Quantitative comparison of measured and simulated O 4 absorptions for one day with extremely low aerosol load over the tropical Atlantic

. In this study we compare measured and simulated O 4 absorptions for conditions of extremely low aerosol optical depth (between 0.034 to 0.056 at 360nm) on one day during a ship cruise in the tropical Atlantic. For such conditions, the 10 uncertainties related to imperfect knowledge of aerosol properties don’t significantly affect the comparison results. We find that the simulations underestimate the measurements by 15% to 20%. Even for simulations without any aerosols the measured O 4 absorptions are still systematically higher than the simulation results. The observed discrepancies can not be explained by uncertainties of the measurements and simulations and thus indicate a fundamental inconsistency between simulations and measurements


Introduction
Remote sensing measurements of the atmospheric absorption of the oxygen dimer (O 2 ) 2 are often used to derive properties of aerosols and clouds. The atmospheric concentration of (O 2 ) 2 (in the following referred to as O 4 ) varies only slightly with 20 temperature, pressure and humidity variations (aside the dependence on altitude). Thus deviations from the O 4 absorptions for clear sky conditions indicate changes of the atmospheric radiative transfer, e.g. due to clouds and aerosols. In recent years, inconsistencies between the measured atmospheric O 4 absorption and radiative transfer simulations were detected for Multi-AXis-DOAS (MAX-DOAS) observations. MAX-DOAS instruments measure scattered sun light under different, mostly slant elevation angles (Hönninger and Platt, 2002). Several studies found that a scaling factor (SF<1) had to be 25 applied to the observed atmospheric O 4 absorptions in order to bring them into agreement with radiative transfer simulations.
Other studies, however, did not find the need to apply such a scaling factor (e.g. Ortega et al., 2016; see also discussion in Wagner et al., 2019, and references therein). One major difficulty in the quantitative interpretation of these comparisons is that usually the atmospheric aerosol properties are not well known (e.g. the vertical extinction profile and/or the optical properties). And even if they were known, it is still a challenge to accurately represent them in atmospheric radiative transfer 30 simulations.
In this study we mainly minimise these difficulties by using atmospheric observations in the presence of very low aerosol loads. During a ship campaign across the tropical Atlantic, very low aerosol optical depth (AOD) was observed on one day (2 May 2019). At 360 nm (the wavelength at which we analyse the atmospheric O 4 absorption), the AOD ranged from 0.034 to 0.056, which is an order of magnitude lower than the optical depth of molecular Rayleigh scattering.

35
Like in previous studies, we compare the observed atmospheric O 4 absorption with the results of radiative transfer simulations. Information about the aerosol properties is derived from sun photometer measurements in combination with ceilometer measurements. Also in our study, considerable uncertainties about the aerosol vertical profile and the aerosol optical properties exist. However, these uncertainties are less important for the interpretation of the comparison results than in previous studies because of the low AOD, and we find large discrepancies between the measured and simulated O 4 40 absorptions.
The paper is organised as follows: In section 2, an overview on the ship campaign and the instruments used in this study is given. Sections 3 to 5 describe the spectral analysis, the cloud classification, and the calculation of the O 4 profile. In section https://doi.org/10.5194/amt-2020-457 Preprint. Discussion started: 19 November 2020 c Author(s) 2020. CC BY 4.0 License. 6, the radiative transfer simulations and the extraction of the aerosol extinction profiles are presented. Section 7 presents the comparison results, and section 8 the summary and conclusions.

2 Overview on the ship campaign and the instruments used in this study
The MAX-DOAS measurements were carried out during a cruise (MSM82/2) of the German research vessel RV Maria S. Merian (https://www.ldf.uni-hamburg.de/merian.html) from Montevideo (Uruguay) to Las Palmas (Spain) from 26 April 50 2019 to 14 May 2019 (see Fig. 1). More details on the ship cruise MSM82/2 can be found in Krastel et al. (2019). In this study, we focus on one day with particularly low AOD (2nd May), which is marked in Fig. 1.
The MAX-DOAS instrument was mounted above the ship's bridge at about 20m altitude above sea level. The telescope was aligned in the driving direction of the ship (Fig. 2). 55

MAX-DOAS instrument
The MAX-DOAS instrument is a so-called Tube MAX-DOAS instrument which was developed and built by the electronic workshop of the Max Planck Institute for Chemistry in Mainz (Donner, 2016). It consists of two major parts: the telescope unit and the spectrometer unit. The telescope unit is mounted outside on the railing of the ship. The spectrometer unit is 60 located inside the ship. Besides the spectrometer it also contains a peltier cooling element which stabilises the spectrometer temperature at 15 °C. Both units are connected via a quartz glass fibre bundle and electric cables. The telescope unit is equipped with a gyroscope to stabilise the elevation angles by continuously adjusting the motor position with an accuracy of ±0.1°.
The spectrometer is an Avantes ULS2048x64-USB2. It covers the spectral range from 299.4 nm to 463.1 nm with a spectral 65 resolution between 0.52 and 0.54 nm as described by the full width half maximum (FWHM). Spectra are measured with an integration time of 1 min at the following elevation angles: -2°, -1°, -0.5°, 0°, 0.5°, 1°, 2°, 3°, 4°, 5°, 6°, 8°, 10°, 15°, 30°, 90°. Note that in this study only measurements with positive elevation angles are used. One elevation sequence is completed within about 21 min. Dark current and offset spectra are taken during night and are used to correct the measured spectra before the spectral analysis.

Sun photometer
A MICROTOPS II sunphotometer provided atmospheric totals on aerosol and water vapor. The instrument, when directed towards the sun (in a handheld operation), captures via diodes the solar intensity in five sub-spectral bands near wavelengths 75 of 380, 440, 670, 870 and 940nm. In combination with the larger reference solar intensity at the top of the atmosphereusing time and (GPS-provided) position data -sun-photometer measurements define the atmospheric attenuation at these solar sub-spectral bands. Four spectral bands (near 380, 440, 670 and 870 nm) sample in trace-gas poor regions, while one spectral band (near 940nm) is strongly affected by water vapor absorption. In the absence of clouds, the solar attenuations in the four trace-gas poor bands can be linked to aerosol -after (surface air pressure defined) contributions from air-molecule 80 (Rayleigh) scattering have been removed. Hereby, the aerosol associated attenuations are quantified by the (vertically normalized) aerosol optical depth (AOD). As the instrument offers AOD values simultaneously at four different solar wavelengths, the typical aerosol particle size is revealed and even AOD contributions from sub-micrometer (mainly from pollution and wildfire) and super-micrometer size aerosol particles (mainly from dust and seasalt) can be distinguished. The determination of the atmospheric water vapor is based on the differential absorption between 870 and 940 nm attenuation 85 data. Any quality measurement usually relies on many repeated samples in order to identify and remove poor data associated https://doi.org/10.5194/amt-2020-457 Preprint.

110
The results of the spectral analysis represent the integrated trace gas concentration along the atmospheric light path, the socalled slant column density (SCD). For O 4 the SCD is expressed with respect to the square of the O 2 concentration (see Greenblatt et al., 1990). Thus, the unit of the O 4 SCD is molec²/cm 5 . For the analysis of the measured spectra, a so-called Fraunhofer reference spectrum is used. In this study, the Fraunhofer reference spectrum is calculated as the average of the zenith spectra before and after the chosen elevation sequence, weighted by the time of the selected measurement from that 115 elevation sequence. Before performing the spectral analysis, these sequential Fraunhofer reference spectra are fitted to a ‚universal' Fraunhofer reference spectrum (29 April 13:43, SZA: 44.8°, elevation angle: 90°) to transfer the spectral calibration of the universal Fraunhofer reference spectrum to the sequential Fraunhofer reference spectra. The universal Fraunhofer reference spectrum was calibrated using a high resolved solar spectrum.
Since the Fraunhofer reference spectrum also contains atmospheric trace gas absorptions, the output of the spectral analysis 120 represents the difference between the SCDs of the selected non-zenith spectrum and the Fraunhofer reference spectrum, the so-called differential SCD (or dSCD).
The typical fit error of the derived O 4 dSCD is between 2·10 41 molec 2 /cm 5 and 4·10 41 molec 2 /cm 5 . Depending on the magnitude of the retrieved O 4 dSCD this corresponds to relative errors between 1 and 4 %.

4 Cloud detection using the MAX-DOAS measurements
Although during most of the afternoon on 2 May clear sky conditions prevailed, also some scattered clouds were present.
They were e.g. detected by the ceilometer in zenith direction (see Fig. 3). In order to derive information about possible cloud https://doi.org/10.5194/amt-2020-457 Preprint.  (2014,2016). Figure A3 in the appendix shows the time series of the retrieved O 4 dSCDs on 2 May for the different elevation angles. During the morning the O 4 dSCDs show strong variability caused by the presence and variability of clouds as also seen in the ceilometer data (Fig. 3)

150
Here M H2O is the mixing ratio of water vapour taken from the ERA interim data. For the mixing ratio of O 2 (M O2 ) a value of 21% is assumed. The O 4 concentration is then represented by the square of the O 2 concentration (Greenblatt et al., 1990). To derive the O 4 VCD, the O 4 concentration is vertically integrated between the surface and 30 km with a vertical resolution of 20 m.

155
The temperature and pressure from the ECMWF ERA-Interim data set at the surface are also compared to the in situ measurements on the ship. It is found that the ECMWF temperature is slightly lower (-0.7 K) and the ECMWF pressure is slightly higher (+2 hPa) than the corresponding in situ measurements, see To estimate the uncertainty of the derived O 4 VCD the temperature and pressure of the whole profiles are varied by ±2 K and ±2 hPa, respectively. The resulting changes of the O 4 VCDs are ±1.5 % and ±0.9 %, respectively. In addition, assuming an uncertainty of the atmospheric humidity profile of 30% leads to an uncertainty of the derived O 4 VCD of 0.9 %. Thus, we estimate the total uncertainty of the O 4 VCD to ±2 %.  Table 2). The surface albedo was set to 0.05.   Figure 4 presents the hourly averaged and range corrected ceilometer backscatter profiles for three periods in the afternoon 200 on 2 May 2019 without cloud contamination. In a first order approximation, these backscatter profiles are proportional to the aerosol extinction. Thus together with the total AOD from the sun photometer measurements, the aerosol extinction profiles can be determined. However, ceilometer measurements are affected by several instrumental limitations, which complicate the direct conversion to aerosol extinction profiles: a) Due to the missing overlap between the outgoing beam and the field of view of the detector, the sensitivity of the 205 ceilometer is very low for altitudes below 500 m. Thus for this altitude range, no information on the aerosol extinction can be derived from the ceilometer measurements. b) In spite of the long averaging period, still strong noise appears for altitudes above 3 km.

Extraction of the aerosol extinction profiles
Due to these limitations, the ceilometer profiles can only be used for a restricted altitude range. In the following we used the ceilometer profiles for the altitude range between 500 m and about 7 to 9 km. Between 500 m and 3000 m, averages for 100 Before the backscatter profiles are normalised with the total AODs measured by the sun photometer, the stratospheric part of the total aerosol profile has to be added. This step is usually not important, because in more polluted areas the total AOD is clearly dominated by the tropospheric part. However, for our study, the total AOD is so low that the stratospheric part

Calculation of effective temperatures for the O 4 absorption
Since the temperature of the troposphere decreases with altitude, and the O 4 absorption cross section depends on temperature, the retrieved O 4 dSCDs might deviate from the true O 4 dSCDs (the integrated O 4 concentration along the atmospheric light paths), because only one O 4 cross section for a fixed temperature is used in the spectral analysis. Thus, 235 before the O 4 dAMFs from the measured spectra are compared to those from the radiative transfer simulations, the effect of the temperature dependence of the O 4 absorption has to be investigated.
The effective temperature of the O 4 measurements is calculated according to: Here [O 4 ] z represents the O 4 concentration at altitude z, bAMF z,α the box-AMF for elevation angle α at altitude z, and T z the temperature at altitude z. T eff,α is the effective temperature for the measured O 4 dSCD at elevation angle α.

Direct comparison between measurements and RTM results
In Fig. 6 the O 4 dAMFs derived from the MAX-DOAS measurements are compared to those obtained from the radiative transfer simulations for elevation sequences not affected by clouds (similar comparisons for the sequences with cloud-260 contaminated measurements are shown in Fig. A11 in the appendix).
In the left part of Most valid profiles are obtained for scaling factors between 0.80 and 0.90, or for a free fitted (variable) scaling factor. For the inversions with larger scaling factors, rather high RMS are found. For most cases, the retrieved AODs are smaller than 295 those measured by the sun photometer. However, here it should be noted that for these low aerosol extinctions, the information content of the measurements is probably too low to constrain the aerosol extinction profiles, especially for high altitudes.
The obtained scaling factors are shown in Fig. 7. Overall good agreement between both comparison methods is found. For 300 all elevation sequences, values of the scaling factor < 1 are found. For the direct comparison, the difference from unity is mostly larger than 15 % and can thus not be explained by the uncertainties of the measurements and simulations.

305
We compared measured and simulated O 4 absorptions for one day with very low aerosol optical depth. For such conditions, the uncertainties caused by imperfect knowledge of the aerosol properties play a smaller role than for comparison under more polluted conditions. One important result of the comparison was that for all measurements, the observed O 4 absorption was higher than the simulation results for an atmosphere without aerosols. In the absence of optically thick clouds, the simulated O 4 dAMFs for 310 an atmosphere without aerosols constitutes an upper limit, since especially for the low elevation angles the inclusion of aerosols leads to a decrease of the O 4 absorption. The observed discrepancies thus indicate a fundamental inconsistency between simulations and measurements.
The measured O 4 absorptions are also compared to simulations including aerosol extinction profiles. The aerosol extinction profiles were constrained by measurements of the sun photometer, the ceilometer and a climatology of stratospheric aerosols.

315
Again, a large discrepancy was found for the absolute values. However, for the relative dependence of the O 4 dAMFs on the elevation angle good agreement could be achieved. For each elevation sequence, the ratio of simulated and measured O 4 dAMFs was calculated. For that purpose the elevation angles >4° were used, for which the O 4 dAMFs are almost insensitive to the profile shape in the lower atmospheric layers. For all elevation sequences, ratios of 0.85 or less were found. Similar ratios were also obtained from the application of our profile inversion algorithm (MAPA) to the measurements. The 320 observed discrepancies cannot be explained by the uncertainties of measurements and/or simulations. Here it is important to note that in the spectral analysis, we explicitly corrected for the (small) temperature dependence of the atmospheric O 4 absorption.
Our results indicate that something fundamental is missing/wrong in either the radiative transfer simulations or the spectral analysis of the atmospheric O 4 absorptions. We did not find a clear reason for the discrepancies. One possible reason for the 325 discrepancies could be a systematically too small O 4 absorption cross section.

H 2 O cross section
Recent studies found evidence for substantial atmospheric H 2 O absorptions in measured spectra in the UV (Lampel et al., 2017, Wang et al., 2017. These absorptions are usually rather small, but especially for measurement conditions with 590 high atmospheric humidity the inclusion of a H 2 O cross section in the spectral analysis can be useful.        Shown are the ratios of the obtained O 4 dSCDs for different selections versus those for Fraunhofer reference spectra interpolated between the zenith measurements before and after the selected elevation sequence. Before: zenith measurent before the sequence is used; After: zenith measurent before the sequence is used; Average before and 665 after: the average of the zenith measurents before and after the sequence is used.