Consistency of total column ozone measurements between the Brewer and Dobson spectroradiometers of the LKO Arosa and PMOD/WRC Davos

Total column ozone measured by Brewer and Dobson spectroradiometers at Arosa and Davos, Switzerland, have systematic seasonal variations of around 1.5% using the standard operational data processing. Most of this variability can be attributed to the temperature sensitivity of approx. +0.1%/K of the ozone absorption coefficient of the Dobson spectroradiometer (in this study D101). While the currently used Bass&Paur ozone absorption cross-sections produce inconsistent results for Dobson and Brewer, the use of the ozone absorption cross-sections from Serdyuchenko et al. (2013) in conjunction with an 5 effective ozone temperature dataset produces excellent agreement between the investigated four Brewers (of which two double Brewers), and Dobson D101. Even though other ozone absorption cross-sections available in the literature are able to reduce the seasonal variability, all of those investigated produce systematic biases in total column ozone between Brewer and Dobson of 1.1% to 3.1%. The highest consistency of total column ozone from Brewers and Dobson D101 at Arosa/Davos of 0.1% is obtained by applying the Rayleigh scattering cross-sections from Bodhaine et al. (1999), the ozone absorption cross-sections 10 from Serdyuchenko et al. (2013), the effective ozone temperature from either ozonesondes or ECMWF, and the measured line-spread functions of Brewer and Dobson. The variability between Brewer and Dobson for single measurements of 0.9% can be reduced to less than 0.5% for monthly means and 0.3% on yearly means. As show here, the proposed methodology produces consistent total column ozone datasets between Brewer and Dobson spectroradiometers of better than 1%. For colocated Brewer and Dobson spectroradiometers, as is the case for the Arosa/Davos total column ozone times series, this allows 15 the merging of these two distinct datasets to produce a homogeneous time series of total column ozone measurements. Furthermore, it guarantees the long-term future of this longest total column ozone time-series, by proposing a methodology how to eventually replace the ageing Dobson spectroradiometer with the state-of-the art Brewer spectroradiometer.


Introduction
The world's longest continuous total column ozone time series was initiated in 1926 at the Lichtklimatisches Observatorium 20 (LKO), at Arosa, in the Swiss Alps. It consists of a suite of six instruments, 3 Dobson and 3 Brewer spectroradiometers, which measure the direct spectral solar irradiance at several narrow wavelength bands between 305 nm and 345 nm. Since 2010, Brewer B072 has been relocated to the Physikalisch-Meteorologisches Observatorium Davos, World Radiation Centrer by June 2021.
The Brewer spectroradiometers were installed at LKO Arosa in 1988(Brewer B040), 1991(Brewer B072), and 1998 to complement the Dobson spectroradiometers already in operation. The objective has been to eventually replace the Dobson spectroradiometers with the more modern and automated Brewer spectroradiometers. However colocated measurements between Dobson and Brewer spectroradiometers have shown seasonal variations of the order of 2% to 3% which have 35 been linked to the effective ozone temperature (Köhler et al. , 2018;Vanicek et al. , 2012;Staehelin et al. , 1998;Redondas et al. , 2014). These known systematic variabilities between the two instrument types have so far precluded a merging of datasets from these instruments and have led to the continuous simultaneous operation of the Dobsons and Brewers at LKO Arosa.
A study by Redondas et al. (2014) on the use of different ozone absorption cross-sections in Brewer and Dobson total column ozone retrieval algorithms has shown that most of the observed discrepancies can be reconciled by taking into account the tem-40 perature dependence of the ozone absorption cross-sections of the Dobson spectroradiometer. The study investigated five ozone absorption cross-sections, accessible at the ACSO (Absorption Cross Sections of Ozone) initiative web page (http://igaco-o3.fmi.fi/ACSO). Redondas et al. (2014) came to the conclusion that the cross-sections measured by Serdyuchenko et al. (2013) substantially reduced the observed seasonal dependence of the differences between Dobson and Brewer.
In this study, we revisit the issue of the consistency of Brewer and Dobson total column ozone measurements discussed in 45 Redondas et al. (2014), with the following additional elements: -An additional dataset of ozone absorption cross-sections measured in the frame of the EMRP project ATMOZ (M. Weber, personal communication).
-A dataset of absorption cross sections measured within the ESA SEOM-IAS project (Scientific exploitation of Operational Missions -Improved Atmospheric Spectroscopy Dabases) (Birk et al. , 2018).

Total ozone column measurement
Brewer and Dobson spectroradiometers retrieve the atmospheric total column ozone from the absorption of direct solar radiation by the atmosphere, as described by the Beer-Lambert law, where λ is the wavelength, I 0 is the solar irradiance at the top of the atmosphere, I the irradiance measured at the surface, τ 60 the optical depth of the atmosphere and µ the slant pass (airmass). The equation can be rewritten as, to retrieve the optical depth τ at a specific wavelength. In the ultraviolet range between 300 nm and 340 nm the main atmospheric absorber is ozone. Even though sulphure dioxide also absorbs in this wavelength range, its amount in the atmosphere above Arosa/Davos is so small that it can be neglected here. The only further attenuation comes from Rayleigh scattering (τ β ), which needs to be subtracted from the total optical depth τ to retrieve the optical depth for ozone, where X is the total amount of ozone in the atmospheric column, α(λ) the ozone absorption cross-section and µ β and µ O3 the effective airmass for Rayleigh scattering and ozone absorption respectively.
While the Brewer spectroradiometer uses a slit mask to measure quasi-simultaneously the solar irradiance at four wave-70 lengths, the Dobson spectroradiometer measures the intensity ratio at two wavelengths pairs by adjusting an optical wedge to attenuate the higher of the two intensities with respect to the other. The total column ozone X is then retrieved by a linear combination of the solar radiation measurements at these four wavelengths,

Brewer and Dobson spectroradiometers
The spectroradiometers operating at Davos and Arosa belong respectively to the global Brewer and Dobson networks and are 80 traceable to their respective network reference instruments, Dobson D083 via the RVI regional Dobson D064 from Hohenpeissenerg, Germany, and the Brewer Triad maintained by the Regional Brewer calibration center -Europe (RBCC-E), at the Izaña Atmospheric Research Center, Tenerife, Canary Islands, Spain . For details on the maintenance and calibration schedules we refer to the publications of Scarnato et al. (2010), and . The calibration reports from the latest calibration campaigns which took place in July 2018 are available at the WMO GAW report portal 85 (https://library.wmo.int/index.php?lvl=notice_display&id=21655#.X5rL8IhKhaQ9) and on request from the authors. The consistency between the Brewer spectroradiometers operated at Arosa/Davos is of the order of 0.5% . A recent analysis for the manually and automatically operated Dobson spectroradiometers at Davos and Arosa have found a similar consistency of 0.5% (Stübi et al. , 2020).
The four Brewer spectroradiometers consist of two single monochromator Brewers MKII B040 and B072, and two double 90 monochromator Brewers MKIII B156 and B163. From the three Dobsons, only D101 and D062 are continuously measuring total column ozone, while D051 is usually reserved for Umkehr measurements and is only occasionally fitted with the zenith prism for direct solar radiation measurements. As the line spread function measurements with the TuPS were not successful on D062 due to the low sensitivity of the instrument, only D101 will be used in this study.
The top of the atmosphere calibration constants F 0 were obtained through direct comparisons with the reference spectrora-95 diometers from each network during site-visits to Arosa/Davos. Following the standard operating procedures, the measurements are then corrected for instrument drifts between calibrations by applying standard lamp corrections which are automatically made several times per day for Brewer spectroradiometers and manually on a weekly schedule for Dobson spectroradiometers.

Line spread functions
The spectral resolution and wavelength calibration of the Brewer spectroradiometers is obtained from the measurements of a 105 set of spectral emission lines from several spectral discharge lamps, as described in Gröbner et al. (1998).The ozone absorption     Table 1 shows the spectral characteristics of the four Brewers and of Dobson D101 used in this study, as well as the nominal wavelength positions of the Dobson according to the Dobson handbook. Figure 1 shows the measured line spread functions of Dobson D101 as well as the nominal slit functions used for the calculation of the operational ozone absorption cross-sections according to Bernhard et al. (2005).

Effective ozone temperature and ozone height from ozone sondes
The effective ozone temperature T eff is calculated from ozone and temperature profiles measured by ozone sondes launched every two to three days at the aerological station of MeteoSwiss in Payerne, 220 km distance from Davos. The effective ozone temperature is obtained from the integral of temperature T (z) and ozone density O 3 (z) profiles according to, where z represents the vertical height above ground. The ozone and temperature above the sonde burst height, which occurs typically around 30 km is obtained by extending the measured ozone and temperature profiles with the standard ozone and temperatures taken from the standard US atmosphere (NOAA, 1976) and normalised to the measurements at the sonde burst height. Figure 2 shows the effective ozone temperature over Payerne from 2016 to 2020. The effective ozone temperature shows 130 a distinct seasonal variation, with an average effective ozone temperature of 225.2 K throughout the period and with a 95% coverage between 215 K in winter and 232 K in summer. The average seasonal variability of the effective ozone temperature is 11.4 K as estimated from a periodic fit to the data.
During the winter, the effective ozone temperature can fluctuate significantly, with changes of up to 15 K within a few days (e.g. from 225 K on 27 January 2016, to 209 K on 3 February 2016 and back to 224 K on 10 February 2016). The effective ozone 135 temperatures on days without ozone sonde launches were obtained from linearly interpolating between available measurements.
The resulting dataset was smoothed with a 10 day running average to reduce the day to day noise.
We have also investigated the effective ozone temperature obtained from the ECMWF reanalysis (http://www.temis.nl/ climate/efftemp/overpass.html) for Payerne, shown as the yellow curve in the figure. The two datasets agree very well at larger temperature above 220 K (deviations below 2 K), while at lower temperatures the temperatures from ECMWF systematically 140 underestimate the effective ozone temperatures obtained from the radio-sondes by up to 4 K. Even though we will concentrate on the dataset obtained from the radio-sondes, we will briefly comment on the use of the ECMWF dataset for the recalculation of the Dobson total column ozone time series later in the manuscript.
The corresponding effective ozone height H eff was also calculated by replacing the temperature profile in Eq. 5 with the height profile. The resulting average effective ozone layer over Payerne is found at 22.3 km, with a minimum of 20.2 km 145 during the winter and a maximum of 23.8 km during the summer. The impact of this variability on the calculation of the ozone slant path for the measurements at Arosa/Davos is at most 0.3% at an airmass of 3.9 (SZA of 76), which is the maximum that can be reached due to the mountains blocking the horizon at both sites. Notwithstanding the small expected impact, for consistency reasons we have incorporated the calculated ozone layer height in the calculations of total column ozone from the Brewers and Dobson in this study. The effective molecular scattering height relevant for Rayleigh scattering was kept at 5 km 150 as used in the nominal operating procedure.

Ozone absorption cross-sections
The operational ozone absorption coefficients in use by the Brewer and Dobson networks are based on the ozone crosssections from Bass and Paur (1985) and are presented in Redondas et al. (2014) and the references therein. Furthermore, we have selected datasets of ozone absorption cross-sections available for the wavelength range 300 nm to 345 nm, which 155 cover the wavelength range used by Brewer and Dobson spectroradiometers to retrieve total column ozone. The selected crosssections are also available at several ozone temperatures, in order to calculate the ozone absorption cross-sections for the actual effective ozone temperature at our sites. If not specified otherwise, these cross-sections were retrieved from the ACSO webpage (http://igaco-o3.fmi.fi/ACSO/cross_sections.html). The following six ozone absorption cross-sections were used in this study: DBM The high resolution dataset of Daumont et al. (1993), Brion et al. (1993), and Malicet et al. (1995) at five temperatures between 218 K and 295 K. Due to the lack of measurements at temperatures below 218 K, the quadratic temperature approximation gives inconsistent values at temperatures lower than 218 K. As a compromise, we have therefore decided to calculate instead a linear temperature dependence function from the three lowest temperatures of the DBM dataset to  function from this dataset to calculate the ozone absorption cross-sections for this study.

Ozone absorption coefficients
The ozone absorption coefficient α i at wavelength λ i is calculated by convolving the selected ozone absorption cross-sections σ(λ, T eff ) for a particular effective ozone temperature T eff with the line spread function S i (λ) of the spectroradiometer,

185
where i specifies the central wavelength at which the measurement is taken. The ozone absorption coefficient ∆α from equation 4 can then be calculated,    Table 2. As can be seen in Table 2, the relative differences between the ozone absorption coefficients are the same for all Brewers.
This indicates that a change in absorption cross sections for a Brewer spectroradiometer operated according to the standard procedure, produces the same relative change in total column ozone. This is an important result for the Brewer network, offsets. However as seen in Table 2, the ozone absorption coefficient increases slightly by 0.3% when using the measured slit functions with respect to the parametrized slits. Similar shifts between -0.6% to +0.3% were found for three Dobson spectroradiometers which were characterised during the project EMRP ATMOZ (Köhler et al. , 2018), indicating that the specific spectral characteristics of a Dobson spectroradiometer have an impact of at least this magnitude on the retrieval of total column ozone. This demonstrates the need for an individual spectral characterisation of each Dobson spectroradiometer 210 to reach uncertainties of 1 % or lower.

Rayleigh scattering coefficient
The Rayleigh scattering coefficients β i at wavelength i are calculated by replacing the ozone absorption cross-section σ(λ, T eff ) in Eq. 6 with the Rayleigh scattering coefficient β(λ), calculated using the parametrization from Bodhaine et al.

Results and discussion
The datasets obtained from each ozone absorption cross-section were compared between the four Brewer spectroradiometers  to 30 June 2020 using the operational ozone retrieval procedure with the ozone absorption cross-sections from (Bass and Paur , 1985). The yellow curve is a periodic fit to the data, while the black curve is obtained by combining the effective ozone temperature with the linear temperature dependence retrieved from the dataset. Right figure: the same relative differences, shown versus effective ozone temperature.

Operational Bass&Paur cross-sections
The left panel of Fig. 4 shows the relative difference between the total column ozone from Brewer B156 to Dobson D101, 250 calculated using the operational procedures. The amplitude of a sinusoidal fit to the data with a period of one year gives an amplitude of 0.86%, with an average offset between the two datasets of 1.0%. If the full variability of the data is examined, then the 95% variability between Brewer B156 and Dobson D101 extends over 3.9% (from -0.75% to +3.2%), which is a significant inconsistency that needs to be addressed. Similar offsets and seasonal variations relative to Dobson D101 are also obtained for Brewers B040, B072 and B163.

255
As expected, this seasonal variability correlates very well with the effective ozone temperature as shown in the right panel of Fig. 4. Indeed, the temperature coefficient of -0.14 %/K retrieved from a linear fit to the data, combined with the amplitude of the average seasonal effective temperature variation of 5.7 K, gives an amplitude of 0.80%, in excellent agreement with the amplitude of 0.86% of the periodic fit to the dataset shown in the left panel of Fig. 4. This indicates that the seasonal variability observed between the Brewers and Dobson D101 is strongly correlated to the effective ozone temperature.

260
The relative differences between the datasets calculated with different ozone absorption cross sections and taking into account the effective ozone temperature are shown in Fig. 5 with respect to the operational datasets for Brewer B156 and Dobson D101. The panels in the figure are based on the values shown in Table 2 and are therefore mainly a visual illustration of the impact of changing to a different ozone absorption cross-section with its specific ozone temperature dependence. Note that the Rayleigh correction of -2.4 DU was also included in the operational ozone dataset for Brewer B156, so it cancels out when 265 taking the ratios shown in Fig. 5.
As expected, the amplitude of the periodic sine-fit shown in the sub-figures agree very well with the temperature coefficients calculated for each ozone absorption cross-section shown in Table 2 multiplied with the effective ozone temperature and also to the seasonal fit of the ozone ratios between Brewer and Dobson shown in black in the figures. This excellent agreement indicates that the observed seasonal variability is mainly due to the temperature dependence of the ozone absorption coefficients 270 calculated for Brewer or Dobson.
The most obvious feature apparent in Fig. 5 is the pronounced seasonal variability of the Dobson D101 datasets due to the inclusion of the effective ozone temperature in the calculations. On the other hand, the average offsets to the operational dataset are for the most part below 1%, with the exception of the ACS data, with an offset of +2.4%. In contrast, the changes for Brewer B156 show a much more reduced seasonal variability relative to the operational dataset, which indicates that the 275 Brewer ozone measurement procedure (wavelength settings and relative weighting) produces an ozone absorption coefficient largely insensitive to the ozone temperature. On the other hand, the offsets between the ozone absorption coefficients calculated from these cross-sections vary between +1.7% for the IGQ dataset to -4.1% for DBM and are therefore much larger than for the Dobson. This larger variability is probably due to the smaller spectral widths of the line spread functions of the Brewer, which is therefore more sensitive to spectral noise in the cross-sections than from the wider Dobson line spread functions (see 280 Eq. 6).

Consistency between Brewer and Dobson total column ozone datasets
The comparisons between the datasets of Brewer B156 and Dobson D101 are shown in Fig. 6 for the five cross-sections. The comparisons between the other three Brewers to D101 are summarised in Table 3 and are shown as figures in the supplement.
A possibe confounding factor could be the different locations of the Brewer and Dobson spectroradiometers during the study 285 period, with Brewer B156 colocated with Brewer B040 from 2016 to 2018 at Arosa, and then moved to Davos for the remaining period. However, as discussed in , the observed differences in total column ozone between the two sites has shown no significant differences, and so no distinction has been made between the instruments located at Arosa or Davos.
As seen in Fig. 6 for Brewer B156 to Dobson D101, the relative differences in total column ozone show substantial variations, depending on which cross-section is used to calculate total column ozone. The largest seasonal variation in total column 290 ozone measured by Brewer B156 relative to Dobson D101 is seen with the IGQ cross-sections, with an amplitude of 0.75% and an average offset of +1.8%. The observed seasonal variation is mainly due to a left-over dependence on effective ozone temperature as seen by the good agreement between the black and yellow curves. This implies that the temperature dependence of the IGQ cross-sections fail to account for the actual temperature dependence of the total column ozone retrieved by either Brewer or Dobson.

295
As can be seen in Table 3, there is a slight systematic difference in the seasonal variability of the single Brewer spectroradiometers relative to Dobson D101, compared to the one seen with the double Brewer spectroradiometers. Since single Brewer spectroradiometers are affected by stray-light at large ozone slant path, we have applied a correction based on the stray-light correction determined during the RBCC-E campaign in 2018, with respect to the reference double Brewer B185.
The corrections are small, but slightly improve the consistency with the results from the double Brewers, as can be seen in 300   Table 3. seasonal amplitude equal or less than 0.2%, (only B156 using IUP has a slightly higher amplitude of 0.4%). In contrast, large differences are observed in the offsets between the total column ozone calculated from these cross-sections for the Brewers and while the best agreement of 0.0% is seen with IUP, and -1.2% with IUP_ATMOZ. These results confirm the findings by Redondas et al. (2014), which also saw similar large discrepancies with DBM, and the best consistency with the IUP crosssections.
The new results with ACS and IUP_ATMOZ complement the previous study by Redondas et al. (2014), and support the fact 310 that the Dobson spectroradiometers are sensitive to effective ozone temperature and need a correction based on the temperature coefficients calculated from the ozone absorption cross-sections.
3.2.1 Total column ozone using the effective ozone temperature from ECMWF The availability of a global dataset of effective ozone temperature becomes crucial if the decision is made to reprocess the whole historical dataset of the global Dobson network. Such a global dataset is available through the afore-mentioned ECMWF effec-315 tive ozone temperature climatology, which is available daily since 1956 on a global scale. Without entering in a detailed discussion on this topic, we have evaluated how the re-processed Dobson total column ozone dataset using either the ozonesonde or ECMWF-based effective ozone temperature dataset agrees at Arosa/Davos. As mentioned previously, the values of effective ozone temperatures from ECMWF are systematically biased low, the differences increasing with decreasing temperatures by up to 4 K. Nevertheless, the total column ozone dataset reprocessed using the ECMWF temperatures is only 0.1% lower on 320 average than the ozonesonde based one, and the standard deviation of 0.2% between the two datasets is also extremely low.
Therefore at least for the northern mid-latitude site Arosa/Davos, either effective ozone temperature dataset can be used for the recalculation of the total column ozone from Dobson D101 with a bias of at most 0.1% and a resulting uncertainty of less than 0.2%. A possible influence of stray light on the total column measurements was investigated by correlating the relative differences between the double Brewers B156 and B163 to Dobson D101 against ozone slant pass. The data was filtered by using only effective ozone temperatures between 225±2 K. As can be seen in Fig. 7, the relative differences vary between about ±1%, with a slight curvature, indicating some small unexplained features between the instruments, which could be due to the actual extraterrestrial constants coming from the independent calibrations performed on Brewers and Dobson. However the instruments 330 do not deviate systematically at larger ozone slant pass, which would be an indication for a stray-light impact on the ozone retrieval. The same comparison using the two uncorrected single Brewers B040 and B072 (not shown), shows the impact of stray-light due to the single monochromator Brewers, which under-estimate ozone at large airmass due to the stray light coming from longer wavelengths.

335
The highest consistency between Brewer and Dobsons is obtained with the IUP ozone absorption cross-sections (Serdyuchenko et al. , 2013) and applying an effective ozone temperature correction. The effective ozone coefficient of the Brewer spectroradiometer, as already noted by Redondas et al. (2014), is only weakly sensitive to a change in effective ozone temperature (0.1% per 10 K, see Table 2) , while for the Dobson, the sensitivity to effective ozone temperature is considerably larger, with 0.9% per 10 K change.

340
Based on the results of the double Brewers B156 and B163, the lowest residual seasonal variability between Brewer and Dobson is obtained with the ACS or IUP_ATMOZ cross-sections, indicating that their temperature dependence are the most consistent of the investigated ozone absorption cross-section datasets. Unfortunately, the observed offsets of more than 1% indicate that there is some spectral inconsistency in the datasets which precludes their use as common ozone absorption crosssections for the Brewer and Dobson networks.

345
Due to this significant sensitivity to effective ozone temperature, the total column ozone calculation procedure for the Dobson spectroradiometer requires this additional ancillary information to produce total column ozone corrected for the seasonal variability observed at LKO Arosa and at other stations. This ancillary information could be obtained from ozone sonde launches as done in this study, or from climatological reanalyses such as from the ECMWF effective ozone temperature reanalysis (http://www.temis.nl/climate/efftemp/overpass.html).

350
As shown here, the use of measured spectral line spread functions to calculate the ozone absorption coefficient of the Dobson D101 improves the agreement with the colocated Brewer spectroradiometers. This well established procedure for Brewer spectroradiometers would also improve the general agreement within the Dobson network, since even the reference Dobsons used for the calibration of network Dobson instruments show relative differences in ozone absorption coefficients of up to 0.6% when the actually measured line spread functions are used instead of the nominal ones (Köhler et al. , 2018).

Implications for the total column ozone series of LKO Arosa/Davos
The results of this study are based on individual measurements by well maintained Brewers and Dobson spectroradiometers within the period 2016 to 2020 covering nearly 5 years of continuous measurements at airmasses between 1.1 and 3.9.
Following the decision made by the WMO Scientific Advisory Group on Ozone and UV (WMO SAG O3UV), the total column ozone time series of LKO Arosa/Davos, which are based on a series of Dobson spectroradiometers, will need to 360 be reprocessed using the IUP ozone absorption cross-sections and applying a correction for the effective ozone temperature following the methodology outlined here. Similarly, the total column ozone measurements from the Brewer spectroradiometers operating at LKO Arosa/Davos, need to be reprocessed, using the updated Rayleigh scattering cross-sections and the IUP ozone absorption cross-sections.
Once these datasets are reprocessed, both datasets from Brewer and Dobson can be merged to form a homogenised total col-365 umn ozone dataset for LKO Arosa/Davos. The resulting uncertainty of the merged dataset can be estimated by the comparison of total column ozone measurements obtained between the colocated Brewer and Dobson instruments during the investigated period. As shown in the bottom left panel of Fig. 6 and in the figures in the supplement, the comparison of coincident single measurements within a 5 minute window between Brewers B040, B072, B156, and B163 to Dobson D101 gives an excellent agreement with offsets smaller than 0.5% and standard deviations of 0.9% or smaller. For daily means, this variability decreases 370 to 0.7%, for monthly means to 0.45%, and for the yearly means over the investigated period the relative difference between the reprocessed datasets of Brewers and Dobson D101 reaches 0.3%.