| 07 Oct 2026
Multi-decadal trend of HFC-23 from FTIR observations at the NDACC Junfraujoch station: Persistent increase and comparison with in situ AGAGE observations, TOMCAT model simulations, and ACE-FTS satellite data
Abstract. The gradual phase-out of stratospheric ozone-depleting substances (ODS), including chlorofluorocarbons (CFCs) and hydrochlorofluorocarbons (HCFCs), has led to the widespread adoption of hydrofluorocarbons (HFCs). HFC-23 (CHF3) ranks third among HFCs in radiative forcing contribution. As a potent greenhouse gas, it falls under the Kyoto Protocol and Kigali Amendment, motivating careful monitoring. HFC-23 is not intentionally manufactured; its atmospheric release stems primarily from HCFC-22 production. This study presents the multi-decadal trend of HFC-23 derived from a newly developed retrieval methodology applied to high-resolution solar absorption spectra recorded by ground-based Fourier Transform InfraRed (FTIR) instruments. Measurements obtained at the Jungfraujoch high-altitude research station (Switzerland) within the Network for the Detection of Atmospheric Composition Change (NDACC) were combined across three successive FTIR instruments to produce a continuous and consistent time series from 1986 to 2024, including the earliest detected occurrence of HFC-23 at the site. The FTIR time series was compared with two independent observation datasets, surface in situ measurements from the Advanced Global Atmospheric Gases Experiment (AGAGE) network at Jungfraujoch and satellite retrievals from the Atmospheric Chemistry Experiment Fourier Transform Spectrometer (ACE-FTS, L2 v5.3), as well as simulations from the TOMCAT global three-dimensional chemical transport model. The resulting annual relative trends (2008–2023) are 3.52 ± 0.37 %.yr−1 (FTIR), 3.52 ± 0.15 %.yr−1 (AGAGE), 3.66 ± 0.22 %.yr−1 (ACE-FTS) and 3.53 ± 0.11 %.yr−1 (model). The consistency among datasets supports our retrieval strategy. This methodology can be applied at additional NDACC sites, enabling broader spatial coverage for monitoring this compound using ground-based FTIR instrumentation.
This article describes a long-term HFC-23 data product from the NDACC FTIR at the Jungfraujoch station. The retrieval approach is clearly described, and the data product looks to be of good quality. I have no suggestions for major changes but there are some potential minor issues that might be addressed (described below). Chief among them could be a misunderstanding on my part regarding the conversion of column amounts to mole fractions, which is discussed in more detail below. Having both pseudo column amounts and mole fractions from the TOMCAT model could help diagnose the issue (if there is an issue). There was also possible evidence of channeling in the example residuals shown, something to carefully consider when targeting a weak absorber like HFC-23. There was no indication of a systematic error from channeling manifesting in the slope of the HFC-23 trend (compared to trends from ground-based AGAGE measurements), but the example residuals related to a new instrument (a Bruker IFS-125HR), which wasn’t used to generate the majority of the data product. This could represent a potential issue moving forward if the old instrument (an IFS-120HR) didn’t have a channeling problem but the new on does.
Line-by-line comments:
>Line 76: thanks to thee development
thee -> the
>Line 166: Gohar et al. (2004) (Gohar et al., 2004), Sharpe et al. (2004) (Sharpe et al., 2004), Chung et al. (2005) (Kyoung Chung, 2005), and Harrison (2013) (Harrison, 2013)
The set of second citations following the inline citations seem unnecessary
>Line 192: Given its weak absorption, it is essential that the contributions from the other chemical species are well characterized and adjusted in the retrieval.
In panel B of Figure 1, it is impossible to see what contributions HFC-125 and PAN are providing to the spectrum. It might be useful to include the HFC-125 and PAN contributions in panel C of Figure 1 to avoid giving the impression that their contributions could simply be ignored.
Looking at the residuals in panel A of Figure 1, there is possible evidence of channeling in the spectrum. I expect one would need to look at residuals in a broader range of wavenumbers to know for sure if it is actually channeling or just a vaguely sinusoidal variation to the residuals. Is it possible to include a cosine term (or maybe multiple terms) in the least squares fitting in Sfit to account for channeling? If there are multiple channeling contributions superimposed, you could get a beating pattern in the channeling feature.
>Line 233: Fig. 3 shows
Previous references to figures write out the word “Figure,” while an abbreviation is used here. Use one or the other, but I presume the journal has a suggested format.
>Line 238: …the monthly average RSD
Acronym not defined on first use.
>Line 248: data from the IFS-120HR (1996–2024) and IFS-125HR (2024) yield values of 7.2% and 1.1%, respectively.
Stating the comparison in this manner seems to imply a greater improvement in precision for the 125HR than perhaps can be expected. It is not an entirely fair comparison, in that there appears to be only 1 data point reported from the 125HR instrument (December 2024), and data points from earlier years (circa 1996) for the 120HR instrument had significantly lower HFC-23 levels, so the standard deviation represented a larger percentage of the total even if the magnitude of the standard deviation was similar to observations in later years. The increasing HFC-23 levels would therefore lead to improved precision over time regardless of the instrument used, and assigning a single value to the 120HR 28 year data set becomes less representative for that instrument.
Additionally, there are a number of data points from the 120HR instrument between 2020 and 2024 that have very small error bars, so small that they are hidden within the data point. I don’t know if that means there is a small number of measurements with similar results for that month or the measurements are simply showing exceptionally low scatter for some reason. Error bars for other months during 2020 to 2024 are quite large. Moving forward, I presume the 125HR results could show similar variability in the magnitude of the standard deviation, which means it seems a bit early to draw any conclusions on precision for the 125HR instrument based on what appears to be the error bars from one month’s worth of measurements (unless the reported value was based on more results than were shown in the monthly average plots).
>Line 249: we find that HFC-23 concentrations become detectable in our observations from 1986 onwards
What is meant by detectable? One interpretation is that the absorption depth neared the noise level and the HFC-23 spectral feature became visible in the residuals, while another is that fitted HFC-23 values became persistently positive within the standard deviation (which would happen before the residuals became readily visible by eye).
>Line 321: The mean offset of 10.1% between ACE-FTS and ground-based FTIR data can be partly attributed to differences in measurement geometry between the Jungfraujoch FTIR instruments (zenith-looking) and those aboard SCISAT (limb-sounding).
>Line 358: …including the transport time required for mixing and vertical transport between surface sources and the upper troposphere.
Agreed for both. I admit I don’t completely follow the nuances of converting column amounts to mole fractions, but to me the various data sets appear to be reasonably consistent even though they don’t quantitatively agree.
AGAGE is measuring at the surface, in the source region, and not surprisingly has the highest mole fraction values.
TOMCAT values use the ground-based observations as inputs but indicate lower values at the Jungfraujoch station’s altitude than at the surface. It is not clear to me if that results from physical processes in the model (e.g., a time lag for transport/mixing from surface emissions and possibly atmospheric processing from reactions with OH), or it is perhaps the signature of transforming column amounts to mole fractions. Or maybe a combination of the two effects. Again, I don’t follow the subtleties of the mole fraction conversion, but it seems to me a column measurement (or a pseudo column measurement, in the case of TOMCAT) will include a contribution from the stratosphere, where the HFC-23 mole fraction is potentially smaller, which could lead to smaller values when converting column measurements to mole fraction. In effect, the weighted average mole fraction (with the weighting provided by the decreasing density with altitude) would yield a smaller value than the mole fraction at the lower end of the altitude range. My apologies if I completely misunderstood the process. Are the mole fractions at the Jungfraujoch station’s altitude not directly available from the TOMCAT model, removing the need to convert TOMCAT column amounts to mole fraction? If mole fractions are directly available from TOMCAT, how do they compare to the values derived from converting columns? If the column conversion is not contributing to the difference from AGAGE results, is it mostly just transport/mixing time lag?
ACE-FTS atmospheric mole fractions always exhibit a time lag to surface measurements (Schmidt et al., Trends in atmospheric composition between 2004-2023 using version 5 ACE-FTS data, JQSRT, 2024). For molecules that are emitted in the Northern Hemisphere, air in the tropical upper troposphere shows a similar time lag for transport/mixing to that experienced by ground stations in the Southern Hemisphere, about 2.25 years. This was observed for several different molecules including HFC-23. I don’t know what time lag to expect for ACE-FTS measurements between 6.5 and 12.5 km in the latitude range 40-50 °N, but I would actually be more concerned if there were no time lag. Judging from Figure 4, the usual time shift of 2.25 years for the AGAGE data should yield reasonably good overlap with the ACE-FTS observations.
For the NDACC FTIR results, my concerns for converting column amounts to mole fraction would also apply and might be more pronounced than for TOMCAT if the model underestimates HFC-23 decrease with altitude in the stratosphere. That effect could (at least partly) explain why the NDACC FTIR results are lower than ACE-FTS results, an indirect consequence of the differences in observation geometry. If I am incorrect on the mole fraction conversion being a potential issue, another option to explain the difference with ACE could be channeling in the NDACC FTIR. As I said previously, there was possible evidence of channeling in the residuals shown in Figure 1, although that example was for the 125HR instrument, while most (all?) of the NDACC FTIR data in Figure 4 are from the 120HR instrument. Is there evidence of channeling in residuals from the 120HR instrument? With a weak absorber like HFC-23, channeling features could introduce significant systematic errors. However, if the channeling remained stable over time, I expect the percentage error would change as the HFC-23 signal grew, and I don’t see evidence of that in the HFC-23 trend from the NDACC FTIR, i.e., the percentage change in HFC-23 mole fraction matches well with the AGAGE ground-based results. That would argue that channeling is not a major concern for the 120HR instrument.
>Line 416: the maximum absorption depth for HFC-23 is approximately 2.5%
This is circa 2025. The peak absorption depth changes significantly over time.