UCX module on 31 July 2019 for improved stratospheric chemistry and composition in the GEOS‐CF product. Observation suite included ozonesondes and satellites (namely ACE‐FTS, MLS, and SAGE III/ISS) to provide a general overview of the global state of the GEOS‐CF stratospheric composition. Since the GEOS‐CF replay O₃ is constrained by observations by nudging toward the GEOS FP assimilated O₃ product, it is expected to agree well with independent observations in the stratosphere. The median O₃ simulated in GEOS‐CF colocated with 20 ozonesonde locations agrees well in the stratosphere (400–10 hPa), and the median percent bias is within ±20% through most of the stratosphere. GEOS‐CF correlates well with SAGE III/ISS observations (r > 0.92) between 100 and 4.6 hPa, but near the stratopause the relationships tend to break down (r = 0.61 at 1 hPa). Overall, the spatial patterns of the GEOS‐CF simulated concentrations agree well with MLS and ACE‐FTS for chlorine (HCl and ClO) and nitrogen (HNO₃; ACE‐FTS only for N₂O, ${{\mathrm{N}\mathrm{O}}_x}^{\ast }$, and NO _y ) species.

With the inclusion of the complex chemistry in GEOS‐CF, during extremely low column O₃ events, such as occurred within the NH and SH polar vortexes of 2020, the GEOS‐CF forecasts can realistically predict key features of stratospheric O₃ variability. GEOS‐CF captures the dynamical and chemical environments of the polar vortexes since heterogeneous reactions on PSCs are represented in the GEOS‐Chem UCX mechanism. Specifically, it simulates low concentrations of HCl within the polar vortex and high concentrations of ClO within the sunlit portion, which leads to the destruction of O₃ within the vortex. While biases can exist in the initial conditions and forecasts, in situations where the bias is unimportant or can be corrected, GEOS‐CF forecasts should prove especially useful. Future development, as more years of GEOS‐CF output become available, will focus on better characterizing this bias. There is also the potential for longer, 10‐day, O₃ forecasts pending future demand.

One new development from GMAO is the expanded GEOS DAS to multi‐constituent assimilation (“CoDAS”). Demonstrated by Wargan, Weir, et al. (2020), the assimilation of stratospheric O₃, HCl, H₂O, and N₂O from MLS with a stratospheric chemistry model can offer a more realistic representation of important species related to stratospheric O₃ recovery, in particular within the polar vortex. Stratospheric H₂O in reanalysis products are historically poor (Davis et al., 2017), and without an observational constraint on H₂O above the tropopause, the GEOS‐CF stratospheric water vapor is also biased compared to independent observations from MLS and ACE‐FTS (Figure S6 in Supporting Information S1). In addition to HNO₃, water vapor is important for PSCs and other heterogeneous processes. Future developments for the GEOS‐CF system include incorporating the CoDAS system to constrain both tropospheric and stratospheric constituents. The first test will include the assimilation of stratospheric O₃ to remove the need for the O₃ nudging technique. With the assimilation of satellite‐retrieved H₂O and other stratospheric species, GEOS‐CF would likely improve on the spatial distribution of these and other related chemical species globally, and especially in and around a polar vortex.

Supporting information

Supporting Information S1

Knowland, K. E. , Keller, C. A. , Wales, P. A. , Wargan, K. , Coy, L. , Johnson, M. S. , et al. (2022). NASA GEOS Composition Forecast Modeling System GEOS‐CF v1.0: Stratospheric composition. Journal of Advances in Modeling Earth Systems, 14, e2021MS002852. 10.1029/2021MS002852

Data Availability Statement

All GEOS‐CF model output is centrally stored at the NASA Center for Climate Simulation (NCCS). Public access to these archives is provided by the GMAO at https://gmao.gsfc.nasa.gov/weather_prediction/GEOS-CF/data_access through model output access tools including OPeNDAP and Hypertext Transfer Protocol (HTTP). The SBUV merged data set is available from https://acd-ext.gsfc.nasa.gov/Data_services/merged/index.html, OMI “TOMS‐like” level 3 gridded product (Bhartia, 2012) available from https://disc.gsfc.nasa.gov/datasets/OMTO3d_003/summary, OzoneWatch is available from https://ozonewatch.gsfc.nasa.gov/, SAGE III‐ISS data is available from the NASA Langley Research Center Atmospheric Sciences Data center (https://eosweb.larc.nasa.gov/project/SAGE%20III-ISS), OMI and MLS data is available at https://disc.gsfc.nasa.gov/, and TOLNet available from https://www-air.larc.nasa.gov/missions/TOLNet/data.html. The ozonesondes are available from http://www.woudc.org and https://gml.noaa.gov/aftp/data/ozwv/Ozonesonde/. ACE‐FTS measurements are available, following registration, from http://www.ace.uwaterloo.ca/data.php, with the data quality information available at https://dataverse.scholarsportal.info/dataset.xhtml?persistentId=doi:10.5683/SP2/BC4ATC.