Advancing CH4 and N2O retrieval strategies for NDACC/IRWG FTIR observations with the support of airborne in situ measurements

Abstract

Atmospheric methane (CH₄) and nitrous oxide (N₂O) are potent greenhouse gases with significant impacts on climate change. Accurate measurement of their atmospheric abundance is essential for understanding their sources, sinks, and the impact of human activities on the atmosphere. Ground-based high-resolution Fourier-transform infrared (FTIR) observations, employed by collaborative international initiatives like the Infrared Working Group (IRWG) within the Network for the Detection of Atmospheric Composition Change (NDACC), play a vital role in retrieving the atmospheric amounts of these gases. Network-wide consistent data products rely on consistent observations and retrievals. Recent developments in spectroscopy, a priori data, and retrieval software and techniques underscore the necessity to revisit the current retrieval strategies for all NDACC/IRWG species. This study investigates various retrieval strategies of CH₄ and N₂O utilizing high-resolution FTIR observations in Boulder, Colorado, and compares them with unique airborne in situ measurements. The initial focus is on characterizing retrieval differences across spectroscopy databases. While it is challenging to identify the best retrievals purely based on spectroscopy, as they produce similar outcomes, notable differences in profile shapes and magnitudes underscore the importance of independent validation. Specifically, when multiyear independent nearby AirCore and aircraft in situ profile measurements are used to evaluate vertical distributions and biases in partial columns, they reveal excellent agreement in relative differences with FTIR retrievals and thereby strengthen confidence in the assessment. The final optimized retrievals for CH₄ and N₂O are presented, incorporating quantitative fitting results and comparisons of vertical profiles as well as partial and total columns. We find that employing a priori profiles using the latest simulations of the Whole Atmosphere Community Climate Model (WACCM) enhances accuracy relative to in situ profiles. While the HITRAN 2020 spectroscopic database is effective for N₂O, ATM 2020 provides better results for CH₄, with a slight improvement observed when paired with the water vapor line list from the German Aerospace Center (DLR); however, this improvement may be site-dependent. Regarding regularization, both first-order Tikhonov and optimal estimation produce comparable outcomes, as long as the fitted profile degrees of freedom remain between 2 and 2.5. Correspondingly, profile result comparisons yield biases of $-\mathrm{0.08}$ $\pm$ 0.38 % and 0.89 $\pm$ 0.28 % for tropospheric and stratospheric layers of CH₄ relative to AirCore, respectively, whereas they yield a bias of 0.39 $\pm$ 0.42 % for aircraft comparisons in the troposphere. For N₂O, the bias in the troposphere using aircraft measurements is approximately 0.18 $\pm$ 0.2 %. Uncertainty budgets combining random and systematic sources are provided. Random errors, mainly stemming from temperature profile uncertainties and measurement noise, dominate in the troposphere for both gases, with a retrieval random error of 0.5 %. Systematic errors primarily arise from HITRAN-based spectral line parameters, predominantly the line intensity and air-broadened half-width. These findings contribute to advancing our understanding of atmospheric composition and will support the improvement of a harmonized approach for all IRWG/NDACC sites.

1 Introduction

Atmospheric methane (CH₄) and nitrous oxide (N₂O) are potent greenhouse gases that significantly contribute to radiative forcing and, consequently, climate change. Accurate and precise measurement of these trace gases is essential for understanding their sources, sinks, and the impact of anthropogenic activities on the Earth's atmosphere. CH₄ is considered a short-lived greenhouse gas with an atmospheric lifetime of approximately 10 years. Despite its relatively short atmospheric lifetime compared to other greenhouse gases, CH₄ has a high global warming potential, 27 to 30 times that of CO₂ over a 100-year timescale, making it a significant contributor to near-term climate forcing. Its sources are diverse, including natural processes, such as wetlands, and anthropogenic activities, such as livestock management and fossil fuel extraction . In contrast, N₂O is classified as a long-lived greenhouse gas, with a lifetime of approximately 114 years. Although it is less abundant, N₂O exhibits a remarkable warming potential, approximately 270 times greater per metric ton than CO₂ over a 100-year time horizon. This amplifies its impact on climate change and its role in stratospheric ozone depletion . Understanding the sources and sinks of both CH₄ and N₂O is crucial for developing effective climate mitigation strategies and policies. Moreover, as both gases are intricately linked to the global carbon and nitrogen cycles, studying their behavior aids in comprehending broader ecological processes and human-induced perturbations to the Earth's atmosphere .

Ground-based Fourier-transform infrared (FTIR) spectroscopy has proven to be a reliable method for extracting atmospheric concentrations of both CH₄ and N₂O through direct-Sun solar absorption measurements. Collaborative international initiatives actively use FTIR observations to retrieve information on these species. The Infrared Working Group (IRWG) within the Network for the Detection of Atmospheric Composition Change (NDACC) has demonstrated the capability to retrieve both gases, along with numerous others, in the mid-infrared with higher spectral resolution and over the long term . Simultaneously, the Total Column Observation Network (TCCON) focuses on precision measurements of column-averaged abundances for CH₄ and N₂O in the near-infrared spectral region . Additionally, the newly formed Collaborative Carbon Column Observing Network (COCCON) has employed portable low-resolution FTIR spectrometers to measure greenhouse gases, including CH₄ and N₂O . While TCCON and COCCON have optimized their retrievals through profile scaling to derive total columns, the IRWG/NDACC is able to obtain vertical profile information with higher spectral resolution, with degrees of freedom (DOF) values of up to 2 or 3.

Presently, sites within IRWG/NDACC generally adhere to the recommendations outlined in . However, as evidenced by , , and , a significant variability exists in the final retrieval strategies employed at each site. The retrieval strategy proposed by for CH₄ involved using the HIgh-resolution TRANsmission molecular absorption database (HITRAN) 2001, which exhibited better agreement between measured and simulated spectra compared to newer databases of that time (e.g., HITRAN 2008; ). This strategy utilized three out of five possible micro-windows and incorporated Tikhonov L1 regularization, which focuses on stabilizing the solution by adding constraints to the optimization problem. In a more recent study, analyzed total columns derived from various line lists, concluding that HITRAN 2001 offers better CH₄ retrievals. However, ATM19 produced slightly different, yet comparable, results to HITRAN 2001. This conclusion is based on the optimal agreement observed between measured and simulated spectra. Considering the availability of updated spectroscopy, diverse retrieval strategies at each site, and new validation data, there is a pressing need to explore an optimal and harmonized strategy within the IRWG/NDACC. Additionally, while previous studies have focused mainly on total columns, IRWG/NDACC observations have the potential to separate both tropospheric and stratospheric columns for N₂O and CH₄. However, this has not yet been extensively validated.

This study investigates diverse retrieval strategies for CH₄ and N₂O, employing several micro-windows, a priori profiles, and regularization methods. High-resolution Fourier-transform infrared (HR-FTIR) observations in Boulder, Colorado, form the basis of our analysis . Initially, we focus on characterizing residuals, representing differences between simulated and observed spectra using different spectroscopy databases. A key aspect of our investigation involves integrating multiyear independent observations from AirCore (CH₄) and aircraft (CH₄ and N₂O) profiles obtained in close proximity to the FTIR site. These independent datasets are crucial for evaluating the vertical distribution (profile shapes) of the retrieval strategies. For CH₄, we examine the bias in tropospheric and stratospheric partial columns, whereas for N₂O, the assessment primarily focuses on tropospheric columns. Boulder offers a unique opportunity with its extensive, multiyear coverage of AirCore, aircraft, and IRWG/NDACC profiles, which sample CH₄ in close proximity, allowing for the integration of numerous observations into our analysis.

2 Observations

2.1 Ground-based FTIR

High-resolution (HR) direct solar infrared absorption spectra have been recorded at the National Science Foundation's National Center for Atmospheric Research (NSF NCAR) in Boulder (Colorado; 40.40° N, 105.24° W; 1600 m.a.s.l., meters above sea level) from 2010 to present. Measurements from 2010 to 2017 were undertaken utilizing a Bruker 120 HR-FTIR spectrometer; however, the instrument subsequently underwent an upgrade throughout a significant portion of 2018, incorporating the latest electronics and optics from the Bruker 125 HR-FTIR. Measurements are conducted under clear-sky conditions. The site officially became a part of the Infrared Working Group (IRWG) within NDACC in 2020, and the observations adhere to the standard measurement protocols established by the network. Optical band-pass filters are employed to optimize the signal-to-noise ratio (SNR) within the near- and mid-infrared spectral range. provide a list of typical band limits for each NDACC filter, along with the gases exhibiting absorption features and typically retrieved, covering a spectral range of 750–5000 cm⁻¹. For CH₄, NDACC filter 3 is used, while for N₂O, both filters 3 and 4 can be used, although filter 3 is predominantly used. The nominal spectral resolution achieved is 0.004 cm⁻¹ (maximum optical path difference (OPD) of $\approx$ 257 cm), utilizing liquid-nitrogen-cooled indium antimonide (InSb) and mercury cadmium telluride (MCT) detectors and a potassium bromide (KBr) beam splitter. For a detailed description of the FTIR operations at Boulder, see .

Figure 1

(a) Individual AirCore CH₄ profiles in Boulder, Colorado, with different colors indicating fractional years between 2018 and 2022. (b) Mean vertical profile of CH₄ accompanied by the standard deviation, with the red circle at the bottom representing the mean volume mixing ratio from an in situ sensor near the HR-FTIR (see text for details). A total of 61 vertical profiles are included in these plots.

[Figure omitted. See PDF]

Table 1

Summary of the five micro-windows tested in the retrieval of CH₄.

Micro-windows (cm−1)		Main species
1	2613.70–2615.40	CH4, HDO, CO2
2	2650.60–2651.30	CH4, HDO, CO2
3	2835.50–2835.80	CH4, HDO
4	2903.60–2904.03	CH4, NO2, HDO, H2O
5	2921.00–2921.60	CH4, H2O, HDO, NO2

The spectra undergo analysis using the SFIT4 algorithm, an advancement from SFIT2 , to extract profiles of CH₄ and N₂O. Key input and common parameters for the retrieval of all gases using SFIT4 include vertical profiles of pressure, temperature, and volume mixing ratios (VMRs) of the atmospheric gases absorbing in the specific micro-windows. The input pressure and temperature vertical profiles are obtained from National Center for Environmental Prediction (NCEP) and are standardized for use in all NDACC retrievals. These are daily average profiles extending up to approximately 50 km. Above that, we use monthly mean pressure and temperature profiles from simulations using the Whole Atmosphere Community Climate Model (WACCM) . As an initial step (and to mitigate water vapor interference), the retrieval of water vapor is conducted first. Subsequently, this water vapor information is utilized in the retrieval process for CH₄ and N₂O to prevent potential bias arising from inaccuracies in the water vapor profile shape . Section  and describe the respective specific retrievals for CH₄ and N₂O in more detail.

Table 2

Different strategies tested in the retrieval of CH₄.

Retrieval	Micro-windows	Spectroscopy details	Regularization	Additional details*
r1	1, 3, 5	HIT00	Tikhonov L1	Baseline following
r2	1, 3, 5	ATM20	Tikhonov L1	ATM20 for all gases
r3	1, 3, 5	ATM20 and HIT00 (CH4)	Tikhonov L1	Same as r2 but HIT00 for CH4
r4	1, 2, 3, 4, 5	ATM20 and HIT00 (CH4)	Tikhonov L1	Five micro-windows
r5	1, 2, 3, 4, 5	ATM20	Tikhonov L1	Five micro-windows and ATM20 for all gases
r6	1, 2, 3, 5	ATM20	Tikhonov L1	Four micro-windows and ATM20 for all gases
r7	1, 2, 3	ATM20	Tikhonov L1	Three micro-windows and ATM20 for all gases
r8	1, 2, 3, 4, 5	ATM20 and DLR (H2O)	Tikhonov L1	DLR water vapor and ATM20 all others
r9	1, 2, 3, 4, 5	ATM20 and DLR (H2O)	OEM	Same as r8 but using OEM
r10	1,2,3,4,5	ATM20	Tikhonov L1	Same as r5 but using monthly mean WACCM6

^* OEM denotes optimal estimation.

Figure 2

(a) NOAA aircraft profiles of CH₄ at Briggsdale, Colorado, with different colors indicating fractional years between 2018 and 2022. (b) The mean and standard deviation of CH₄ profiles derived from the NOAA aircraft, along with the surface CH₄ volume mixing ratio (VMR) from the Boulder reservoir, including the standard deviation. Panel (c) is the same as panel (a) but for N₂O. Panel (d) is the same as panel (b) but for N₂O. A total of 81 vertical profiles are included in these plots.

[Figure omitted. See PDF]

2.2 AirCore CH₄ profiles

We employ CH₄ AirCore data acquired from the National Oceanic and Atmospheric Administration (NOAA) Global Monitoring Laboratory (GML) to examine and assess the vertical distribution of CH₄ in the troposphere and stratosphere. Regular monthly AirCore flights are initiated in Boulder, Colorado; many of these flights occur at approximately 13:00 LT (local time), to synchronize with the OCO-2/A-train satellite overpass, with air samples collected anywhere from 17 to 121 km near the FTIR site (measured from the descent point). The AirCore technique employed by NOAA involves collecting air samples within a stainless-steel tube (approx. 100 m in length and 1 cm in diameter). This tube passively collects atmospheric air during balloon descents from the middle stratosphere to the ground. Subsequently, the collected air is analyzed using a cavity ring-down spectrometer, ensuring a trace mole fraction precision of better than 0.4 ppb for CH₄ . Because storage of the air sample in the tubing does not diffuse quickly between the balloon payload landing and analysis, approximately 100 discrete samples are measured in the AirCore tubing during a 4 h storage time, offering detailed insights into the vertical distribution of CH₄, carbon monoxide (CO), and carbon dioxide (CO₂). For our study, we leverage Level-2 CH₄ from the current NOAA AirCore dataset v20230831, which includes dry-air mole fraction profiles retrieved since 2012 . Our investigation includes profiles collected from 2018 to 2022, enabling the evaluation of different retrieval strategies using a substantial set of coincident FTIR, AirCore, and aircraft observations.

Figure 3

Retrieval fit example of CH₄ for 18 November 2019 with a solar zenith angle of 60.0°. Spectral contributions for various species are displayed above each micro-window, spanning from micro-window 1 (a) to micro-window 5 (e), employing Tikhonov regularization and WACCM6 as the a priori. The observed and fitted lines are represented using blue and green, respectively. The bottom plots show the residuals (observed minus fitted), with distinct colors indicating different spectroscopy types as labeled. The numbers within the labels denote the root-mean-square fitting for each micro-window across various spectroscopy databases.

[Figure omitted. See PDF]

Figure a shows individual CH₄ AirCore profiles obtained in Boulder, Colorado, from 2018 to 2022. As expected, there is a noticeable rise in the CH₄ mole fraction from 2018 onwards, reflecting the increase in the global background (https://gml.noaa.gov/ccgg/trends_ch4/, last access: 20 May 2025). These profiles exhibit distinct features, such as elevated near-ground values in certain instances, suggestive of local sources. Further, within the boundary layer and free troposphere, the profiles demonstrate well-mixed characteristics with some variations in the upper troposphere and the lowest stratosphere of the AirCore profiles, culminating in a sharp decrease at the tropopause. Figure b presents the mean vertical profiles from 2018 to 2022. To derive the mean profile, the individual profiles are regridded to common altitude scheme levels within a range from 1.6 to 25 km, and mean values are then estimated within a 0.2 km resolution. Ultimately, mean and standard deviation profiles are calculated. The shaded blue area in Fig. b represents the standard deviation in the AirCore profile dataset used here, while the red circle at the bottom represents the mean values from an in situ sensor located at the Boulder reservoir about 5 km from the HR-FTIR using the overlap period between 2020 to 2022 (https://www.bouldair.com/boulder.htm, last access: 20 May 2025), providing insight into the variability around the Boulder region.

Table 3

Summary of the four micro-windows used in the retrieval of N₂O.

Micro-windows		Main species
1	2481.30–2482.60	N2O, H2O, CO2, CH4
2	2526.40–2528.20	CH4, N2O, CO2, HDO, CH4
3	2537.85–2538.80	N2O, CH4, HDO
4	2540.10–2540.70	N2O

Table 4

Different strategies tested in the retrieval of N₂O.

Retrieval	Spectroscopy details	Regularization	Additional details
r1*	HIT08	OEM	Filter 3 and WACCM4
r2	HIT20	OEM	Filter 3 and WACCM4
r3	ATM20	OEM	Filter 3 and WACCM4
r4	HIT20	OEM	Filter 4 and WACCM4
r5	HIT08	OEM	Filter 3 and WACCM6
r6	HIT20	Tikhonov L1	Filter 3 and WACCM6
r7	HIT20	Tikhonov L1	Filter 4 and WACCM6
r8	HIT20	Tikhonov L1	Filters 3, 4, and WACCM6

^* Current retrieval strategy as suggested by the IRWG.

Table 5

Overview of the mean root-mean-square (RMS) values across all five micro-windows (MWs) and the overall RMS for the CH₄ retrieval using various spectroscopy databases for 2021. The last column displays the degrees of freedom (DOF) values. Values in parentheses are the standard deviation.

	RMS
Spectroscopy	MW1	MW2	MW3	MW4	MW5	Overall	DOF
HIT00	0.140 (0.022)	0.125 (0.022)	0.160 (0.038)	0.169 (0.033)	0.168 (0.038)	0.156 (0.030)	2.46 (0.15)
HIT20	0.148 (0.024)	0.164 (0.035)	0.115 (0.016)	0.205 (0.042)	0.504 (0.058)	0.270 (0.039)	2.31 (0.13)
ATM20	0.139 (0.018)	0.122 (0.017)	0.116 (0.018)	0.204 (0.036)	0.183 (0.032)	0.161 (0.025)	2.42 (0.16)
ATM20_DLRWV	0.140 (0.017)	0.134 (0.022)	0.117 (0.018)	0.206 (0.039)	0.210 (0.051)	0.170 (0.032)	2.41 (0.16)
ATM20_HIT00CH4	0.143 (0.022)	0.120 (0.019)	0.160 (0.037)	0.164 (0.031)	0.142 (0.024)	0.150 (0.027)	2.47 (0.14)

2.3 Aircraft-based observations of CH₄ and N₂O profiles

To further evaluate CH₄ and N₂O retrievals, we use the vertical profiles obtained at a remote location near Briggsdale (40.6347° N, 104.3269° W), Colorado, approximately 100 km northeast of the HR-FTIR, as part of the NOAA Earth System Research Laboratories (ESRL) Global Greenhouse Gas Reference Network Aircraft Program (https://gml.noaa.gov/ccgg/aircraft/, last access: 20 May 2025). This program employs contracted aircraft equipped with instruments to collect 12 discrete air samples in vertical profiles, primarily over North America, and monitor over 50 trace gases throughout the boundary layer and free troposphere over the long term. The network provides crucial insights into the sources, sinks, and transport mechanisms of greenhouse and other trace gases. Additionally, these profiles are a resource for the model evaluation of vertical mixing and independently constrain the mean biospheric uptake of CO₂ throughout the annual cycle. The routine data collection within the program further facilitates the evaluation of satellite retrievals, which is often complemented by ground-based remote-sensing observations. Figure  illustrates the CH₄ and N₂O aircraft profiles acquired at Briggsdale, Colorado, spanning from 2018 to 2022. In line with CH₄ AirCore profiles, there is minimal variation observed in the profiles up to 6 km, with occasional enhancements near the ground on certain days. In the case of N₂O, there is a clear increase from 2018 onwards, also consistent with the increase in the global background (https://gml.noaa.gov/ccgg/trends_n2o/, last access: 20 May 2025). The N₂O profiles exhibit a well-mixed pattern, with no notable enhancements near the ground. Similar to Fig. , the mean profiles are presented on the right side for each gas, depicting mainly the increase in the global background. However, the mean profile is displayed within the altitude range of 1.6 to 7 km a.s.l., with a resolution of 0.4 km due to the original vertical resolution being lower than that of the AirCore.

3 Retrieval strategies

3.1 CH₄

Multiple retrieval strategies have been devised, encompassing various combinations of micro-windows, spectroscopy, and regularization techniques. The fitting results are employed to assess different spectroscopy databases, and all retrieval strategies are compared with AirCore and aircraft observations to determine the most effective approach.

An optimized selection of micro-windows (MWs) in the mid-infrared has been carried out by , and a summary of these retrieval strategies is provided in Table . In alignment with , we employed the same five potential micro-windows for our analysis, as detailed in Table .

The molecular absorption databases employed in this study include the latest official release of HITRAN 2020 (HIT20; ); the ATM 2020 (ATM20) spectroscopic database used by TCCON , which is available at https://mark4sun.jpl.nasa.gov/toon/linelist/linelist.html (last access: 22 December 2023); and HITRAN 2000 (HIT00; including the 2001 updates), as recommended by . In their study, also tested HITRAN 2008, which was the most recent database at the time, and found significant residuals with HITRAN 2008, whereas HIT00 showed much lower residuals. More recently, examined the impact of different databases and similarly concluded that HIT00 provides superior fitting results, with the exception of ATM (ATM19 at the time). Additionally, we utilized combinations of these databases with others for specific purposes. For instance, water vapor spectroscopy is based on the line list provided by the German Aerospace Center (DLR) , while ATM20 is used in conjunction with the HIT00 line list for CH₄. Section  provides a more detailed overview of the various spectroscopy databases and their corresponding results.

Figure 4

Retrieval fit example of N₂O on 31 January 2019 with a solar zenith angle of 63.8°. Spectral contributions for various species are displayed above each micro-window, spanning from micro-window 1 (a) to micro-window 4 (d), employing Tikhonov regularization, WACCM6 as the a priori, and filter 4. The observed and fitted lines are represented using blue and green, respectively. The bottom plots show the residuals (observed minus fitted), with distinct colors indicating different spectroscopy types as labeled. The numbers within the labels denote the root-mean-square fitting for each micro-window across various spectroscopy databases.

[Figure omitted. See PDF]

For the current IRWG gas retrievals, mean a priori profiles were primarily obtained using WACCM version 4 for the 1980–2020 period (referred to as WACCM4). WACCM simulations underwent enhancements in physical, chemical, and aerosol parameterizations , resulting in version 6 (WACCM6), and new mean a priori profiles have been created spanning the 1980–2040 period (https://wiki.ucar.edu/pages/viewpage.action?pageId=452264118, last access: 20 May 2025). Comparing CH₄ between these two WACCM versions at different sites (e.g., Mauna Loa, Hawaii; Thule, Greenland) with independent profiles obtained during the airborne Pole-to-Pole Observations (HIPPO; https://www.eol.ucar.edu/field_projects/hippo, last access: 20 May 2025; ) campaign and the Atmospheric Tomography Mission (ATom; https://daac.ornl.gov/ATOM/campaign/, last access: 20 May 2025; ) field campaigns indicates that the latest version-6 profiles are more suitable for recent years (not shown). To streamline the retrieval process, all retrievals utilized the most recent version of the WACCM profiles. For regularization, the process of adding constraints or penalties to the optimization problem to ensure a more stable and reliable solution, we employed both the first-order Tikhonov regularization, following the recommendation of , and the conventional optimal estimation (OEM) approach. In Tikhonov regularization, the regularization strength ($\mathit{\alpha }$) is optimized to maintain the degrees of freedom (DOF) value at approximately 2, with potential variations depending on conditions. Optimization is achieved through the typical L-curve method, as proposed by . The Tikhonov regularization procedure is directly implemented in the latest version of SFIT4, accessible at https://wiki.ucar.edu/display/sfit4/SFIT+Core+Code (last access: 22 December 2023). For the OEM approach, regularization involved diagonal a priori covariance matrices. Variance magnitudes were empirically tuned to prevent profile oscillations and, similarly, to achieve a target DOF of around 2. The OEM technique has the advantage that the retrieved vertical profile relaxes back to the a priori profile shape, as spectral information diminishes with altitude. In contrast, the Tikhonov constraint tends to preserve the deviation from the a priori at higher altitudes where spectral information is still present. Additional details on the retrieval workflow in Boulder are provided in .

Table 6

Summary of the RMS values for N₂O retrieval in 2021, encompassing all five micro-windows (MWs) and the overall RMS. The last column presents the DOF values. Values in parentheses are the standard deviation.

	RMS
Spectroscopy	MW1	MW2	MW3	MW4	Overall	DOF
HIT08	0.149 (0.031)	0.130 (0.019)	0.133 (0.029)	0.132 (0.032)	0.148 (0.030)	2.56 (0.11)
HIT20	0.148 (0.031)	0.131 (0.019)	0.131 (0.028)	0.128 (0.032)	0.147 (0.029)	2.57 (0.11)
ATM20	0.149 (0.031)	0.130 (0.019)	0.133 (0.029)	0.131 (0.032)	0.148 (0.029)	2.57 (0.11)

Figure 5

(a) Illustrative comparison between FTIR and AirCore CH₄ profiles on 14 May 2019, utilizing ATM20, WACCM6, and Tikhonov regularization (retrieval r5 in Table ). Various colors denote different HR-FTIR retrieval times (UT), with the high-resolution AirCore profile presented in black and the smoothed profile in blue. (b) The relative difference (FTIR minus AirCore) for each corresponding time.

[Figure omitted. See PDF]

We establish a baseline retrieval strategy (r1) by employing three micro-windows (1, 3, and 5), along with the first-order Tikhonov regularization approach, and the HIT00 spectroscopy database . A total of 10 retrieval strategies have been developed to explore various approaches and determine whether different methods impact the results when compared with AirCore and aircraft profiles. As shown in Sect. , HIT20 was excluded due to its significantly higher residuals in the fitting results. Detailed information on each retrieval strategy is provided in Table .

3.2 N₂O

The N₂O retrievals suggested by the IRWG involve the use of four micro-windows within the 2480–2540 cm⁻¹ spectral range. The specific micro-windows and associated interfering species are detailed in Table . Our focus is not directed towards the identification or refinement of micro-windows, as they are well-harmonized across the network, yielding satisfactory results; instead, our objective is to assess the performance of the latest spectroscopy line parameters. Similarly, the spectroscopy evaluation includes testing HITRAN 2008 (HIT08) (the current official IRWG recommendation), HIT20, and ATM20. Within the retrieval strategies outlined in Table , we incorporate both the Tikhonov L1 and OEM approaches, as the regularization approach is not uniformly harmonized . Additionally, two optical band-pass filters, denoted as filters 3 and 4 within the IRWG, are available for utilization in the N₂O retrieval process. In this case, we conducted separate N₂O retrievals using each filter independently, as well as a retrieval approach in which both filters were used independently, but the results were combined without distinguishing between filters. In this instance, we incorporate retrieval strategies utilizing both WACCM4 and v6, as there is a comparatively lower number of sensitivity studies compared to CH₄.

Figure 6

Relative differences in CH₄ profiles, calculated as the difference between FTIR and AirCore values divided by the AirCore values, employing diverse retrieval strategies. The mean and median profiles are shown using blue and red lines, respectively. The green lines illustrate the relative differences obtained in the lower troposphere using aircraft comparisons, as shown in Fig. .

[Figure omitted. See PDF]

4 Results

4.1 Intercomparison among spectroscopy databases

This section addresses the quality of fits resulting from various spectroscopy databases. Specifically, we compare five databases for CH₄: (1) HIT00, (2) HIT20, (3) ATM20, (4) ATM20 with DLR water vapor (ATM20_DLRWV), and (5) ATM20 with HIT00 for CH₄ (ATM20_HIT00CH4). To maintain the focus on spectroscopy, all five retrievals follow the same procedure, employing Tikhonov regularization and WACCM6 as the a priori, with the only variation being the choice of spectroscopy. Figure  visually shows the spectral contributions of various species during the CH₄ retrieval across the five micro-windows. As an example, the top panel displays these contributions for each micro-window on 18 November 2019 at a solar zenith angle of 60.0°. The observed and fitted lines, presented in blue and green, respectively, provide insights into the quality of the retrieval. In the corresponding bottom plots, the residuals (observed minus fitted) are presented, with distinct colors indicating the different spectroscopy databases. There are clear differences for some micro-windows. For instance, HIT20 exhibits substantial systematic residuals for micro-windows 2 and 5. The large residual at 2921.33 cm⁻¹ in HIT20 for micro-window 5, previously identified by for HIT08, suggests that HIT20 still needs to be improved for this line strength. To further identify differences, the numbers within the labels provide the root-mean-square fitting for each micro-window across diverse spectroscopy databases. Micro-window 1 stands out as the only one for which all databases perform similarly, while others exhibit variations. Nonetheless, ATM20 demonstrates slightly better results within micro-window 1. Apart from HIT20, which displays high residuals in micro-window 2, all other databases show similar residuals. In micro-window 3, ATM20, ATM20_DLRWV, and HIT20 produce improved results, whereas in micro-window 4, lower residuals are observed with ATM20_HIT00CH4, HIT20_HIT00CH4, and HIT00, indicating that HIT00 remains the preferred choice for this micro-window. As previously noted, HIT20 continues to show large residuals in micro-window 5, whereas HIT00 still provides better residuals. A summary of root-mean-square values (RMS) for all micro-windows for a full year (2021) of retrievals is provided in Table . Overall, HIT20 displays larger RMS values, suggesting it may not be an optimal choice, while all other spectroscopy databases show similar RMS values within the standard deviation. These findings align with , who found that HIT00 yields better RMS values but is similar to ATM20. However, there are notable differences in the profile shapes and magnitudes (see Sect. ), highlighting the importance of the assessment conducted with independent observations.

Figure 7

Relative difference between FTIR and NOAA aircraft CH₄ profiles using several retrieval strategies.

[Figure omitted. See PDF]

For N₂O, we compare the HIT08, HIT20, and ATM20 databases. Our objective is to assess whether HIT20 or ATM20 exhibits better residual quality than HIT08, currently employed in the IRWG/NDACC strategy. Figure  illustrates the spectral contributions of various species during the N₂O retrieval across four micro-windows. Similar to CH₄, the top panels showcase an example, specifically on 31 January 2019, at a zenith angle of 63.8°. The bottom residual plots do not reveal a significant difference among the three databases. While the RMS suggests a slight improvement with HIT20, an overview table (Table ) presenting results for the entire year (2021) indicates no significant improvement. Nonetheless, this suggests that either HIT20 or ATM20 could be considered in the IRWG recommendation.

4.2 Intercomparison with AirCore and aircraft profiles

Between 2018 and 2022, a total of 61 AirCore sampling systems were launched in Boulder, 36 of which were coincident with FTIR measurements. The total number of coincident dates between FTIR and aircraft profiles is 51. To enable a quantitative comparison of in situ profiles and partial columns, vertical profile measurements from AirCore and aircraft are regridded onto the FTIR retrieval's altitude grid using first-order spline interpolation. Subsequently, these profiles are smoothed with the daily mean FTIR averaging kernels to account for the FTIR altitude sensitivity . Figure a presents an illustrative comparison between FTIR profiles obtained on a given day and an AirCore profile. The high-vertical-resolution AirCore profile is depicted using black triangles, while the smoothed AirCore profile is shown using blue triangles. FTIR retrievals from the same day are represented by circles in various colors. In Fig. b, the relative difference in percentage (or bias), defined as the difference between FTIR and AirCore values divided by the AirCore values, is shown for each corresponding time. To maximize the number of profile comparisons (and considering the relatively stable nature of CH₄ and N₂O), we compare FTIR profiles obtained within a 6 h time window of the AirCore descent time. The mean distance between the AirCore descent location and the FTIR site is 76.8 $\pm$ 26.5 km, while the mean distance between the FTIR site and the AirCore landing point is 101.1 $\pm$ 28.5 km. The same approach is undertaken when comparing with aircraft observations but the aircraft profiles generally extend up to around 6 km a.s.l., whereas the AirCore profiles extend up to 20 km or higher. As mentioned in Sect. , aircraft observations are conducted approximately 100 km northeast of the FTIR site.

Figure 8

(a) Averaging kernel for CH₄, representative of 18 November 2019, at a solar zenith angle of 60.0°. Panel (b) shows the total column averaging kernel in blue and the cumulative sum of the degrees of freedom (DOF) values in black. The shaded rectangles indicate the cumulative sum of the DOF values. The first DOF spans from 1.6 to 9.8 km, whereas the second DOF ranges from 9.8 to 22.1 km. This example employs five micro-windows, Tikhonov regularization, and WACCM6 as the a priori.

[Figure omitted. See PDF]

4.2.1 CH₄

Figure  shows the relative differences in the CH₄ vertical profiles obtained for each retrieval strategy when compared with AirCore profiles. The subpanel names correspond to the retrievals listed in Table . The individual profile relative differences are displayed in gray, while the mean and median profiles are depicted using red and blue, respectively. To highlight the similarity between AirCore and aircraft profile results, the median relative differences in the troposphere from aircraft profiles are displayed in green in the same figure. The relative differences with aircraft profiles in the troposphere (up to 6 km) are further detailed in Fig. . Despite the absence of coincidence between AirCore and aircraft profiles, the relative differences exhibit remarkable similarity with respect to both magnitude and shape, enhancing confidence in the assessment of multiple retrievals. It is interesting to observe that changing only the CH₄ database spectroscopy leads to a substantial change in the magnitude of profile differences. For example, retrievals r1 and r2 show opposite signs in their relative differences within both the troposphere and stratosphere, despite having identical retrieval parameters, apart from the spectroscopy. The key distinction between retrievals r1 and r2 is that r1 employs HIT00, whereas r2 relies on ATM20. Interestingly, using a combination of ATM20 and HIT00 for CH₄ yields profile differences that are very similar to those obtained using only HIT00, suggesting that the line parameters for CH₄ primarily dictate the profile shape. The inclusion of three or five micro-windows does not change the shape of the profile difference; however, r5, which uses five micro-windows, shows smaller relative differences in both the troposphere and stratosphere compared to r2, which uses the same settings but with only three micro-windows. Retrieval strategies showing favorable bias outcomes are encompassed in r7 to r10. Strategy r7 utilizes micro-windows 1, 2, and 3, while r8 employs all five micro-windows, deviating from the micro-windows 1, 3, and 5 recommended by . We observe slightly higher variability in r7, whereas r8 and r9 exhibit more compact variability, particularly in the troposphere. Retrieval r8 is identical to r9, except that r9 uses the OEM approach. Additionally, using monthly mean CH₄ a priori profiles does not significantly improve biases, as seen when comparing r10 to r5. It is important to note that past studies have predominantly focused on total column analysis. In such cases, retrieval strategies might produce biases in the troposphere and stratosphere with opposite signs, potentially canceling each other out and reducing the overall bias in total column measurements. In contrast, this work focuses on examining the differences at various altitudes within the vertical profiles, which, due to vertical sensitivity, ultimately lead to distinct biases in tropospheric and stratospheric columns, as well as in total column measurements.

Table 7

CH₄ bias results among the different retrieval strategies for the different partial columns using AirCore profiles as the reference. The results in bold indicate the best performance.

Retrieval no.	Partial column (km a.s.l.)	Mean bias (%)	Median bias (%)	CI (%)
r1	1.6–10.0	0.88	0.84	0.31
	10.0–20.0	$-\mathrm{1.39}$	$-\mathrm{1.26}$	0.32
	1.6–20.0	0.48	0.34	0.23
r2	1.6–10.0	$-\mathrm{0.97}$	$-\mathrm{1.32}$	0.36
	10.0–20.0	1.91	1.99	0.27
	1.6–20.0	$-\mathrm{0.46}$	$-\mathrm{0.66}$	0.27
r3	1.6–10.0	1.79	1.70	0.30
	10.0–20.0	$-\mathrm{1.70}$	$-\mathrm{1.60}$	0.32
	1.6–20.0	1.18	1.15	0.23
r4	1.6–10.0	1.71	1.80	0.28
	10.0–20.0	$-\mathrm{2.37}$	$-\mathrm{2.30}$	0.36
	1.6–20.0	0.99	1.05	0.21
r5	1.6–10.0	$-\mathrm{0.64}$	$-\mathrm{0.49}$	0.27
	10.0–20.0	1.20	1.19	0.25
	1.6–20.0	$-\mathrm{0.32}$	$-\mathrm{0.18}$	0.21
r6	1.6–10.0	$-\mathrm{0.75}$	$-\mathrm{0.99}$	0.34
	10.0–20.0	1.86	1.99	0.28
	1.6–20.0	$-\mathrm{0.29}$	$-\mathrm{0.40}$	0.26
r7	1.6–10.0	0.97	0.52	0.61
	10.0–20.0	0.52	1.00	0.60
	1.6–20.0	0.89	0.61	0.43
r8	1.6–10.0	$\mathbf{-}$0.08	$\mathbf{-}$0.17	0.38
	10.0–20.0	0.89	0.86	0.28
	1.6–20.0	0.09	0.12	0.29
r9	1.6–10.0	$\mathbf{-}$0.15	$\mathbf{-}$0.18	0.37
	10.0–20.0	1.19	1.26	0.33
	1.6–20.0	0.08	0.05	0.28
r10	1.6–10.0	$-\mathrm{0.64}$	$-\mathrm{0.49}$	0.27
	10.0–20.0	1.19	1.07	0.26
	1.6–20.0	$-\mathrm{0.32}$	$-\mathrm{0.18}$	0.21

The retrieval methodology implemented in SFIT4 encompasses the vertical profile but is confined in sensitivity to specific altitude ranges. Figure a shows typical averaging kernels (row kernels) employed in the CH₄ retrieval, while Fig. b portrays the total column averaging kernel and the cumulative sum of the DOF values, denoting the number of independent layers in the retrieved profile. Shaded rectangles separate the cumulative sum of the DOF values, with the first DOF spanning 1.6–9.8 km and the second DOF ranging from 9.8 to 22.1 km. Note that all profile regularization constraints of the retrieval strategies have been adjusted to attain similar DOF values, although averaging kernels may differ, particularly between those utilizing Tikhonov and OEM regularization due to the inherent nature of the approaches. To comprehensively evaluate these layers, we conduct a comparative analysis with AirCore and aircraft data, using mixing ratios weighted by air mass density within the specified altitude ranges (see , for details on weighted mixing ratios). Tables  and  list the mean and median CH₄ relative differences among various retrieval strategies for different partial columns using AirCore and aircraft profiles, respectively. The last column represents the 95 % confidence interval (CI) calculated. As indicated earlier, tropospheric biases estimated from both AirCore and aircraft observations reveal congruent findings. Strategy r8 demonstrates satisfactory performance, with mean biases of $-\mathrm{0.08}$ $\pm$ 0.38 % and 0.89 $\pm$ 0.28 % for the tropospheric and stratospheric layers, respectively, whereas it shows a mean bias of 0.39 $\pm$ 0.42 % for aircraft comparisons in the troposphere. Strategy r9, which is identical to r8 but employs the OEM approach, produces very similar results. Additionally, we assessed the a priori profile by comparing it with AirCore and aircraft profiles. When evaluating WACCM6 across different partial columns, the biases are $-\mathrm{1.9}$ $\pm$ 1.0 % and $-\mathrm{1.5}$ $\pm$ 1.3 % for the troposphere and stratosphere, respectively. Retrievals yield improved and lower biases, with r8 and r9 showing significant reductions.

Table 8

CH₄ bias results among the different retrieval strategies in the lower troposphere using aircraft profiles as the reference. The results in bold indicate the best performance.

Retrieval no.	Mean bias (%)	Median bias (%)	CI (%)
r1	1.90	1.92	0.32
r2	$-\mathrm{0.76}$	$-\mathrm{0.79}$	0.40
r3	2.89	2.88	0.34
r4	2.84	2.84	0.31
r5	$-\mathrm{0.26}$	$-\mathrm{0.10}$	0.36
r6	$-\mathrm{0.47}$	$-\mathrm{0.43}$	0.39
r7	1.08	1.24	0.64
r8	0.39	0.30	0.42
r9	0.35	0.41	0.41
r10	$\mathbf{-}$0.24	$\mathbf{-}$0.22	0.37

4.2.2 N₂O

To evaluate N₂O, we employ a comparable approach but use only low- to mid-tropospheric values from aircraft profiles. Figure  shows the vertical profile differences across all retrieval strategies, and a summary of tropospheric weighted mixing ratio biases is presented in Table . In the current retrieval strategy (r1) recommended by the IRWG, there is a systematic positive bias of 1.58 $\pm$ 0.28 % at all tropospheric altitudes. Transitioning to the latest HIT20 spectroscopy (r2) reduces the overall tropospheric bias to 0.12 $\pm$ 0.27 %, although the profile's relative differences shift from negative values below 3 km to positive values above 3 km. Interestingly, employing ATM20 (r3) also results in a systematic positive bias of 1.42 $\pm$ 0.27 %, similar to HIT08. Similar outcomes are observed when using both filters independently. Switching to WACCM6 yields a slight improvement. For example, in retrieval r5, which utilizes HIT08 with the updated WACCM6 profile, the difference is 1.27 $\pm$ 0.28 %. The combination of HIT20 and WACCM6 further improves the comparison, as seen in retrievals r6 ($-\mathrm{0.02}$ $\pm$ 0.25 %) for filter 3 and r7 (0.36 $\pm$ 0.22 %) for filter 4, both employing the Tikhonov approach. Retrieval r8, which uses both filters, exhibits minimal bias (0.18 $\pm$ 0.20 %), demonstrating the feasibility of using filters 3 and 4 across the IRWG/NDACC network. Additionally, combining both filters maximizes the number of available retrievals. The OEM retrieval approach was also tested with HIT20 and WACCM6, yielding similar results to the Tikhonov approach. Specifically, the OEM-based retrieval showed a bias of 0.02 $\pm$ 0.23 %, which falls within the uncertainty of r8 using the Tikhonov approach. The averaging kernels and the cumulative sum of degrees of freedom (DOF) values for N₂O are shown in Fig. . The first DOF is captured from the surface to about 8 km, whereas the second DOF is observed between 8 and 17 km. Because N₂O appears to be well mixed among all tropospheric altitudes, the results shown above are representative of the entire troposphere. Unfortunately, long-term stratospheric AirCore N₂O data are not yet available in the same fashion as CH₄ to evaluate the stratospheric component using the same approach.

Figure 9

Relative difference between FTIR and NOAA aircraft N₂O profiles using several retrieval strategies.

[Figure omitted. See PDF]

Table 9

N₂O bias results among the different retrieval strategies in the lower troposphere using aircraft profiles as the reference. The results in bold indicate the best performance.

Retrieval no.	Mean bias (%)	Median bias (%)	CI (%)
r1	1.58	1.62	0.28
r2	0.12	0.15	0.27
r3	1.42	1.50	0.27
r4	0.56	0.54	0.26
r5	1.27	1.35	0.28
r6	$\mathbf{-}$0.02	0.15	0.25
r7	0.36	0.36	0.22
r8	0.18	0.30	0.20

Figure 10

(a) N₂O FTIR mean row averaging kernels. (b) Total column averaging kernel and the cumulative sum of the DOF values on 31 January 2019 using five micro-windows, Tikhonov regularization, and WACCM6 a priori profiles. The first DOF is between 1.6 and 8 km, whereas the second DOF is between 8.0 and 17.12 km.

[Figure omitted. See PDF]

4.3 Recommended retrieval strategies and error budget

The final suggested retrievals for CH₄ and N₂O are presented in Table , based on both fitting results using various spectroscopy line parameters and, especially, on the assessment of vertical profiles and columns using independent AirCore and aircraft profiles. These retrievals correspond to r8 and r9 for CH₄ (Table ), which are identical except that r8 employs Tikhonov regularization, whereas r9 uses OEM. For N₂O, retrievals follow the r8 strategy (Table ). Although Table  lists DLR water vapor for CH₄ retrievals, we believe that a standardized approach using ATM20 for all gases would provide satisfactory results, while also being more practical and easier to implement on a broader scale. SFIT4 provides uncertainty budgets that combine random and systematic sources following the formalism outlined in . The primary random and systematic error profiles for CH₄ and N₂O are illustrated in Figs.  and , respectively. These vertical profile uncertainties are expressed as percentages relative to the mean mixing ratio, with the total respective errors represented by dotted black lines. The error calculation in SFIT4 is detailed in , with a specific analysis for water vapor at Boulder. The dominant random error in the troposphere for both gases stem from uncertainties in the temperature profile and measurement noise characterized by the SNR in the spectral region of interest. Additional error parameters, such as interfering species, the apodization function, and the solar zenith angle, have a comparatively lesser impact. Major systematic error components arise from absorption line parameters, specifically line intensity ($S$), air-broadened half-width ($\mathit{\gamma }$), and the temperature dependence of $\mathit{\gamma }$ ($n$). The lower-limit uncertainties reported in HIT20 are used for systematic error retrieval, with values of 5 %, 5 %, and 10 % for CH₄ and 2 %, 10 %, and 5 % for N₂O for $S$, $\mathit{\gamma }$, and $n$, respectively. It should be noted that these errors are expected to be significantly smaller for CH₄ due to the use of the empirically modified line list ATM20. Smoothing error is treated separately and excluded from the total error analysis, as it is often not well-known and is, thus, frequently simplified. Tables  and  summarize the mean random, systematic, and total uncertainties in the retrieval of CH₄ and N₂O for three partial columns. These uncertainties are expressed as percentages relative to the mean weighted mixing ratio of the corresponding partial column (using weighted mean errors) and the total column errors. Random error, associated with precision or statistical uncertainty, reflects the inherent variability in measurements and is crucial for meeting precision requirements in observations for CH₄, such as growth rate, seasonal cycle amplitudes, and interhemispheric gradients. This type of uncertainty is typically quantified as the standard deviation or standard error of a set of measurements. For example, optimized precision to about 0.3 % using $\mathrm{1}\mathit{\sigma }$ standard deviation and diurnal variation in a 7 min integration, aligning with the retrieval random error of 0.5 % shown in Table .

Table 10

Suggested CH₄ and N₂O final retrieval strategies based on fitting results and assessment with AirCore and aircraft profiles.

	CH4	N2O
Micro-windows	1, 2, 3, 4, 5 (Table )	1, 2, 3, 4 (Table )
Spectroscopy	ATM 2020 and DLR (H2O)	HIT20
A priori	WACCM6 (all main species)
Retrieval constraint	Tikhonov L1 or OEM
IRWG filter	Filter 3	Filters 3 and 4
Additional details	Regularization strength optimized for DOF values of $\approx \mathrm{2.0}$–2.5
	pre-retrieved H2O, and 257 cm OPD
	Pressure and temperature from NCEP (at least daily)

Figure 11

Mean vertical profiles of the primary random (a) and systematic (b) uncertainty components in the retrieval of CH₄ for the year 2021.

[Figure omitted. See PDF]

Figure 12

Same as Fig.  but for N₂O.

[Figure omitted. See PDF]

Table 11

Mean values of random, systematic, and total uncertainties in the CH₄ retrieval for the entire year of 2021 using the final suggested retrieval strategy. These values represent the weighted errors in the corresponding partial column, with the total column errors presented in the bottom row.

Altitude (km)	Random (%)	Systematic (%)	Total (%)
1.6–9.8	0.58 $\pm$ 0.06	4.07 $\pm$ 0.41	4.12 $\pm$ 0.41
10.7–19.6	0.46 $\pm$ 0.03	7.22 $\pm$ 0.47	7.24 $\pm$ 0.47
Total column	0.44 $\pm$ 0.06	3.45 $\pm$ 0.05	3.48 $\pm$ 0.05

Table 12

Mean values of random, systematic, and total uncertainties in the N₂O retrieval for the entire year of 2021 using the final suggested retrieval strategy (r8). These values represent the weighted errors in the corresponding partial column, with the total column errors presented in the bottom row.

Altitude (km)	Random (%)	Systematic	Total (%)
		(%)
1.6–9.8	0.29 $\pm$ 0.09	5.21 $\pm$ 0.85	5.22 $\pm$ 0.85
10.7–19.6	0.66 $\pm$ 0.13	10.74 $\pm$ 0.75	10.76 $\pm$ 0.75
Total column	0.16 $\pm$ 0.05	2.58 $\pm$ 0.09	2.59 $\pm$ 0.09

5 Conclusions

Over recent years, there has been noticeable variability in the retrieval strategies for CH₄ among different ground-based FTIR instruments within the IRWG of NDACC. Acknowledging advancements in spectroscopy and the need for retrieval consistency among the IRWG of NDACC, our study evaluates various retrieval strategies, encompassing not only CH₄ but also N₂O. Using observations from Boulder, Colorado, we initially assess fitting results across diverse spectroscopy databases, including the latest HITRAN databases, the ATM spectroscopy utilized by TCCON and IRWG, a combination of them, and a water vapor line list from DLR. We also evaluate previously identified suitable micro-windows. While some cases exhibit clear improvements in residuals for certain spectroscopy, in other instances, residual differences are less distinct, despite significant variations in profile retrievals. Thus, we emphasize the importance of the subsequent step: comparing independent long-term AirCore and aircraft in situ profile observations with the various FTIR retrieval strategies. By utilizing this dataset, we strengthen the evaluation of various retrieval strategies for CH₄ and N₂O, an approach rarely undertaken within the IRWG. In total, we assess 10 CH₄ retrieval strategies, varying micro-windows, spectroscopy, and regularization (Tikhonov or optimal estimation) approaches, as well as 8 N₂O retrievals, primarily focusing on spectroscopy, regularization, and a priori profiles. Notably, both AirCore and aircraft profiles exhibit remarkable similarity in their relative differences compared to all FTIR retrievals, strengthening our confidence in the assessment. Moreover, bias analyses across various atmospheric layers for both CH₄ and N₂O contribute to refining retrieval methodologies. Clear biases were observed when undertaking a comparison with independent profiles, making the exclusion of certain retrievals straightforward. We propose optimized retrieval settings for CH₄ and N₂O, leveraging a first-order Tikhonov method and incorporating a priori profiles from the latest WACCM model simulations to enhance accuracy. While the HITRAN 2020 spectroscopic database is effective for N₂O, it exhibits suboptimal residuals for CH₄, indicating the need for further refinements. In contrast, ATM 2020 yields better results for CH₄, with a slight improvement when combined with the DLR water vapor line list. However, this enhancement may be site-dependent. Furthermore, profile comparisons reveal biases within different atmospheric layers. For the proposed retrievals, we obtain a mean bias of $-\mathrm{0.08}$ $\pm$ 0.38 % and 0.89 $\pm$ 0.28 % for the tropospheric and stratospheric layers of CH₄ using AirCore, respectively, whereas the mean bias is 0.39 $\pm$ 0.42 % for aircraft comparisons in the troposphere. For N₂O, the bias in the troposphere is approximately 0.18 $\pm$ 0.2 %. Furthermore, we provide uncertainty budgets, including random and systematic sources, to comprehensively understand error sources, guiding future refinement efforts. Random errors of about 0.5 %, mainly attributed to temperature profile uncertainties and measurement noise, dominate in the troposphere for both gases. Overall, our findings significantly contribute to advancing the understanding of atmospheric composition and advocate for an effective harmonized approach among all IRWG/NDACC sites. Although Boulder benefits from unique long-term AirCore and aircraft profile observations, these capabilities are not widespread across IRWG/NDACC sites, underscoring the importance of deploying similar observations systems elsewhere in the NDACC network.

Code and data availability

The profile retrieval algorithm, SFIT4 , is freely available and can be obtained at 10.18758/CLYG3EUO. The Level-2 NOAA AirCore dataset (v20230831) is publicly available via the NOAA data repository: 10.15138/6AV0-MY81. Atmospheric observations from the NOAA GML Carbon Cycle Aircraft Vertical Profile Network can be obtained from 10.15138/39HR-9N34. Additional data used in this study can be provided upon request from the corresponding author. The figures in this paper were created using Python. Plotting scripts are available from the corresponding author upon request.

Author contributions

IO and JWH conceptualized the project. IO carried out the data analysis, generated all figures and plots, and prepared the initial draft of the manuscript. BCB, KM, and DS contributed the AirCore, aircraft, and Lauder FTIR datasets, respectively, and provided substantial input throughout the study. All co-authors contributed to discussions, offered feedback on the manuscript, and assisted in its preparation.

Competing interests

The contact author has declared that none of the authors has any competing interests.

Disclaimer

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors.

Acknowledgements

We thank the many people at NOAA GML that contributed to the AirCore and aircraft datasets used in this study, including Andy Crotwell, Molly Crotwell, Ed Dlugokencky, Jack Higgs, Xin Lan, Pat Lang, Thomas Legard, Monica Madronich, Eric Moglia, John Mund, Don Neff, Tim Newberger, Colm Sweeney, Pieter Tans, and Sonja Wolter.

Financial support

This study has been supported under contract by the National Aeronautics and Space Administration (NASA) award no. NNX17AE38G. This work also received support from the NOAA Climate Program Office's Atmospheric Chemistry, Carbon Cycle, and Climate program (grant no. NA23OAR4310285).

Review statement

This paper was edited by Nicholas Deutscher and reviewed by two anonymous referees.