Methanol (CH3OH) is the second most abundant organic molecule in the atmosphere after methane, with concentrations between 1 (Singh et al., 2001) and 20 ppbv (Heikes et al., 2002), despite a lifetime that has been estimated to lie between 4.7 days (Millet et al., 2008) and 12 days (Atkinson et al., 2006). Plant growth is the largest source of methanol with a 65–80 % contribution to its emissions (Galbally and Kirstine, 2002; Jacob et al., 2005). The atmospheric production of CH3OH through peroxy radical reactions represents up to 15–23 % of its sources (Madronich and Calvert, 1990; Tyndall et al., 2001). Other sources of methanol are plant matter decaying (Warneke et al., 1999), biomass burning (Dufour et al., 2006; Paton-Walsh et al., 2008), fossil fuel combustion, vehicular emissions, solvents and industrial activities.

Methanol influences the oxidising capacity of the atmosphere through reaction with the hydroxyl radical (Jiménez et al., 2003), its main sink, leading to the formation of water vapour and either CH3O or CH2OH radicals, which both react with O2 to give HO2 and formaldehyde (H2CO) (Millet et al., 2006). The photo-oxidation of formaldehyde, a key intermediate in the oxidation of numerous volatile organic compounds, leads to the formation of HO2 radicals and carbon monoxide (CO). As a consequence, CH3OH is considered as a source of CO with a yield close to 1 (Duncan et al., 2007). The main sources and sink of methanol are characterised by significant seasonal modulations. This results in a strong signal for CH3OH, with maximum and minimum abundances observed in the Northern Hemisphere at the beginning of July and in December, respectively (Rinsland et al., 2009; Stavrakou et al., 2011; Wells et al., 2012; Cady-Pereira et al., 2012), reflecting the seasonality of biogenic sources.

In the past decade, ground-based (Schade and Goldstein, 2001, 2006; Karl et al., 2003; Carpenter et al., 2004), ship (Warneke et al., 2004) and aircraft (Singh et al., 2006; Fehsenfeld et al., 2006) in situ measurements combined with space-based measurements, including the Infrared Atmospheric Sounding Interferometer (IASI) onboard the MetOp-A satellite (Razavi et al., 2011), the TES (Tropospheric Emission Spectrometer) nadir-viewing Fourier transform spectrometer (FTS), on board the Aura satellite (Beer et al., 2008), and the solar occultations recorded by the Atmospheric Chemistry Experiment-FTS (ACE-FTS, Bernath et al., 2005; Dufour et al., 2006, 2007) have supplied numerous observations of CH3OH, which have provided valuable insights on the distribution and budget of methanol at the global scale. In addition, previous studies have reported the measurement of methanol from ground-based infrared solar absorption observations performed at Kitt Peak (31.9∘ N, 111.6∘ W, 2090 m a.s.l.; Rinsland et al., 2009) and at Saint-Denis (Reunion Island, 21∘ S, 55∘ E, 50 m a.s.l.; Stavrakou et al., 2011; Vigouroux et al., 2012). However, there still remain large uncertainties in our knowledge of the methanol global sources and sinks, as indicated by the large discrepancies existing between different measurement-based estimates of the total sources (Galbally and Kirstine, 2002; Tie et al., 2003; von Kuhlmann et al., 2003a, b; Jacob et al., 2005; Millet et al., 2008; Stavrakou et al., 2011).

In this paper, we report the first long-term methanol time series (17 years) derived from ground-based high-resolution infrared spectra recorded with a Fourier transform infrared (FTIR) spectrometer operated under clear sky conditions at the high-altitude International Scientific Station of the Jungfraujoch (ISSJ, Swiss Alps, 46.5∘ N, 8.0∘ E, 3580 m a.s.l.; Zander et al., 2008) providing a valuable tool for model and satellite validation. Most of the available spectra have been recorded within the framework of the Network for Detection of Atmospheric Composition Change monitoring activities (NDACC; see http://www.ndacc.org) complementing the NDACC measurements at northern mid-latitudes. A detailed analysis was conducted to optimise the retrieval strategy of atmospheric methanol in order to minimise the fitting residuals while maximising the information content. A thorough discussion of the retrieval strategy, data characterisation (information content and error budget), long-term trend and seasonal cycle of total and partial columns of methanol above Jungfraujoch is presented here. This paper is organised as follows. A detailed description of the optimised retrieval strategy is given in Sect. 2. The characterisation of our data by their eigenvectors and error budget is discussed in Sect. 3. Finally, in Sect. 4, we present and discuss the results, focusing on the intra-annual and intra-day variability of methanol at ISSJ along with comparisons with in situ measurements, satellite occultations and model calculations.

Regular FTIR observations have been carried out at the ISSJ with a homemade spectrometer since 1984, complemented in the early 1990s and then definitely replaced by a commercial Bruker IFS120HR instrument (Zander et al., 2008). This spectrometer is equipped with HgCdTe and InSb cooled detectors, allowing us to cover the 650 to 4500 cm-1 region of the electromagnetic spectrum. Since 1991, the FTIR instruments are affiliated with the NDACC network.

The Bruker observational database consists of more than 6500 spectra recorded between 1995 and 2012 with an optical filter covering the 700 to 1400 cm-1 domain encompassing the fundamental C-O stretching mode υ8 of methanol at 1033 cm-1. Spectral resolution, defined as the reciprocal of twice the maximum optical path difference, alternates between 0.004 and 0.006 cm-1. Signal-to-noise (S / N) ratios vary between 250 and 1800 (average spectra resulting from several successive individual Bruker scans, when solar zenith angles vary slowly). The optimisation of the retrieval strategy was based on a subset of 314 spectra covering the year 2010.

The CH3OH column retrievals and profile inversions have been performed using the SFIT-2 v3.91 fitting algorithm. This retrieval code has been specifically developed to derive mixing ratio profiles of atmospheric species from ground-based FTIR spectra (Rinsland et al., 1998). It is based on the semi-empirical implementation of the Optimal Estimation Method (OEM) developed by Rodgers (1990). Vertical profiles are derived from simultaneous fits to one or more spectral intervals in at least one solar spectrum with a multilayer, line-by-line calculation that assumes a Voigt line shape (Drayson, 1976). The model atmosphere adopted above the Jungfraujoch altitude consists of a 39 layers scheme with progressively increasing thicknesses, from 3.58 km to reach the 100 km top altitude. The pressure-temperature profiles are provided by the National Center for Environmental Prediction (NCEP, Washington DC, USA, http://www.ncep.noaa.gov/) while the solar line compilation supplied by F. Hase (KIT) (Hase et al., 2006) has been assumed for the solar absorptions. Line parameters used in the spectral fitting process were taken from the HITRAN 2008 spectroscopic compilation (Rothman et al., 2009). Methanol lines were added to the HITRAN compilation for the first time in 2004 (Rothman et al., 2005). The parameters for the 10 µm region are described in the paper by Xu et al. (2004) and were derived from measurements with two high-spectral resolution FTS instruments.

Two spectral windows both encompassing the υ8 C–O stretch absorption band of methanol have been defined. Synthetic spectra (6.1 mK or 0.0061 cm-1, zenith angle of 80∘) have been computed for the first and second order absorbers in both selected windows and are illustrated in Fig. 1. The first interval ranges from 992 to 1008.3 cm-1 and is based on windows used in previous investigations. A 992–998.7 cm-1 window was employed for the retrieval of CH3OH from Kitt Peak FTS spectra (Rinsland et al., 2009) and a 984.9–998.7 cm-1 window was used for the initial retrievals of methanol from ACE-FTS occultation observations (Dufour et al., 2007). The latest ACE-FTS CH3OH retrievals (version 3.5) use an extended window from 984.9 to 1005.1 cm-1. Measuring in the limb, ACE-FTS measurements start to saturate for wavenumbers above 1005.1 cm-1 for occultations with higher than average O3 levels. As ground-based observations do not have this problem, we included supplemental methanol features up to the 1008.3 cm-1 limit. The second interval, ranging from 1029 to 1037 cm-1 is used by Vigouroux et al. (2012).

Absorption by the main ozone isotopologue (16O-16O-16O or O3) captures nearly 93 and 98 % of the IR radiation in the “1008” and “1037” windows respectively and is close to saturation in the latter one. Methanol features are much weaker, with mean absorption of 1.7 and 1.8 % in the “1008” and “1037” windows respectively. Additional absorptions are associated with O3 isotopologues, such as O3(668) or (16O-16O-18O), O3(686) or (16O-18O-16O), O3(676) or (16O-17O-16O) and O3(667) or (16O-16O-17O) as well as carbon dioxide (CO2) and water vapour (H2O). Since the CH3OH absorption lines are quite weak, only spectra with solar zenith angles greater than 65∘ and up to 80∘ have been analysed. During the retrievals, both windows were for the first time fitted simultaneously.

The a priori mixing ratio profile for the CH3OH target is a zonal mean (for the 41–51∘ N latitude band) of 903 occultations recorded by the ACE-FTS instrument (version 3.5) between 27 March 2004 and 3 August 2012, extending from 5.5 to 30 km tangent altitudes. The profile was extrapolated to 1 ppbv to the surface (Singh et al., 2001; Heikes et al., 2002), and to 0.05 ppbv (Singh et al., 2006; Dufour et al., 2007) for upper layers. The covariance matrix is specified for each layer as a percentage of the a priori profile and an ad hoc correlation length, which is interpreted as a correlation between layers decaying along a Gaussian. For methanol, we adopted a 50 % km-1 diagonal covariance and a Gaussian half width of 4 km for extra diagonal elements. A priori profiles for all interfering molecules are based on the WACCM (version 5, the Whole Atmosphere Community Climate Model, e.g. Chang et al., 2008) model climatology for the 1980–2020 period and the ISSJ station. The vertical profiles of CH3OH, O3 and O3(668) are fitted during the iterative process while the a priori distributions of O3(686), O3(676), O3(667), H2O and CO2 are scaled. Since the fitting quality is significantly different in both windows, two different values for the signal-to-noise ratio for inversion have been selected, i.e. 180 and 40 for the “1008” and “1037” domains, respectively.

When fitted independently, we observe a compact correlation between the corresponding CH3OH total columns retrieved from both windows with a small bias of 15 ± 13 % (2σ). When comparing ozone total columns respectively retrieved from the strategy described in this work and from the retrieval strategy applied within the NDACC network (window limits: 1000–1005 cm-1, Vigouroux et al., 2008), no significant bias emerges from the comparison between the two ozone total column sets, with a mean relative difference of -0.8 ± 2.4 % (2σ), demonstrating a proper fit of the main interference involved in our methanol retrieval strategy. Additional functions are also included in the fitting process to account for deviations from a perfectly aligned FTS. As an effective apodization function, we assumed a polynomial function of order 2 (Barret et al., 2002). The effective apodization parameter (EAP) gives the value of the effective apodization function at the maximum optical path difference and is synonymous of a well-aligned instrument when it is close to 1.0. The inversion of the EAP has been included in our retrieval as well as in the NDACC's retrieval strategy of ozone. The EAP derived from both strategies proved to be consistent, with a mean relative difference of 0.7 ± 2.6 % (2σ). Those three latter points give confidence in the combination of the two selected windows and in our optimised retrieval strategy.

Information content has been carefully evaluated and typical results are displayed on Fig. 2. The information content is significantly improved, with a typical degree of freedom for signal (DOFS) of 1.82, in comparison with DOFS of about 1 in previous studies (e.g. Rinsland et al., 2009; Vigouroux, et al., 2012). In Fig. 2, the first eigenvector and eigenvalue (see left panel, in orange) show that the corresponding information is mainly coming from the retrieval (99 %). The increase of information content allows us to retrieve a tropospheric column (Tropo, from 3.58 to 10.72 km) with only 1 % of a priori dependence as well as two partial columns with less than 30 % of a priori dependence (second eigenvector), i.e. a low-tropospheric (LT, from 3.58 to 7.18 km) and an upper troposphere–lower stratosphere (UTLS, from 7.18 to 14.84 km).

The error budget is calculated following the formalism of Rodgers (2000), and can be divided into three different error sources: the smoothing error expressing the uncertainty due to finite vertical resolution of the remote sounding system, the forward model parameters error, and the measurement noise error. The right panel of Fig. 2 gives the corresponding error budget, with identification of the main error components, together with the assumed variability. Error contributions for total and all three partial columns are reported in Table 1.

Through a perturbation method, we also accounted for other error sources: systematic errors, such as the spectroscopic line parameters and the misalignment of the instrument, while uncertainty on the temperature and on the solar tracking is considered to be source of random error. Table 1 provides an error budget resulting from major instrumental and analytical uncertainties. For the spectroscopic line parameters, we included in our error budget the uncertainty on line intensities provided by the HITRAN database. As methanol line intensities matter, a rough idea of the accuracy of the intensities can be obtained from Table 8 of the Xu et al. (2004) study, as it reports an RMS deviation of 7 %. It should be noted that the uncertainty on ozone and its isotopologues lines, according to HITRAN-08 parameters, amounts to between 5 and 10 % (Rothman et al., 2009). However, an extremely high accuracy of ozone spectroscopic parameters is required in order to retrieve methanol columns properly. We noted that the SFIT-2 algorithm fails to perform a satisfying retrieval when using spectroscopic parameters with ozone lines intensity incremented by 10 %, suggesting that the error on the concerned lines intensity is more likely to be closer to 5 (or even lower) than to 10 %. Therefore, we accounted for an error on ozone and its isotopologues line intensities of 5 % in our error budget.

We accounted for an error of 10 % on the instrument alignment at the maximum path difference. By comparing the two official NDACC algorithms, Hase et al. (2004) and Duchatelet et al. (2010) have established that the forward model may induce a maximum error of 1 % on the retrieved columns for a suite of FTIR target gases. The uncertainty on the pressure–temperature profiles is provided by NCEP with an error of 1.5 K from the ground to an altitude of about 20 km. Concerning the upper levels, the uncertainty increases with altitude, from 2 K around 25 km until 9 K at the top. The uncertainty on the solar zenith angle (SZA) is estimated at 0.2∘.

We also provide in Table 1 the mean relative standard deviation for each daily mean for days with three or more measurements. It is found to be of the same order of magnitude as the random error. The dominant contribution to the systematic error is the error on methanol spectroscopic lines, while the measurement noise error is the main component of random error. Both systematic and random errors are given in Table 1, with 7 % and around 5 % respectively on the total columns.

Since the improvement in information content allows us to compute partial columns with only a 30 % a priori dependence and as the random error on the tropospheric column is about four times the error on total columns (see Table 1), we focus our trend analysis on total, LT and UTLS columns. Therefore, an analysis of the seasonal variation of methanol in the lower troposphere and the UTLS has been performed, including comparisons with in situ measurements (Legreid et al., 2008) and to ACE-FTS occultation observations, respectively. Comparisons with simulations obtained from the IMAGESv2 global chemistry-transport model (Stavrakou et al., 2011) have also been conducted.

A long-term time series of methanol has been determined from the analysis of a 17-year time series of infrared solar absorption spectra recorded with a commercial Fourier transform spectrometer Bruker IFS120HR, operated at the high-altitude International Scientific Station of the Jungfraujoch (ISSJ, Swiss Alps, 45∘ N, 8.0∘ E, 3580 m a.s.l.; Zander et al., 2008) providing a valuable tool for model and satellite validation and complementing the NDACC measurements at northern mid-latitudes.

The results were analysed using the SFIT-2 v3.91 fitting algorithm and thanks to the combination of spectral windows used in previous studies for the retrieval of methanol from FTS spectra (Dufour et al., 2007; Rinsland et al., 2009; Vigouroux, et al., 2012), we have significantly improved the information content. With a typical DOFS of 1.82, a total column and two partial columns time series are available, i.e. a lower-tropospheric (LT, 3.58–7.18 km) and an upper tropospheric–lower stratospheric one (UTLS, 7.18–14.84 km). Both random and systematic error sources have been identified and characterised using the spectra recorded in the year 2010, and are found to be respectively 5 and 7 % for the total column.

The analysis of the time series does not reveal a significant long-term trend but shows a high peak-to-peak amplitude of the seasonal cycle of 129.4 ± 5.5 % (1σ) for total columns. Methanol total and partial columns are characterised by a strong seasonal modulation with minimum values and variability in December to February and maximum columns in June–July. First analysis of methanol diurnal variation shows an increase of methanol in the morning and a decrease during the afternoon for all seasons but summer.

Comparisons with methanol measurements obtained with other techniques (in situ and satellite) give satisfactory results. The FTIR lower tropospheric data compared to in situ measurements generally shows a good agreement regarding the data dispersion. Concerning the UTLS partial columns, there is a close to statistical agreement with ACE-FTS occultations despite higher ACE columns of methanol in March and May.

The IMAGESv2 simulations underestimate the peak-to-peak amplitude for total and lower-tropospheric columns. Despite the absence of a systematic bias between our results and the IMAGESv2 simulations, comparisons show seasonal differences with an overestimation of winter methanol and an underestimation during summertime, which might be explained by an overestimation of the vertical gradient of methanol mixing ratios by the model. Regarding UTLS columns, the peak-to-peak amplitude and timing of the maximum (June–July) in both IMAGESv2 simulations are in very good agreement with the FTIR results.

Even though the role of plant growth in methanol budget is confirmed by its seasonality, large uncertainties remain in the methanol budget. Thanks to the improvement of the information content of our retrieval and therefore our vertical resolution, our partial column time series should contribute to better constraints for model simulations and therefore may lead to a better understanding of methanol budget.