Detection of O 4 absorption around 328 nm and 419 nm in measured atmospheric absorption spectra

Retrieving the column of an absorbing trace gas from spectral data requires that all absorbers in the corresponding wavelength range are sufficiently well known. This is especially important for the retrieval of weak absorbers, whose absorptions are often in the 10−4 range. Previous publications on the absorptions of the oxygen dimer O2-O2 (or short: O4) list absorption peaks at 328 nm and 5 419 nm, for which no spectrally resolved literature cross-sections are available. As these absorptions potentially influence the spectral retrieval of various trace gases, such as HCHO, BrO, OClO and IO, their shape and magnitude needs to be quantified. We assume that the shape of the absorption peaks at 328 nm and 419 nm can be approximated by their respective neighboring absorption peaks. Using this approach we obtain estimates for the wavelength of the absorption and its magnitude. Using Longpath Differential Optical Absorption Spectroscopy (LP-DOAS) observations and Multi-Axis (MAX)-DOAS observations, 10 we estimate the peak absorption cross-sections of O4 to be (1.7±0.2)×10−47 cm molec−2 and determine the wavelength of its maximum at 328.51± 0.15 nm. For the absorption at 419.0± 0.4 nm a peak O4 cross-section value is determined as (3.7± 2.7)× 10−48 cm molec−2.


Introduction
The collision-induced absorption of the O 2 -O 2 dimer (or short: O 4 ) needs to be considered in various wavelength regions for in situ and remote sensing absorption spectroscopy of various atmospheric trace gases.Furthermore, the O 4 absorption governs the budget of tropospheric singlet oxygen O 2 ( 1 ) (Schurath, 1986), which can potentially impact the oxidation of atmospheric trace gases, like CO (Platt and Perner, 1980) and SO 2 (Sorokin, 2010).In addition, the absorption of O 4 itself can be used to deduce information about the actually observed light paths for passive remote sensing applications by applying inverse modelling methods and radiative transfer models.Ellis and Kneser (1933) and Dianov-Klokov (1959, 1964) reported absorption peaks observed for liquid oxygen at 328 and 419 nm (for an overview, see Table 2).Most of the absorption peaks in the UV-vis range of liquid oxygen can be also found for gas-phase oxygen, as measured by Janssen (1885), Salow and Steiner (1936), Greenblatt et al. (1990), Hermans et al. (1999), Thalman and Volkamer (2013) and others.It was first observed in the atmosphere by Perner and Platt (1980).These absorptions are potentially shifted by less than a nanometre compared to the liquid phase.A clear difference of the relative magnitude for the absorption of O 4 in the liquid and gas phases cannot be seen from the available data.
Spectrally resolved cross-section data suitable for spectroscopy applications exist for the absorption peaks at 344,361,380,447,477,533 and 577 nm and continue further into the red spectral range.Due to instrumental limitations (detection limits and/or covered spectral range), spectrally resolved cross-section data for the absorption peaks at 328 nm (transition 3 − g + 3 − g → 1 + g + 1 + g (ν = 3); see also Table 2) and 419 nm (transition 3 − g + 3 − g → 1 + g + 1 g (ν = 2)) are, to our knowledge, not reported in literature.Salow and Steiner (1936) measured the intensity of the absorption at 328 nm to be 4.2 % of the O 4 absorption at 360 nm, i.e. 15 % of the O 4 absorption at 344 nm.For an O 4 differential slant column density (dSCD) of 4 × 10 43 molec 2 cm −5 , as e.g.found in MAX-DOAS observations under low elevation angles, this corresponds to an optical depth of 1.7 × 10 −2 at 360 nm (Thalman and Volkamer, 2013) and thus an optical depth of 7 × 10 −4 at 328 nm.This could introduce systematic biases in the spectral retrievals of BrO, HCHO, OClO and SO 2 , when this wavelength range is included in the respective spectral analysis, as e.g. in satellite retrievals of HCHO (De Smedt et al., 2008), various observations of BrO in the boundary layer (see Vogel et al., 2013, and references therein) and ground-based measurements of SO 2 (e.g.Schreier et al., 2015 andWang et al., 2017).Convoluted to a spectral resolution of 0.5 nm, the overlaying absorption structure of HCHO at 328 nm has a peak absorption cross section of 4.5 × 10 −20 cm 2 molec −1 (Chance and Orphal, 2011).Typical ground-based HCHO dSCDs observed on the tropical open ocean are about 1-3 × 10 16 molec cm −2 (e.g.Peters et al., 2012) and lower at higher latitudes (e.g.De Smedt et al., 2008), which corresponds to an optical depth of 4 − 14 × 10 −4 at 329 nm.Therefore it is important to consider the possible O 4 absorption at around 328 nm if the retrieval wavelength range for HCHO is extended towards shorter wavelengths below 330 nm.The recommended HCHO setting from Pinardi et al. (2013) suggests a fit interval from 336.5 to 359 nm and thus does not include the O 4 absorption at around 328 nm.This recommendation is for an analysis with fixed Fraunhofer reference spectrum.If sequential references are used (as it is typically done today for the spectral retrieval of weak absorbers, as e.g. in Wang et al., 2017), then the spectral range can and should be extended towards shorter wavelengths (e.g.323 nm).
In addition to that, for the spectral retrieval of BrO, Vogel (2012) found a lower limit for the choice of reliable fit intervals at about 330.6 nm, which could have been caused by the neglected O 4 absorption at around 328 nm -the FWHM (full width half maximum) of the known absorption peaks at 344 and 360 nm amounts to 4 nm.The method from Vogel et al. (2013) was applied here on measured atmospheric spectra along different light path lengths.This fit interval was then widely used, e.g. in Lübcke et al. (2014), General et al. (2015), Gliß et al. (2015), Bobrowski et al. (2015) and Lübcke et al. (2016).Widening the spectral retrieval interval for both species could reduce the fit error of BrO and HCHO dSCDs, as relatively large absorption bands are found for both species below 330 nm.
For the absorption at 419 nm no estimate for its intensity is known.However, this spectral region is also known for uncertainties of the available water vapour cross section as reported in Lampel et al. (2015b), which overlays potential O 4 absorptions.Using the same ratio of peak absorption cross sections of O 4 at 477 and 446 nm, an absorption band at 419 nm could potentially have a peak optical depth (OD) of 2 × 10 −4 for an O 4 dSCD of 4 × 10 43 molec 2 cm −5 .This spectral range is of interest, as it is typically included in the spectral analysis of IO (e.g.Großmann et al., 2013;Prados-Roman et al., 2015).The potential to obtain residual spectra with an root mean square (RMS) below 1 × 10 −4 has been demonstrated (e.g.Lampel et al., 2015a;Zielcke, 2015) and allows us to measure IO at sub-ppt concentrations, corresponding to an OD along 10 km light path of < 7 × 10 −4 .Such concentrations have been reported on the open ocean by Prados-Roman et al. (2015) and were found above their respective limit of detection by Zielcke (2015) in polar regions (0.2-0.4 ppt using LP-DOAS and MAX-DOAS observations).
In this work, spectral data from three different instruments and environmental settings, utilizing two different remote sensing geometries, are presented in order to estimate the magnitude of the O 4 absorption peaks at around 328 and 419 nm.The different techniques used and the different locations of the measurements make it very unlikely that instrumental artefacts or local atmospheric influences are responsible for the observed absorption structures.

Measurement campaigns
In this section we shortly describe the measurement campaigns, during which DOAS data were collected that are used in this paper.

Antarctica LP-DOAS 2012
In 2012, a measurement campaign was undertaken based at New Zealand's research station Scott Base in Antarctica.The station is situated on Ross Island at 78 • S and the campaign lasted throughout austral spring, from the end of August until the end of November.
The LP-DOAS instrument used in this campaign has been used in previous studies and therefore has already been extensively described elsewhere, e.g. in Pöhler et al. (2010) and Eger et al. (2018).
In brief, light is coupled in and out of a telescope using a Y-shaped fiber optic bundle consisting of seven individual fibers.The common end is mounted near the focal point of the telescope, the single central fiber is connected to the spectrometer and the six outer fibers are attached to a 800 µm mono fiber which leads to the light source.A diffuser plate can be driven into the light path in front of the telescope end of the fiber in order to record spectra of the lamp (so-called optical shortcut spectra).For most of the campaign, a 75 W xenon arc lamp (Osram XBO, not ozone-free) was used as the light source.During the last 2 weeks, a 500 W arc lamp (PLI Hanovia HSAX5002) was used, which has a higher luminous density and thus leads to a lower limit of detection in the deeper UV region around 300 nm.Both lamps were coupled into the fiber using a single fused silica lens.Besides several colour filters to reduce stray light, the lamp housing also featured a shutter in front of the fiber coupling to block the lamp in order to be able to record background spectra with the telescope, i.e. spectra of the light entering the telescope due to scattering on the surface or air and not from the dedicated light source.The spectrometer, an Acton 300i, was used with a resolution of 0.50 nm and spectra were recorded by a back-illuminated CCD camera from Roper Scientific (Spec-10:2KBUV).
Two light paths were set up approximately 1.5 m over sea ice, for which two retro reflector arrays were deployed.The retro reflector arrays reflecting the light back into the telescope consisted of 12 (short path) and 50 (long path) individual elements made of fused silica with a diameter of 63.5 mm each.One array was located closer to the station, to be able to measure during periods with fog or snow drift, the other one further away to achieve lower detection limits.They were located at distances of 1.46 km (short) and 4.01 km (long).
The measurement procedure was as follows: spectra were acquired alternatingly on both light paths.For each light path, spectra were recorded at four different wavelength intervals.The region of interest for this study is between 271 and 355 nm.For each of those regions, 25 individual spectra with a saturation of 70 % were averaged to one measurement spectrum, with a maximum duration of 1 s per individual spectrum.Typical exposure times for one individual spectrum were between 6 and 500 ms depending on the wavelength range, visibility conditions and the used arc lamp.Thirty-two measurement spectra are averaged afterwards to achieve an optimal signal-to-noise ratio.

Polarstern 2014-2016
The MAX-DOAS instrument used during these cruises PS88-PS98 (ANT XXX, ARK XXIX, ANT XXXI1 ) on board RV Polarstern from October 2014 to April 2016 is described in Lampel et al. (2017a) and the same upper limits for the spectral stability of the instrument also apply here for each leg.As during previous cruises, the exposure time per spectrum was set to 2 min.Spectra were recorded at seven elevation angles of 90 • (zenith), 40, 20, 10, 5, 3 and 1 • as long as solar zenith angles were below 85 • .To reduce RMS, four elevation sequences were co-added before the DOAS analysis.

Penlee Point Atmospheric Observatory
2015-2016 MAX-DOAS measurements were performed from 3 April 2015 to 3 March 2016 at the Penlee Point Atmospheric Observatory (PPAO) on the south-west coast of the UK (e.g.Yang et al., 2016).Similar to the instrument used in Lampel et al. (2015a) during the MAD-CAT campaign, the EnviMeS2 MAX-DOAS instrument is based on an Avantes ultra-low straylight AvaSpec-ULS2048x64 spectrometer (f = 75 mm) using a back-thinned Hamamatsu S11071-1106 detector.The spectrometer is temperature stabilized ( T < 0.02 • C).The UV spectrometer covered a spectral range of 296-459 nm at a FWHM spectral resolution of ≈ 0.55 nm (at 334 nm) or ≈ 6 pixel.The spectral stability was sufficiently high with a maximum spectral shift within 1 day of less than 3 nm after correction for the tilt effect (Lampel et al., 2017b).During the night, mercury (Hg) discharge lamp spectra were recorded automatically in order to measure the instrument's spectral response function.No significant change of the response function was observed during the campaign.
The elevation sequence included elevation angles of −2, −1, 1, 2, 3, 5, 10, 20, 40 and 90 • , heading towards an azimuthal south-westerly direction of 245 • .After 22 January 2016, the azimuthal viewing direction was changed to a south-easterly direction of 147 • .Spectra with a total exposure time of 1 min were recorded at an adaptive integration time per scan in order to obtain spectra with a maximum saturation of 50 %.From 5 June to 27 August 2015 the total exposure time per spectrum was reduced to 10 s.
The inherent non-linearity of the measured intensity values with respect to the actual incoming intensity of the spectrometers was corrected for by multiplying all intensities with a non-linearity correction polynomial.This polynomial was determined from a set of spectra recorded at different exposure times, which were recorded previous to the campaign using a temperature stabilized "white" LED light source.
In order to have a coherent data set and to reduce the RMS noise of the fits, spectra of subsequent elevation angle sequences during 1 day were co-added during preprocessing in order to obtain a consistent MAX-DOAS data set with spectra with a total exposure time of 4 min.

Method
Two DOAS methods were applied to quantify the O 4 absorption around 328 nm.The MAX-DOAS measurements have often longer effective light paths, which are not initially known.LP-DOAS measurements have the advantage that the absolute light path length is known, but they often yield larger fit residuals.

MAX-DOAS
The MAX-DOAS elevation sequences were evaluated against a current Fraunhofer reference using the sum of the two nearest zenith sky spectra in order to minimize the effect of stratospheric absorbers.At the same time, this approach minimizes the effect of instrumental instabilities on the data evaluation.
The literature absorption cross sections listed in Table 1 were convoluted with the measured Hg emission line at 334 nm.The Ring spectrum was calculated using DOASIS (Kraus, 2006), the correction spectrum for vibrational Raman scattering of molecular nitrogen (VRS; Lampel et al., 2015a) was calculated from the Fraunhofer reference itself, shifting the spectrum by the corresponding vibrational energy quantum.
Neither HONO and SO 2 nor OClO were not detected in significant amounts.Water vapour absorption in the UV as reported in Lampel et al. (2017a) is small below 358 nm (σ < 3×10 −28 cm 2 molec −1 ) and thus negligible, especially in polar regions.

LP-DOAS
The optical depth was calculated by dividing atmospheric and optical shortcut spectra after their respective background spectra were subtracted.The optical densities were then high-pass filtered and afterwards the fit was applied.The applied fit scenario settings are shown in Table 1.As for the MAX-DOAS evaluation, the literature cross sections were convoluted with a measured Hg emission line at 334 nm.

Hypothetical O 4 absorption cross sections
The absorption peak shape of the O 4 absorption at 328 and 419 nm cannot be deduced directly from field measurements, as a large number of other absorbers (HCHO, BrO, SO 2 , HONO, OClO, NO 2 , O 3 , H 2 O and others) potentially overlay the respective O 4 absorption peak.Their abundance is unknown in field observations.Additionally, their absorption cross sections may not be known precisely enough (e.g.water vapour; Lampel et al., 2015bLampel et al., , 2017a) ) to determine their abundances in other spectral ranges and thus constrain their overall absorption in order to obtain an extended O 4 absorption cross section from MAX-DOAS or LP-DOAS field measurements.Dedicated laboratory studies will be needed.
The shape of the O 4 absorption peaks at 344, 360, 380 and at 446 and 477 nm is similar as shown in Fig. 1a when plotting the absorption cross section (here from Thalman and Volkamer, 2013) over the difference in wave numbers to the peak absorption.We therefore guessed the shape of the potential absorption bands at 328 and 419 nm by shifting the O 4 absorption peaks at 344 and 446 nm by 1414 and 1476 cm −1 , respectively.These shifts were chosen arbitrarily based on previous publications (Salow and Steiner, 1936;Ellis and Kneser, 1933), which list the wavelengths of the respective absorption peaks.This approach is reasonable, as the width of the O 4 absorption peaks is defined by the lifetime of the collision complex (Long and Ewing, 1973;Thalman and Volkamer, 2013) and is thus related to the energy of the respective absorption peak.The absorption peak shape at 477 nm was parameterized by Sneep and Ubachs ( 2003), but we are not aware of parameterizations or quantum-mechanical calculations yielding absorption peak shapes at other wavelengths at room temperature.

Peak positions
As we assume the maximum of the O 4 absorption peak at 328.2 nm (Salow and Steiner, 1936) and at 419 nm (Ellis and Kneser, 1933), its exact wavelengths needed to be determined again from the measurements.This was done based on experimental MAX-DOAS data, here from PPAO: if introduced as a free parameter in the a DOAS fit, the individual shift of the hypothetical O 4 absorption cross section can be used in order to derive a more accurate estimate for the spectral position of the absorption peak.Measurements, which showed a ratio of fitted dSCD and dSCD fit error of more than 8 were considered for this analysis.The resulting peak position was found at 328.59 ± 0.15 nm and thus results in an overall shift of the absorption at 344 nm of +1359 ± 13 cm −1 .In the blue wavelength range it was found at 419.13 ± 0.40 nm and thus results in an overall shift of the absorption at 446 nm of +1469 ± 23 cm −1 .For the absorption around 419 nm a larger scatter of shift values was observed, probably due to uncertainties of the overlaying water vapour absorption.(Ellis and Kneser, 1933), SaSt (Salow andSteiner, 1936), DiKl (Dianov-Klokov, 1959, intensities manually extracted from Fig. 4) and ThVo (Thalman and Volkamer, 2013).The values in the column for relative intensities are normalized to the value of the absorption peak at 344 or 446 nm, respectively.Transitions according to Greenblatt et al. (1990).4.2 Absorption band at 328 nm

LP-DOAS
The LP-DOAS data evaluation reveals the suspected absorption structure at 328 nm nicely, given the extensive averaging of the data as described in Sect.3.2.An exemplary DOAS fit is shown in Fig. 2. Compared to the ozone and O 4 absorption, the other species feature only a negligible optical depth in this example.Besides ozone and the O 4 absorption band at 343.8 nm, the 328 nm absorption of O 4 is the most prominent absorption feature in this case and spectral region.
In order to compare the relative absorption strength of the 328 and 344 nm bands, the entire data set was filtered for evaluations with a residual RMS better than 1.2 × 10 −4 since the structure of interest can only then be retrieved with sufficient accuracy.Typical optical densities of the 328 nm absorption are between 4 and 6 × 10 −4 along the 8 km light path and column densities reach up to a maximum of around 20 times the instrumental detection limit.
Figure 3 shows the correlation of the LP-DOAS column densities of the suspected 328 nm absorption feature with the well-known absorption feature at 344 nm.Data for the short light path are shown in orange, while data for the long path are shown in blue.A correlation between the suspected and the known absorption is visible.As expected, higher column densities for the longer light path and lower ones for the shorter light path.A linear fit performed on the entire data set indicates a relative absorption strength of 0.183 ± 0.015 Table 3. Results for the O 4 absorption at 328 nm for different fit ranges and settings.MAX-DOAS data were selected according to RMS < 4 × 10 −4 and LP-DOAS data below an RMS of 1.2 × 10 −4 , which then results in a different value for n, the number of valid observations.The correlation coefficient R 2 was calculated in each case.The peak magnitude of the absorption cross section at 328 nm was calculated using the O 4 cross section published by Thalman and Volkamer (2013) using the maximum cross-section value of 9.5 × 10 −47 cm 5 molec −2 at 343.8 nm.The wavelength dependence of the O 4 AMF for the MAX-DOAS observations was not corrected here.Figure 2. Exemplary LP-DOAS fit result of a measurement from 22 November 2012 on the 8 km light path.The O 4 absorption at 328 nm is clearly visible.The retrieved column densities are O 4 (2.67 ± 0.04)×10 43 molec 2 cm −5 and O 4 328 nm (6.09 ± 0.38)×10 42 molec 2 cm −5 .Table 3. Results for the O 4 absorption at 328 nm for different fit ranges and settings.MAX-DOAS data were selected according to RMS < 4 × 10 −4 and LP-DOAS data below an RMS of 1.2 × 10 −4 , which then results in a different value for n, the number of valid observations.The correlation coefficient R 2 was calculated in each case.The peak magnitude of the absorption cross section at 328 nm was calculated using the O 4 cross section published by Thalman and Volkamer (2013) using the maximum cross-section value of 9.5 × 10 −47 cm 5 molec −2 at 343.8 nm.The wavelength dependence of the O 4 AMF for the MAX-DOAS observations was not corrected here.the influence of the wavelength dependence of the air mass factors (AMFs), which is discussed shortly in the following paragraph.

Uncertainties
For PPAO, the distribution of fit errors of the O 4 absorption 5 at 328 nm has its maximum at 5 × 10 41 molec 2 cm −5 (similar as in Fig. 4, corresponding to a peak OD of 4.6 × 10 −5 using the shifted absorption peak from 344 nm), while the www.atmos-chem-phys.net/18/1/2018/Atmos.Chem.Phys., 18, 1-13, 2018 (see also Table 3, cross-section values are listed) at 328 nm compared to the known 344 nm band when allowing for a constant offset.Fixing this offset to zero results in a slightly larger relative absorption strength of 0.212 ± 0.006.The given error is the error of the linear regression analysis and does not reflect the actual measurement error of each individual observation.A variation of the total O 4 SCD along both light paths can be observed, which is found simultaneously for both O 4 absorption bands at 328 and 344 nm, leading to the distribution of the measurements shown in Fig. 3.This points towards a general property of the absorber O 4 and can be explained by temperature variations (up to 35 K within the measurement period) and pressure variations.We estimate that almost twothirds of the observed variation are due to the changing absorption cross section with temperature (Thalman and Volkamer, 2013), one-third due to the change in number density according to the ideal gas law.Pressure variations between 962 and 1006 hPa during the observations also contribute.Variations depending on temperature and pressure were also reported by Wang et al. (2017) and Wagner (2017) for groundbased MAX-DOAS observations and simulated for satellite observations by Park et al. (2017) and Dörner et al. (2017).A further and more detailed analysis of these dependencies is not within the scope of this publication.

MAX-DOAS
For the PPAO data set, several sensitivity studies were performed in order to estimate the effect of ozone absorption and the contribution of VRS on the results.
To account for non-linear effect of strong ozone absorption, we also included a wavelength-scaled version of the ozone absorption as well as its square term, as suggested in Puk ¸īte et al. ( 2010) and Puk ¸īte and Wagner (2016).This was mainly necessary when extending the fit interval below 320 nm to include additional absorption bands of HCHO.One part of the absorption of ozone is due to changes in the slant stratospheric ozone column during the measurement of the elevation sequence and thus assumed to be symmetric and not introduce a systematic bias on the evaluation of the O 4 absorption band.The tropospheric contribution is of similar magnitude.
The additional intensity caused by VRS did not significantly change the result as seen from Table 3.
BrO was analysed in the spectral range from 332 to 358 nm and not found above the detection limit (1 × 10 13 molec cm −2 ) at PPAO.During the Polarstern cruises the amount of HCHO (evaluated in the same spectral range) was typically smaller than at PPAO; however, also significant amounts of BrO were observed during Arctic spring (May-June; up to 1.4 × 10 14 molec cm −2 at 3 • elevation) and Antarctic spring (December-January; up to 5 × 10 13 molec cm −2 at 3 • elevation).We therefore exclude the possibility that the absorption at 328 nm is caused by erroneous BrO or HCHO absorption cross sections.
As the MAX-DOAS measurements took place in regions without strong anthropogenic pollution, tropospheric NO 2 absorption does not significantly contribute to the overall magnitude of the observed residual spectra.
The resulting estimates for the absorption cross section of O 4 at 328 nm can be found in Table 3. Good correlations of the absorptions at 328 and 344 nm can be found for LP-DOAS observations as well as for both MAX-DOAS data sets.For MAX-DOAS different settings for the spectral retrieval were tested and yielded similar results, but for the larger fit interval slightly larger (+10 %) absorption cross-section peak values are estimated.The low correlation for the setting not using the Taylor expansion approach by www.atmos-chem-phys.net/18/1671/2018/Atmos.Chem.Phys., 18, 1671-1683, 2018  • elevation reference spectrum.Spectra were coadded to obtain a total exposure time per spectrum of 4 min.The 328 nm O 4 absorption structure found here is 14 times as large as the DOAS fit error.Without considering this absorption, the HCHO dSCD in this case is larger (+2.5 × 10 14 molec cm −2 or 1.4 %), and the RMS of the residual increases by +22 %. "RS-temp" denotes the difference between Ring pseudo-absorption spectra calculated at 273 and 243 K as listed in Table 1.
Puk ¸īte et al. ( 2010) on the large fit interval was expected as here strong ozone absorption can produce significant residual structures, which can interfere with other fitted species.
Note that the values in this table were not yet corrected for the influence of the wavelength dependence of the air mass factors (AMFs), which is discussed shortly in the following paragraph.

Uncertainties
For PPAO, the distribution of fit errors of the O 4 absorption at 328 nm has its maximum at 5 × 10 41 molec 2 cm −5 (similar as in Fig. 4, corresponding to a peak OD of 4.6 × 10 −5 using the shifted absorption peak from 344 nm), while the distribution of O 4 dSCDs at 328 nm has its maximum at 6 × 10 42 molec cm −2 .Thus most observations are above the detection limit.The influence of strong ozone absorption is largely compensated for by the Taylor expansion approach and even al-lows fits at sufficiently low RMS down to 311.5 nm.The deduced magnitude of the O 4 absorption at 328 nm is slightly larger for larger fit intervals, compare Table 3.
The effect of VRS is negligible especially at the lower end of the fit interval, as can be also seen from Fig. 4. It is, however, correlated with the Ring signal as previously reported for the blue spectral range and increases slightly the number of valid observations, while the deduced magnitude of the O 4 absorption stays constant within the fit error.
The difference in MAX-DOAS AMFs of O 4 for low elevation angles at of 328 nm compared to 344 nm is expected to lead to an underestimation of the absolute O 4 absorption cross section at 328 nm with the approach presented above.Using a set of 10 representative aerosol profiles with aerosol optical thicknesses ranging from 0 to 5 and simulating the resulting O 4 dSCDs at 328 and 344 nm using SCIA-TRAN (Rozanov et al., 2014), an underestimation of 14 % is yielded.This is slightly less than one could have expected for the pure Rayleigh case, yielding ≈ 1 − (328/343) 4 = 16 %.Applying this correction of 14 % to the observed data from Table 3, this means that the LP-DOAS result for the regression without allowing for an offset fits best the corrected MAX-DOAS result of 1.72 × 1.14 = 1.96.This makes sense as for the reference measurements (at a light path length of 0 m), the column density is by definition zero.Estimating the potential systematic measurement errors by the measurement error itself, LP-and MAX-DOAS results agree with each other with and even without the correction for the wavelength dependence of the AMF, as the measurement errors and therefore the scatter of data points are of the same magnitude as the expected difference between the two DOAS measurement types.
As seen from Table 2 the relative intensity of the absorption peak at 328 nm relative to 344 nm of 0.18 is found in between the values listed in Salow andSteiner (1936) of 0.15 andDianov-Klokov (1959) with a value of 0.22.

Absorption band at 419 nm
If assuming a large O 4 dSCD of 1 × 10 44 molec 2 cm −5 and using the ratio of the magnitudes of the other 3 − g + 3 − g → 1 + g + 1 g absorption bands at 446 nm and 477 nm of about 12.7 (Thalman and Volkamer, 2013), the peak magnitude of the O 4 absorption at 419 nm could be expected to reach an optical density of 4 × 10 −4 .It is, however, unclear whether this extrapolation is valid.For the 3 − g + 3 − g → 1 + g + 1 + g absorption bands at 361, 344 and 329 nm, this is approximately the case, as seen in Sect.4.2.
The absorption structure reported by Ellis and Kneser (1933) around 419 nm is overlayed by water vapour absorption at around 416 nm with a peak absorption of 3 × 10 −3 for a water vapour dSCD of 4 × 10 23 molec cm −2 at a spectral resolution of 0.5 nm using the HITRAN2012 line list (Rothman et al., 2013).From MAX-DOAS and LP-DOAS observations it was reported that these absorption lines are overestimated by a factor of 2 in HITRAN2012 (Lampel et al., 2015b), while the overall shape is relatively well reproduced.Thus the water vapour absorption contains some uncertainty, which could have an effect on the detection of the O 4 absorption band at 419 nm.This is the main reason why the wavelength of the maximum O 4 absorption around 419 nm is difficult to estimate.
Therefore the absorption band at 419 nm was fitted in different wavelength intervals, including and excluding the larger water vapour absorption peaks around 442 nm.The variation of the absolute O 4 absorption at 419 nm in these two intervals was less than 25 %.
To circumvent the potential influence of water vapour absorption in this spectral region, an overdetermined system of linear equations was set up in order to quantify the contribution of water vapour as well as O 4 absorption on the apparent O 4 absorption at 419 nm.The "true" dSCDs, S, were determined at a wavelength close to the wavelength of interest, in this case for the water vapour absorption bands and the O 4 absorption band at around 477 nm.These are both more than 10 times stronger than their respective absorptions between 410 and 420 nm.This can be done as enough variation in water vapour concentrations is found for all MAX-DOAS campaigns included here, thus not leading to linearly dependent data points of water vapour and O 4 dSCDs. (1) The resulting c O 4 is 0.0785 ± 0.0070.The correction for wavelength dependence of the AMF is estimated for the pure Rayleigh case, yielding ≈ 1 − (419.0/446.4) 4 = 22 %.In other words, the absorption peak at 419 nm is 10.4 ± 0.9 times smaller than the absorption peak around 445 nm.
However, as the exact shift of the O 4 absorption structure is only poorly restricted from the MAX-DOAS measurements, the remaining uncertainty strongly depends on the position of the absorption peak.Using different fit settings and different positions of the absorption peak (±0.3 nm) yields results for c O 4 within 0.08 ± 0.05.When ignoring the possible interference with water vapour absorption, 0.12 is obtained.This then results in a peak value of the O 4 absorption cross section (after correction of radiative transfer effects) of (5.0 ± 3.5) × 10 −48 cm 5 molec −2 at a wavelength of 419.13 ± 0.42.Despite the large error, this is significantly smaller than the relative intensity listed by Dianov-Klokov (1959) but agrees well with the relative scaling of the larger absorption peaks at 477 and 447 nm as listed in Table 2.

LP-DOAS observations
The O 4 absorption around 419 nm is close to its detection limit.The maximum of the distribution of O 4 fit errors is 7 × 10 41 molec 2 cm −5 , and the maximum of the distribution of O 4 dSCDs is 6 × 10 42 molec 2 cm −5 .
Using the available data sets, a reliable conclusion on the magnitude of the O 4 absorption is difficult to draw.With an estimated value of 0.07 ± 0.05 the ratio of magnitudes to the next larger O 4 absorption peak seems to be similar as the ratio of the magnitudes of the other 3 − g + 3 − g → + g + 1 g absorption bands at 446 and 477 nm of about 0.079 (Thalman and Volkamer, 2013).

Conclusions
We analysed atmospheric measurements of LP-and MAX-DOAS setups from different field campaigns in order to estimate the magnitude and wavelength of previously reported O 4 absorption peaks at 328 and 419 nm, for which no spectrally resolved literature cross sections are currently available and which have not been reported from atmospheric observations so far.
The main conclusion is that both O 4 absorption peaks at 328 and 419 nm can be observed using current MAX-DOAS and LP-DOAS setups and therefore have the potential to introduce biases in the spectral retrieval of weak absorbers.Therefore, further laboratory studies are needed to quantify the magnitude of these small absorption peaks and to obtain a continuous absorption cross section in between these absorption peaks.We suggest addressing at the same time uncertainties in recent water vapour absorption cross sections (see e.g.Lampel et al., 2015b;Volkamer et al., 2015;Wilson et al., 2016 andLampel et al., 2017a) at around 419 and 360 nm in a controlled laboratory environment with wellknown light path lengths at different pressure and humidity values.The availability of such a data set has the potential to improve the spectral retrieval of various trace gases substantially as pointed out below.
The O 4 absorption peak at 328 nm was unambiguously identified.Its magnitude agrees with a previous publication by Salow and Steiner (1936) and is found to be 0.19 ± 0.02 of the magnitude of the next absorption peak at 344 nm.This results in a maximum peak absorption cross section based on Thalman and Volkamer (2013) of (1.96 ± 0.20) × 10 −47 cm 5 molec −2 at 328.59 ± 0.15 nm.
The impact on incoming sun radiation in the spectral region from 323 to 331 nm is small and amounts to 7.8 × 10 −4 Wm −1 for zenith sun and an O 4 vertical column density (VCD) of 1.3 × 10 43 molec 2 cm −5 using the solar atlas by Chance and Kurucz (2010).It has therefore only a small impact on the radiative balance in the atmosphere.
It is interesting to note that the potential SO 2 oxidation by singlet oxygen (Sorokin, 2010) requires singlet oxygen molecules with the vibrational eigenstate ν > 2. In particular this weak absorption near 328 nm (and near 315) leads to the formation of singlet oxygen molecules in the third (and fourth) vibrationally excited state (see Table 2) and thus could play a role for SO 2 oxidation in the atmosphere.
The impact on trace gas retrievals depends on the fit settings of the respective trace gas and instrumental properties.For DOAS measurements in pristine to semi-polluted regions, a significant impact is expected for spectral retrievals of HCHO, BrO, SO 2 and OClO, which could encompass this spectral region.Previous publications often avoided this spectral region.We suggest that the reason for the previously observed discrepancies was often the O 4 absorption, which was not accounted for, rather than the increasing influence of tropospheric and stratospheric ozone absorption towards shorter wavelengths.
Incorporating this O 4 absorption in the spectral retrievals of the above-mentioned absorbers will lead to a substantial improvement of the respective detection limits as additional absorption bands can be included in the spectral retrieval.In our evaluations, extending the fit range lower limit from 332.5 to 322.5 nm led to a reduction of the fit error by ≈ 35 % for HCHO and BrO.It furthermore significantly reduced the previously observed interferences between the BrO and HCHO absorption structures (Pinardi et al., 2013;Volkamer et al., 2015).
The O 4 absorption peak at 419 nm was difficult to identify using the method presented here, as it is difficult to exclude the possible influence of water vapour absorption which overlays the O 4 absorption structure.For water vapour absorption in this spectral range significant uncertainties were reported by (Lampel et al., 2015b).Its magnitude is estimated to be about 0.08 ± 0.05 of the absorption peaks at 446 nm.No published data are available for the absorption at 419 nm.Based on Thalman and Volkamer (2013), this results in a peak absorption cross section of (5.0 ± 3.5) × 10 −48 cm 5 molec −2 at a wavelength of 419.13 ± 0.42.Also in the case of the 419 nm absorption, the impact on trace gas retrievals depends on the fit settings for the respective trace gas and instrument parameters, but an influence can be expected for the spectral retrievals of weak absorbers in pristine regions, such as IO and NO 2 .
However, the O 4 absorption peak at around 419 nm cannot explain the observed differences between different water vapour absorption cross sections in recent literature (compare Lampel et al., 2015b) but could contribute to previously observed systematic residual structures.
Data availability.The spectra used for the data evaluation in this paper (about 10 GB) can be obtained on request from the author.
Information about the Supplement According to the procedure described above, a merged absorption cross section of O 4 based on Thalman and Volkamer (2013) was calculated.The absorption peak at 344 nm was shifted by +1366 cm −1 to shorter wavelengths and scaled by 0.1863 according to Table 3 and additionally the radiative transfer correction by 14 % (as in Sect.4.2.3).These crosssection values were then added to the original O 4 absorption cross section at 293 K below 331 nm.To avoid negative cross-section values below 337.5 nm after convolution to instrument resolution, absorption cross-section values between 331 and 337.5 nm were set to zero.For the shifted absorption peak from 446.4 to 419.0 nm values between 409.25 and 426.0 nm were used, while the original O 4 absorption cross section at 293 K between 390.0 and 435.5 nm was set to zero.The peak was scaled by 0.0785 and corrected for radiative transfer dependence with wavelength by 22 %.The resulting file is provided as a Supplement.
Competing interests.The authors declare that they have no conflict of interest.

Figure 1 .
Figure 1.Shifted (a by wave number; b by wavelength) and normalized O 4 absorption cross sections at 446 and 477 nm according to Thalman and Volkamer (2013).

Figure 3 .
Figure 3. LP-DOAS correlation of the suspected 328 nm absorption feature with the known absorption at 343.8 nm.Data for the short light path are shown in orange and for the long path in blue.Linear regression results are shown with and without for a constant offset term.

Figure 4 .
Figure 4. MAX-DOAS fit fromPPAO, 6 July 2015, 11:18 UTC, at 3 • elevation and a 90 • elevation reference spectrum.Spectra were coadded to obtain a total exposure time per spectrum of 4 min.The 328 nm O 4 absorption structure found here is 14 times as large as the DOAS fit error.Without considering this absorption, the HCHO dSCD in this case is larger (+2.5 × 10 14 molec cm −2 or 1.4 %), and the RMS of the residual increases by +22 %. "RS-temp" denotes the difference between Ring pseudo-absorption spectra calculated at 273 and 243 K as listed in Table1.

Figure 5 .
Figure 5. Correlation of O 4 dSCD at 344 nm and the fitted O 4 OD at 328.5 nm assuming the same shape of the absorption cross section as at 344 nm, using the PPAO MAX-DOAS data set.The wavelength dependence of the O 4 AMF was not corrected here.

Table 1 .
Retrieval wavelength intervals for DOAS measurements.Values in parentheses were used for sensitivity studies only.

Table 2 .
Relative intensities and peak wavelengths of O 4 absorption peaks below 500 nm: ElKn