Comparison of ground-based Brewer and FTIR total column O 3 monitoring techniques

We compare the currently most precise, groundbased total O3 measurement techniques: Brewer and FTIR. We give an overview of the similarities and the differences between the measurements and the retrieval approaches of both experiments. We compare coincident measurements performed at the Atmospheric Observatory of Iza ña from 2005 to 2007 and demonstrate that, if the properties of the instruments are well characterised, the scatter between both experiments is as small as 0.5%. This is in agreement with the theoretical predictions and confirms empirically that both techniques are able to monitor total O 3 amounts with a precision of better than 0.4%. However, we found systematic differences between both techniques of around 4.5%, which we think are mainly due to discrepancies between the applied UV and infrared spectroscopic parameters.


Introduction
Over the last 20 years significant progress was made concerning the precision and accuracy of atmospheric remote sensing measurements.This development is due to improvements in both instrumental setups and in retrieval algorithms.High precision measurements performed on a continuous basis are needed in order to detect potential atmospheric trends as soon as possible.A good example is the atmospheric ozone content.In the coming decades some kind of ozone recovery is expected, however, it is difficult to predict how, when, and to what extent it will occur (Weatherhead and Andersen, 2006).Continuous high precision measurements of Correspondence to: M. Schneider (matthias.schneider@imk.fzk.de)O 3 are very important in this context.They provide for improved atmospheric O 3 models by detecting trends between the models and the measurements as soon as possible.
Currently there are two continuous ground-based measurement techniques that claim to monitor total O 3 with a precision of 1-2 DU: FTIR and Brewer.Their precision and accuracy are theoretically estimated in Schneider and Hase (2008) and Redondas and Cede (2006), respectively.Ground-based measurements are an important component of the global atmospheric monitoring system, since they are decisive for the validation of satellite data.Only space-based measurements achieve a continuous global coverage, but their quality is coupled to the precision and accuracy of the groundbased long-term reference experiments.In particular socalled "super-sites" play an important role in this context.They concentrate different measurement techniques at the same site offering a variety of parameters to be used for the validation of satellite data.In addition, these super-sites are very important for the development and quality assurance of the different ground-based experiments.The possibility to inter-compare the different ground-based techniques under routine conditions allows for the validation of new developments and for a timely identification of potential instrumental drifts.In this work we compare total O 3 amounts obtained from a state-of-the-art FTIR observing system and from the European reference Brewers.The measurements are performed at the Izaña Observatory, Canary Island of Tenerife, Spain.Currently the Izaña FTIR system provides the most precise FTIR O 3 data world-wide, which is due to both advanced instrumental equipment and an optimised O 3 retrieval strategy (Schneider and Hase, 2008).Concerning the Brewer, Izaña is the Regional Brewer Calibration Centre for Europe (http://www.rbcc-e.org/) of WMO/GAW (World Meteorological Organisation/Global Atmospheric Watch), M. Schneider et al.: Comparison of ground-based Brewer and FTIR technique which guarantees highest quality standards.Both FTIR and Brewer activities are part of NDACC (Network for Detection of Atmospheric Composition Change; Kurylo (1991Kurylo ( , 2000)); http://www.ndacc.org/).Currently, such high quality and continuously performed FTIR and Brewer measurements only coincide at the Izaña Observatory.Therefore, it is a predestinated site for an inter-comparison study of both techniques.
In the following we list the main similarities and differences of the FTIR and Brewer technique, and discuss their respective advantages and disadvantages (Sect.2).In Sect. 1 we briefly introduce the site where the measurements used in this work were made.In Sects. 2 to 6 we compare operational data from two different Brewer spectrometers and the FTIR spectrometer.Finally, we summarise the reasons that make the Izaña data unique and conclude about the implications of our results (Sects.7 and 8).

FTIR versus Brewer technique
The principle of both techniques is the same: both the FTIR and the Brewer spectrometer detect direct solar light.The experimental differences are the spectral regions that are analysed and the spectral resolution of the measurements.The respective retrieval algorithms deduce the atmospheric O 3 amounts by analysing the absorption signatures imprinted onto the extraterrestrial solar spectrum.For this purpose the Brewer and FTIR technique use different retrieval approaches, which are imposed by the nature of the respective experimental data.

The experiments
The Brewer spectrometer detects spectral irradiance in six channels in the UV (303.2,306.3,310.1,313.5,316.8,and 320.1 nm) each covering a bandwidth of 0.5 nm (resolution power λ/ λ of around 600).The spectral analysis is achieved by a holographic grating in combination with a slitmask which selects the channel to be analysed by a photomultiplier.At Izaña only MK-III Brewer instruments with double monochromators are applied which widely reduces the impact of straylight on the measurements.The Brewer system works in a completely automatic way, and usually measures continuously during the whole day.The first channel at 303.2 nm is only used for spectral wavelength checks by means of internal Hg-lamps, the second channel is used for measuring SO 2 and the remaining four channels at longer wavelength determine the O 3 amounts.The finally reported O 3 result is the mean value of a set of 5 observations.The standard deviation within these 5 observations is used for the acceptance of the measurement, here we consider a measurement to be valid if this standard deviation is lower than 2.5 DU.The Brewer's field of view (FOV) is about 2.7 • (diameter of solar disc is 0.5 • ).Consequently, all direct solar irradiance is coupled into the spectrometer even for a moderate misalignment of the solar tracker.On the other hand, a certain fraction of the diffuse radiance (circumsolar) is measured together with the direct irradiance.This signal of the diffuse radiance increases with the amount of scattering, i.e. mainly with SZA and aerosols, and alters the retrieved O 3 amounts.
The Brewer experiment needs some instrumental characteristics to be determined by calibration experiments: the wavelength calibration and the slit function (instrumental line shape function), and the extraterrestrial constant (ETC).The slit function is determined once per year.This is done in a laboratory by means of low pressure discharge lamps (e.g.Fioletov et al., 2005;Gröbner et al., 1998).The exact wavelength settings are monitored in an automated mode every 40 min by means of the internal Hg-lamps.The slit function and the exact wavelength settings are necessary to convolve the highly resolved Bass and Paur (1985) O 3 absorption coefficients with the instrumental function.The ETC is the weighted sum of the logarithms of the intensity that would be detected by the different channels in the absence of any terrestrial atmosphere (according to Eq. 6 and text thereafter).The Izaña sky conditions allow the determination of the ETC for each Brewer independently by the Langley method.However, in this work we use the ETC transferred by the traveling world standard Brewer #017.This ETC calibration is done once per year.The traveling world standard is applied to guarantee a world-wide consistency within the Brewer network.The stability of the ETC calibration is monitored continuously in an automated mode using an internal halogen lamp.The difference between the weighted counts of the halogen lamp at the time of calibration and the smoothed mean of two weeks is used to correct the ETC (Fioletov et al., 2005).
The mid-infrared FTIR measurement program covers the spectral region from 700 to 4200 cm −1 (corresponding to 14.3 to 2.4 µm), excluding the region between 1400 and 1700 cm −1 , which is not useful for ground-based measurements due to strong water vapour absorption.The spectral range is split up in 6 filter regions.For the longer wavelength region we use a band-pass filter for the 700-1400 cm −1 region.The spectra are obtained by a Fourier analysis of the recorded interferograms.The spectral resolution depends on the maximum optical path difference between the interfering light beams.For operational O 3 measurements the FTIR spectra are measured with a resolution of 0.005 cm −1 , which corresponds to a resolution power λ/ λ at 1000 cm −1 of 2×10 5 .The spectral windows applied for the O 3 retrieval are similar to those of (Schneider and Hase, 2008).They consist of a total of 3600 spectral bins and contain many individual O 3 rotational-vibrational lines (at 991.25-993.80 cm −1 , 1001.47-1003.04cm −1 , and 1005.00-1006.91cm −1 ) and 4 CO 2 lines (at 962.80-963.80 cm −1 , 964.24-965.26cm −1 , 967.20-968.21cm −1 , and 968.59-969.60 cm −1 ) of different intensity.The CO 2 signatures allow the retrieval of a temperature profile which is expected to widely improve the precision of the retrieved total O 3 amount (Schneider and Hase, 2008).It is important to mention that all these individual rotational-vibrational lines are resolved: in the middle infrared the shape of absorption lines is dominated by pressure broadening with a typical HWHM of 0.04 cm −1 at surface level, which is nearly one order of magnitude larger than the operational resolution of the FTIR spectrometer.The numerous fine-structured spectral features provide for an automatic wavelength calibration.The FTIR spectrometer has a field of view of only 0.2 • , i.e. it only analyses sunlight coming from the center of the solar disc (diameter of 0.5 • ).Consequently, a misalignment of the solar tracker directly affects the observing geometry.In Schneider and Hase (2008) it is shown that for elevation angles below 20 • this is the leading error source.Although on a much higher level of spectral resolution if compared to the Brewer, it is important to monitor the instrumental line shape (ILS) of the FTIR spectrometer.This is done regularly by performing low pressure cell calibration measurements as described in Hase et al. (1999).Even then the residual ILS error is estimated to be the most important error source concerning FTIR total O 3 amounts measured at solar elevation angles above 20 • .

The retrieval approaches
The basic equation for analysing solar absorption spectra is Lambert Beer's law: Here I (λ) is the measured intensity at wavelength λ, I ET the extraterrestrial intensity, τ O 3 the optical depth due to O 3 , and τ x the optical depth due to all other atmospheric components (other trace gases, aerosols etc.).For O 3 it is: whereby σ O 3 (λ, s) is the absorption cross section and n O 3 (s) the concentration of O 3 at location s.The integration is performed along the path of the direct sunlight.The cross section σ O 3 depends on temperature and in the infrared additionally on pressure.
The integration of n O 3 perpendicularly throughout the atmosphere gives the total O 3 column amount ( O 3 ): To derive O 3 from the measurements both techniques apply different approaches.

Principles of the Brewer retrieval
The Brewer algorithm considers the absorption by O 3 and SO 2 , scattering by molecules, and extinction by aerosols.
The algorithm applies an airmass factor (µ x ), according to: as the ratio between the slant and the vertical total column amount.Here R, , and h x are the Earth's radius, the apparent solar zenith angle (90 • -elevation angle), and the effective altitude of the absorbing or scattering component (h=22 km for O 3 , SO 2 , and aerosols, and h=5 km for Rayleigh scattering).Equation ( 4) is a simplification of the real situation.It assumes that the absorbing compounds are concentrated at a single altitude h x and it disregards that the refraction index depends on altitude.The errors produced by these assumptions are important for low solar elevation angles (for an elevation angle of 10 • this error is 2-3% Bernhard et al., 2005).
Taking the logarithm of Eq. ( 1) and applying Eqs. ( 2) and (3) yields the following relation between the intensities at channel i (I i ) and the amount of the extinction components: The four channels at the longer wavelength are combined (Evans et al., 1987), ) and the weighting coefficients w i are selected to minimise the influence of SO 2 : w [1;4] =[1.0, −0.5, −2.2, 1.7].This choice also widely eliminates absorption features which depend in local approximation linearly on wavelength (λ) like Rayleigh scattering and aerosol extinction, since 4 i=1 w i λ i ≈ 4 i=1 w i =0.In Eq. ( 6) we replaced 4 i=1 w i log I ET,i , the extraterrestrial coefficient, by ETC.With the wavelength and slit function calibration we can calculate the convolved extinction coefficients (σ x of Eq. 6).The ETC is transferred from a reference instrument (Fioletov et al., 2005).Changes with time in the sensitivity of the instrument are reflected in changes of this extraterrestrial coefficient.Equation ( 6) together with Eq.(4) provides the total vertical O 3 amount.For the derivation of Eq. ( 5) we have to assume a constant so-called effective σ x throughout the atmosphere.Any temperature or pressure dependence is neglected.The "operational" algorithm applies a σ O 3 corresponding to an effective height of O 3 of 22 km and a fixed effective temperature of the O 3 layer of −45 • C.These simplifications produce systematic and random errors.In the case of the Izaña Observatory the effective height is about the same as the one used by the operational algorithm, but the effective temperature ranges from −50 • C in winter months to −45 • C in summer months.With the parameterisation of Van Roozendael et al. (1998) we expect a typical error of −0.4% during the summer months due to these simplifications.

Principles of the FTIR retrieval
The FTIR retrieval (PROFFIT, Hase et al., 2004) applies a precise radiative transfer model (KOPRA, Höpfner et al., 1998;Kuntz et al., 1998;Stiller et al., 1998).KOPRA is a line-by-line model which simulates the measured spectra.It includes a ray tracing module (Hase and Höpfner, 1999) to simulate how the solar light passes through the atmosphere.The model uses a discretised atmosphere (here we apply 41 levels between the Earth surface and the top of the atmosphere).The optical depth for each layer, enclosed between a pair of adjacent levels, is calculated by performing the integration of Eq. ( 2) between two levels.The applied absorption cross section σ for each individual line and level are parameterised according to the HITRAN spectroscopic database (Rothman et al., 2005).The parameterisation takes care of the pressure and temperature dependency of σ , i.e. it depends on the pressure and temperature actually present at the corresponding level.Summing up the τ values of the different layers leads to the simulated spectra at the location of the observer (combination of Eqs. 1 and 2).The radiative transfer model determines the changes in the spectral fluxes y for a changing state vector x.These derivatives are sampled in a Jacobian matrix K: Inverting K of Eq. ( 7) would allow an iterative calculation of the atmospheric state from the measurement alone.However, generally the problem is under-determined, i.e. the columns of K are not linearly independent.To overcome this problem an optimal estimation (OE) approach is applied (Rodgers, 2000): since the actual atmospheric state cannot be determined unambiguously from the measurement the OE approach determines the most probable state for the given measurement.The approach is based on the Bayesian theorem and consists in maximizing a total probability density function (pdf).The total pdf is the product of two pdfs: a first, describing the probabilities of the residuals of the spectral fit, and a second, describing the a-priori known probabilities of the absorbers' distributions.
This approach produces vertical O 3 concentration profiles (n O 3 (z)) for several O 3 isotopologues ( 48 O 3 , asymmetric and symmetric 50 O 3 and 49 O 3 ) individually , since each isotopologue offers distinct absorption lines.The total column amounts are subsequently calculated by Eq. ( 3).In addition, the measured spectrum contains sufficient information to determine the main issues of the actual temperature profile.Actually the O 3 retrieval consists of a joint OE of O 3 , 50 O 3 / 48 O 3 , interfering species, and temperature profiles.Further details about the O 3 FTIR retrieval and its error budget are given in Schneider and Hase (2008).In the following sections we always refer to the sum of all O 3 isotopologues when we talk about the FTIR O 3 amounts.
Applying highly-resolved infrared spectra it is easy to separate the extraterrestrial component from the fine-structured absorption signatures of atmospheric trace gases.In the infrared the solar radiation is reasonably close to a black body radiation at 6000 K, i.e. it has a rather smooth dependence on wavelength.The presence of the few solar lines causes no significant problem, since they are well known (e.g.Hase et al., 2006) and fully resolved by the FTIR spectrometer.
The spectral windows applied for the FTIR retrieval contain many spectral bins between the fine-structured absorptions of atmospheric trace gases to construct an empirical background continuum.

Summary
Table 1 collects the principle differences of both techniques and resumes their respective advantages.The Brewer technique depends on several calibration experiments: (a) the wavelength setting and slit function calibration.The convolved absorption cross sections strongly depend on this calibrations.(b) The transfer of the extraterrestrial constant (ETC) from the world standard Brewer.The ETC is necessary to separate the extraterrestrial from the atmospheric signal.Errors in the ETC add directly to the slant column amounts (see Eq. 6), i.e. they are especially important for low slant column amounts.On the other hand, every single FTIR spectrum is self-calibrating with respect to the wavelength and intensity: (a) the FTIR spectrometer fully resolves many rotational-vibrational lines, it provides for a very accurate and automatic wavelength calibration of the measured spectra.(b) the extraterrestrial spectrum can be easily simulated and there is sufficient information to derive the overall instrumental transmittance from every single measurement.Nevertheless, for high precision measurements of absorbers with sharp signatures (stratospheric absorbers), it is important to monitor the actual ILS of the FTIR instrument continuously.For elevation angles above 20 • the remaining ILS uncertainties are the leading error source.
The Brewer algorithm applies the same σ O 3 throughout the year and throughout the whole atmosphere.It, furthermore, assumes a O 3 profile where all O 3 is concentrated in a single layer at 22 km.The FTIR retrieval, on the other hand, takes into account the actual O 3 and temperature profile and applies different σ O 3 s and temperatures for 41 different atmospheric levels.This is important, since σ O 3 depends on temperature and pressure.Furthermore, the FTIR retrieval uses a comprehensive ray tracing model.
The field-of-view of the Brewer instrument is rather large so perfect tracking is less important.However, it simultaneously analyses a significant amount of diffuse light, which is an important error source for low elevation angles.The fieldof-view of the FTIR is smaller than the solar disc.For low elevation angles (below 20 • ) the FTIR data quality depends critically on a perfectly working solar tracker.
A great advantage of the Brewer technique is that the measurements are performed with a very compact and mobile instrument and that all measurements including the calibrations are made nearly automatically.The measurements are performed continuously at many sites over the globe.For these reasons Brewer O 3 data are often used as reference in inter-comparison studies between different instruments and techniques, including space-based instruments.FTIR spectrometers only have limited mobility.They are generally installed inside a laboratory or a big shipping container.On the other hand, they are very versatile instruments measuring a great variety of atmospheric species.Concerning O 3 , the FTIR measurements allow the different isotopologues to be distinguished.FTIR measurements at Izaña are typically performed three times per week.

The Iza ña super-site
The Brewer and FTIR measurements are performed at the Izaña Observatory, which is located on the Canary Island of Tenerife, 300 km from the African west coast at 28  Kurylo (1991Kurylo ( , 2000)); http://www.ndacc.org/).
The Brewer measurements at Izaña started in May 1991.Since November 2003 they represent the Regional Brewer Calibration Centre for Europe (http://www.rbcc-e.org/) of WMO/GAW (World Meteorological Organisation/Global Atmospheric Watch).This Calibration Centre is essential for a coordinated European Brewer network that is needed for both present and future consistency of ground-based total ozone observations and for validation of satellite instruments.Furthermore, it plays an important role in the development and testing of new measurement techniques for the whole Brewer network.The Brewer data are often used as a reference for validating other ground-and satellite-based instruments.
The  high resolution FTIR spectrometer commercially available.
Compared to the IFS 120M it has (a) a more stable instrumental line shape and (b) a 30% higher signal to noise ratio.
In Schneider and Hase (2008) it is shown that a stable ILS is an important requisite to reach total O 3 precision of better than 1-2%.To guarantee highest quality of our FTIR O 3 products in this work, we only use O 3 amounts inverted from IFS 125HR measurements.At the Izaña Observatory clean air and clear sky conditions are prevailing around all the year.Firstly, it is located in the region below the descending branch of the Hadley cell, typically above a stable inversion layer.Secondly, it is situated on an island far away from any significant industrial activities.Consequently it offers excellent conditions for atmospheric observations by remote sensing techniques and it is predestinated for calibration and validation activities.Due to its geographic location it is in particular valuable for the investigation of stratosphere-troposphere exchange associated with the subtropical jet (e.g.Kowol-Santen et al., 1999;Cuevas et al., 2007) and large scale transport from the tropics to higher latitudes.
Figure 1 shows the evolution of the total O 3 amounts in 2005 and 2006 as measured by the site standard Brewer #157,

Brewer #157 versus Brewer #185
In this section we compare total O 3 data from the site standard Brewer #157 and the traveling standard Brewer #185.Between January 2005 and February 2007 there are 4300 measurements performed simultaneously by both instruments.The correlation of these O 3 data is shown in Fig. 2. We find a correlation coefficient ρ of 0.998, a regression line slope of 0.996, and an offset of the regression line of 1.5 DU.The scattering around this regression line provides an estimate of the precision of the Brewer instrument with respect to O 3 amounts: The difference between both Brewer instruments is 0.2±0.4%.If we assume that the leading error sources of the O 3 amounts obtained from the two Brewer instruments are independent, we can estimate the precision of one Brewer instrument to 0.4/ √ 2=0.3%.However, it is likely that some error sources of two experiments which apply the same technique are correlated.Consequently the precision of 0.3% is only a best case estimation.for all coinciding measurements; bottom panels: for measurements with low slant columns (<400 DU).The left panels show all individual coincidences and the right panels the statistics of three month averages: the black shaded area covers the range within which the mean value is situated with a probability of 95%; the grey-shaded area indicates the standard deviation.The scale of the y-axis of the right panels is expanded by a factor 2. The left panels show all individual coincidences and the right averages over values within a radius of 12.5% of the slant column amount: the black shaded area covers the range within which the mean value is situated with a probability of 95%; the grey-shaded area indicates the standard deviation.The scale of the y-axis of the right panels is expanded by a factor 2.

Temporal evolution
In the following we investigate how the difference between both Brewers evolves with time.The left panels of Fig. 3 depict the difference for all individual coincidences versus time.When contemplating this plot it is important to note the scale on the y-axis.It is given in percent and consequently we are looking on temporal variations in the range of a few permil.Since the noise in the data is also situated in the permil range we can only expect to make some useful observations by contemplating averages over a certain amount of data.Here we average data over 3 months in the sense that www.atmos-chem-phys.net/8/5535/2008/Atmos.Chem.Phys., 8, 5535-5550, 2008 Fig. 5. Correlation between total column amounts for different FTIR approaches (a first similar to Barret et al. (2002) and a second according to Schneider and Hase, 2008).Black circles are the individual measurements, the red line is the linear regression lines of the least squares fit.
we apply data within a radius of 1.5 month around the considered date.The statistics of such averaged data are shown in the right panels.The grey bars represent the standard deviation (std) of the 3-month ensembles.They can be interpreted as overall precision of Brewer #157 and #185 within a three month period.The black bars represent the 95% confidence interval.This interval represents the area where the real mean value of a 3-month ensemble is situated with a probability of 95%.The radius of this area is approximately 2× std √ n (n is the number of ensemble members).Often std √ n (the standard error of the mean) is used to determine the confidence radius.However, a so-calculated area would only contain the real mean with a probability of 68%.The upper panels show the situation if all 4300 coinciding Brewer measurements are applied.In this case even for the three month averages there is no significant temporal variation of the difference between the two Brewers.It seems that both instruments produce very consistent data even over several years.A small difference between 2005 and 2006 is that the overall precision of Brewer #157 and #185 for 3 month periods is slightly better in 2006 compared to 2005 (0.4% compared to 0.5%; see grey bars).
In addition we perform a separate analysis for low slant column amounts.This is done for two reasons: (1) FTIR measurements are generally performed at relatively low slant column amounts (75% of all FTIR measurements are at solar elevation angles above 35 • ).On the other hand, Brewer measurements are performed during the whole day, and thus, Brewer data include many O 3 amounts deduced from high slant column amounts.The separate analysis of low slant column assures that the Brewer data are characterised under similar condition as the FTIR data.(2) Systematic ETC errors are amplified for low slant column amounts, since a systematic error in the ETC produces a systematic bias in the retrieved slant column amounts (see Eq. 6).The bottom panels of Fig. 3 shows data only if the respective measurement was made for a slant column lower than 400 DU.This slant column amount corresponds to an solar elevation angle of typically above 50 • .Consequently, even at the subtropical site of Izaña, these data are not available in winter and the time series is limited to data from February to November.In these graphs we can observe a slight difference between 2005 and 2006.In 2005 both Brewers measure nearly the same, whereas in 2006 the site standard Brewer #157 produces O 3 amounts, which are around 0.5% lower than the Brewer #185 amounts.This observation is robust since it is based on more than 700 coincidences between the Brewers.The areas where the mean difference is situated with a probability of 95% (indicated by black bars in the left panels of

Dependence on slant column amounts
In the following we analyse in more detail the differences between the years 2005 and 2006.For this, we have a closer look at the dependence of the observed differences on the present slant column amounts.Figure 4 shows this dependence for the two periods: April 2005 to November 2005 and December 2005 to January 2007.The left panels show the individual data points.The right panels show statistics of the averages.Here the averages are calculated by including all data situated within a radius of 12.5% of the slant column amount, i.e. for a slant column amount of 500 DU we calculated the average of all data between 437.5 and 562.5 DU and for a slant column amount of 1000 DU we apply data between 875 and 1125 DU.The upper panels show the situation for the first period (1100 coincidences).There is a weak dependence of the difference between both Brewers on the slant column amount.For very low slant column amounts the Brewer #157 produces systematically larger amounts than the Brewer #185 (up to 0.5%) and for slant columns above 350 DU the sign changes and the #157 values are smaller than the #185 values.The lower panels show the situation for the year 2006 (period between December 2005 and January 2007; 3200 coincidences).It differs from the year 2005.Now for low slant column amounts the Brewer #157 data are 0.4% lower than the #185 amounts, and for slant columns above 600 DU both Brewer agree nearly perfectly.Furthermore, we found that in 2006 the data are less noisy than in 2005.These plots show that the inconsistencies between the Brewers with respect to 2005 and 2006 are reflected in a different dependence on the slant column amounts.Such a dependence indicates that there are weak inconsistencies in the ETCs applied for the Brewers (see Eq. 6).The dependence on the slant column amount is in particular large for 2005, indicating larger inconsistencies between both Brewers than in 2006.(Barret et al., 2002) versus FTIR (Schneider and Hase, 2008) In this Section we compare total O 3 data from two different FTIR retrieval approaches: first, using a method similar to Barret et al. (2002) (in the following called BA02) and second, similar to Schneider and Hase (2008) (in the following called SH08).By this means we perform an internal consistency check of the FTIR data.Figure 5 shows the correlation between the two approaches.Both retrievals apply the same O 3 absorption signatures but slightly different retrieval strategies.The first consists in an optimal estimation of O 3 profiles alone and the second in an joint optimal estimation of O 3 , 50 O 3 / 48 O 3 , and temperature profiles.A correlation coefficient of "only" 0.984 is relatively low.The agreement is poorer if compared to the agreement between both Brewer instruments.This is mainly due to errors from the BA02 method.In Schneider and Hase (2008) it was shown that the SH08 approach provides for significantly more precise data than the BA02 approach.It nearly eliminates the error due to uncertainties in the applied temperature profiles, which are the dominant error source of the BA02 approach.

FTIR
On the other hand, the SH08 approach is more sensitive to errors due to ILS uncertainties (Schneider and Hase, 2008).By comparing the data of both approaches we expect to get some information about the stability of the ILS from 2005 to 2007.

Temporal evolution
The temporal evolution of the difference between the two FTIR retrieval approaches is depicted in Fig. 6.Like for Fig. 3 the left panels show the individual measurements and the right panels show the statistics for the 3-month averages: the black bars represent the range of the mean values at a 95% confidence level, the grey bars represent the standard deviation.The upper panels show all data and the bottom panels only data for slant column amounts below 400 DU.First of all there is much more noise on the data when compared to Fig. 3.The relatively large noise is mainly due to temperature errors present in the BA02 approach.Again it should be remarked that the BA02 approach does not fully exploit the potential of the FTIR measurements.It is not an optimised approach, but we can exploit its different ILS sensitivity to perform an internal estimation of the FTIR's ILS characterisation.This internal estimation is achieved by averaging over a certain amount of data.This process significantly reduces the temperature error, which is mainly a random error.The remaining systematic signal should then be due to a systematic error source, for which both retrieval approaches have different sensitivities: e.g. the ILS error.The statistics of 3-month averages are depicted in the right panels of Fig. 6.We observe some systematic differences be- There are several possible explications for the relatively poor instrumental stability in 2005.In January 2005, after the installation of the instrument, the ILS was very well characterised (difference between BA02 and SH08 data of 0%).However, in the following there was a sequence of experimental complications.The setting of the ground or the whole surface below the container may have still been in progress during the first months of the measurements.In February 2005 there was a snow storm in Izaña and a subsequent breakdown of the power supply for several days.As a consequence the temperature in the container decreased to around 0 C • .These mechanical and thermal stresses may have produced a degradation of the optical alignment.Furthermore, the low temperature damaged the hygroscopic entrance windows which had to be replaced.Unfortunately we only performed two independent ILS calibration measurements between January and November 2005.This is definitively not sufficient for a high quality characterisation of the ILS, in particular if we consider the aforementioned adverse conditions.In November 2005 we installed new firmware and reinstalled the detectors for technical servicing.The latter significantly reduced a channeling of 0.58 cm −1 .Since December 2005 the instrument is operating continuously and there were no further modifications necessary.In particular there was no further significant temperature breakdown inside the measurement container.Since the end of 2005 it is kept continuously at 22 C • .

Dependence on slant column amounts
The ILS error of the FTIR O 3 data slightly depends on the O 3 slant column amount (Schneider and Hase, 2008).The error is larger for low slant column amounts than for high slant column amounts.Since the BA02 and SH08 approach have different ILS error sensitivities, this dependence should be observable in the difference of BA02 and SH08 O 3 amounts.Figure 7 depicts the dependence of the difference between BA02 and SH08 O 3 on the slant column amount.In 2005 (upper panels) we observe a clear increase of the negative difference for low slant column amounts, whereas in 2006 this increase is much smaller.Thus, Fig. 7 confirms our conclusions drawn from Fig. 6: in 2005 the FTIR instrument is less well characterised than in 2006.Fig. 6.Time series for the differences between the different FTIR approaches (FTIR (BA02)-FTIR (SH08))/FTIR (SH08)).Content of panels, symbols, and colours is the same as in Fig. 3.

Brewer versus FTIR
In this section we compare total O 3 measurements of Brewer and FTIR.To exclude influences due to temporal variabilities we require that both Brewer and FTIR measurements should coincide within 30 min.Between January 2005 and February 2007 we identified a total of 305 FTIR measurements which fulfill these coincidence criteria with the Brewer measurements: 240 for the Brewer #157, 165 for the Brewer #185.Both Brewer instruments measure generally during the whole day.Figure 8 depicts the correlation between the site standard Brewer #157 and the FTIR O 3 amounts.Within all 240 coincidences we found one day which clearly is an outlier (2 March 2006; marked by blue circle in both panels of Fig. 8).So far we have no explication for this outlier.We do not consider this data point for the following analysis since it is not representative but would widely influence the results.The left panel of Fig. 8 shows the correlation between the Brewer data and the FTIR data obtained from the BA02 approach.The agreement between the Brewer and these FTIR data is quite good: a correlation coefficient of 0.982 and a difference of 4.0±1.1%.This agreement is much better than reported by other studies (e.g.Barret et al., 2002;Schneider et al., 2005).It demonstrates the high quality of the Brewer and FTIR measurements performed at Izaña.In particular, it confirms the potential of the FTIR technique when the instrumental aspects of the recipe as presented in Schneider and Hase (2008) are considered.
However, there is still a large potential to further improve the agreement.As shown in Schneider and Hase (2008) the BA02 error budget is dominated by errors in the assumed temperature profiles and applying the SH08 approach should widely reduce the overall FTIR errors.The correlation of the FTIR SH08 data with the Brewer #157 data is shown in the right panel of Fig. 8.It yields a correlation coefficient of 0.992.The difference between both instruments is 4.9±0.7%.The SH08 data agree significantly better with the Brewer data than the BA02 data.If we only correlate Brewer and FTIR SH08 data measured after November 2005 we get an even larger improvement.Then the correlation coefficient for the SH08 approach is 0.996 and the difference −4.7±0.5%,whereas the BA02 approach only yields a correlation coefficient of 0.984 and a difference of −4.0±1.0%(Fig. 9).Limiting the comparison to measurements taken after November 2005 is justified since in 2005 the FTIR instrument is not optimally characterised (see Figs. 6 and 7 and discussion in Sect.5).
The data measured after November 2005 reveal the real potential of the FTIR technique.Then the scatter between the FTIR (SH08) and the Brewer data is only 0.5%.This is an excellent value for two independent remote sensing experiments performed over more than 1 year.It can be interpreted as the root-square-sum of the precisions of the Brewer and FTIR instrument.As aforementioned the Brewer precision is not better than 0.3%.Consequently the Brewer/FTIR comparison is the empirical proof that the ground-based FTIR technique has the potential to measure O 3 amounts with a precision of better than 0.4%.For the BA02 approach the agreement does not improve significantly if we limit the comparison to data measured after November 2005.These data are dominated by errors caused by the assumed temperature profiles, which are much larger than the errors caused by inconsistencies in the instrument's performance.
The Brewer-FTIR comparison confirms the theoretical prediction of Schneider and Hase (2008).However, there are significant systematic differences between both datasets: the FTIR observing system measures systematically 4.7% more O 3 than the Brewer system.These systematic differences may be due to incorrect characterisations of the instruments, simplifications in the Brewer retrieval algorithm, or discrepancies between the applied infrared (Rothman et al., 2005) and UV spectroscopic parameters (Bass and Paur, 1985).
Figure 10 shows the temporal evolution of differences between Brewer #157 and the FTIR data obtained by the SH08 approach.The upper panels show the data from all 240 coinciding measurements.Here the 95% confidence ranges (black bars in right panels) are much broader than in Fig. 3 due to a much smaller number of individual measurements applied for averaging.We observe clear differences between 2005 and 2006.In March 2005 FTIR O 3 amounts are typically 5.7% higher than Brewer #157 amounts, while in 2006 these differences are typically 4.6%.We observe a continuous increase of the Brewer #157 O 3 with respect to the FTIR O 3 between March and November 2005.The stability of the agreement since the end of 2005 until February 2007 is remarkable.Both techniques agree within 0.5% during more than 14 months.This excellent agreement is also achieved in 2005 if we consider only 3-month periods (see grey bars in right panels).Only in January and October/November 2005 it is larger.In both situations the ensembles include data for different FTIR and Brewer instrumental characteristics: the internal FTIR check (Fig. 6) shows steps in February and November 2005 and the internal Brewer check (Fig. 3) shows a step in December 2005.If we limit to slant column amounts below 400 DU (bottom panels) we make the same observations: within 3-month periods the data agree always within 0.5% and starting with January 2006 the agreement keeps stable over more than one year.Concerning the differences between 2005 and 2006 the continuous increase during 2005 is even more pronounced if compared to the upper panels (from −6.0% to −4.8%).summer 2006 the difference is temporarily more negative than April or October of the same year.This feature may be produced the Brewer algorithm's assumption of a fixed effective O 3 temperature (see end of Sect.2.2.1).
6.1 Temporal evolution 10 and 11 have many similarities but also two differences.Firstly, in 2005 the systematic overestimation of the Brewer O 3 amounts by FTIR is smaller for the Brewer #185 than for the Brewer #157.A reason could be the differences in 2005 as observed #185-FTIR are mainly due to inconsistencies in the FTIR data and the differences observed in #157-FTIR are due both inconsistencies in the FTIR and the Brewer #157 data.
the #185-FTIR data are noisier than the #157-FTIR data.An elevated in the Brewer #185 data was already observed during the ETC calibration in October 2005 (K.Lamb, private communication) and is due to the internal filters applied by the Brewer instrument.There are five grey filters assuring that an adequate light intensity enters the photomultiplier (filter #1 is weakly attenuating and filter #5 is strongly attenuating).The problem is that these filters are not ideal grey filters, i.e. their attenuation depends weakly on wavelength.This not ideal attenuation is different for each filter and causes a filter dependent error in the retrieved O 3 amounts.The Brewer #185 is more sensitive than the Brewer #157.Even at large solar elevation angles the filter #3 is sufficiently attenuating for the Brewer #157, whereby the Brewer #185 frequently needs to switch between filter #3 and #4, which is the reason for the elevated noise in the Brewer #185 data.It is possible to reduce this noise by correcting the error caused by the not ideal grey filters.This correction is not part of the standard retrieval algorithm, and thus not applied for the here presented data.The correction would also reduce the systematic difference of (Brewer-FTIR)/FTIR by 0.5% (to around −4% in 2006/2007).

Dependence on slant column amounts
Figure 12 shows the dependence of the difference between the FTIR and Brewer #157 on the slant column amounts.Figure 13 shows the same as Fig. 12 but for the difference between FTIR and Brewer #185.We also found that dependence on slant column amounts is larger in 2005 if compared to 2006.

Some remarks on the quality of Iza ña's O 3 measurements
Many other observatories could reach a similar precision as Izaña, but there is, to our knowledge, currently world-wide no other site where total O 3 column amounts are measured simultaneously by different ground-based techniques with the same high precision as presented in this work.The reasons for the currently unique high quality of the Izaña O 3 data are manifold: (a) the outstanding meteorological conditions that prevail at the Izaña Observatory benefit the quality of the FTIR measurements.There are only negligible intensity fluctuations while recording an FTIR interferogram.At less favorable measurement sites these fluctuations are larger due to changes of the atmospheric transparency (caused by clouds, short-scale inhomogeneities in water vapour, tropospheric aerosols, contrails etc.).These fluctuations are generally neglected, which produces errors in the retrieved O 3 amounts.To achieve an excellent FTIR data quality at less favorable sites the impact of the intensity fluctuations has to be reduced.Keppel-Aleks et al. (2007) describes a method to do so.(b) The application of state-of-the-art instrumentation.The Brewer triad is formed by three state-of-the-art instruments with double monochromator.Intercomparison studies with single monochromator instruments indicate a significantly poorer Brewer precision (Fioletov et al., 2005).The FTIR system consist of a very precise solar tracker (Huster, 1998), a Bruker IFS 125HR (stable ILS), and applies photo-voltaic detectors.These are the most important instrumental requirements needed for high precision FTIR measurements.Instabilities in the ILS are one of the dominant remaining error sources.These instabilities are responsible for the internal FTIR inconsistencies observed during 2005.(c) The high precision of both Brewer and FTIR depends on regularly performed calibration measurements.In case of the Brewer, the ETC and wavelength are monitored internally by calibration lamps several times per day.Once per year the ETC is transferred from the traveling world standard Brewer.The slit function remains very stable.It is measured every year.In case of the FTIR, regularly performed ILS calibration measurements (Hase et al., 1999) are important.Since November 2005 these measurements are performed every 2-3 months.(d) Finally, it is important to apply an optimised retrieval algorithm: the FTIR only reaches a precision better than 0.5% if the retrieval strategy as proposed by Schneider and Hase (2008) is applied.

Summary and conclusions
The operational Brewer data presented here are very consistent over more than two years.Although, in 2005 the ETC determined for the Brewer #157 seems to be slightly less accurate than in 2006.However, these errors very small and are in agreement with the expected ETC uncertainties.Stateof-the-art Brewer instruments (double combine high precision, and mobility.They are perfectly for calibrating O 3 instruments at different sites.Furthermore, Brewer spectrometers are relatively economic measure O 3 throughout the day and during the whole year. The FTIR presented here be classified in two groups.A first group consisting of the data measured in 2005, when the instrument was not optimally characterised, and a second group consisting the data measured after November 2005.The internal consistency check of Sect. 5 reveals the problems in 2005.The strong slant dependence the difference between the two FTIR retrieval approaches (Fig. 7) indicates the optimal characterisation of the FTIR instrument.In the FTIR data do not represent the real potential of the FTIR But even then we found a very small scatter with respect to the Brewer data of 0.8%.Since December 2005 FTIR inis very well characterised.Then the scatter with respect to the Brewer data is only 0.5%, which confirms the prediction of Schneider and Hase (2008).This the theoretical quality assessments of other FTIR products, for which no direct empirical verification can be achieved.We think that our results are of particular interest in the context of ground-based high precision measurements of greenhouse gases, which are performed within TCCON (Washenfelder et al., 2006) and will serve for the validation of OCO (Crisp et al., 2004).TCCON aims for a precision of 0.1% in the O 2 corrected CO 2 column.
Our study empirically derives a precision of Izaña's Brewer and FTIR data of around 0.4%.The high precision provides unique opportunities for investigating systematic differences of both techniques and for studying atmospheric O 3 trends.
We find a significant systematic difference between the FTIR and Brewer data of 4-5%.Concerning the FTIR, a wrong ILS characterisation could produce a systematic error.This is the case in 2005, when an incorrect ILS produces an overestimation of the FTIR O 3 amounts of up to 1%.Concerning the Brewer, uncertainties in the ETC and slit function calibration or not ideal filters introduce systematic errors.An incorrect Brewer ETC assumptions as well as a wrong FTIR ILS characterisation would depend on the observed slant column amount.Such a dependence is clearly observed in 2005 but only very small in 2006 and 2007.Consequently, we conclude that the 2006/2007 data are less affected by systematic errors of instrumental nature.However, 0.5% of the difference of 4.5% as observed in 2006/2007 is caused by the non ideal internal Brewer filters (see end of Sect.6.1).We think that in 2006/2007 in the worst case 1% of the observed systematic difference is due to errors in the assumed ILS, ETC, or non ideal filters.The rest is caused by the spectroscopic parameters.Our work indicates a systematic inconsistency between the infrared and UV spectroscopic coefficients of (4±1)%.This observation is in good agreement to the laboratory study Picquet-Varrault et al. (2005).
For the study of the expected O 3 recovery, measurements maintain a high precision over long time scales are needed.But even for regularly and carefully calibrated instruments small instrumental drifts may occur.In this context super-sites like Izaña are very important.The combination of different high quality experiments (like Brewer and FTIR) at a single site makes both experiments much more valuable for trend studies as if they were performed individually at different sites: the possibility to continuously inter-compare both techniques significantly reduces the risk of detecting artificial trends that may be caused by small instrumental drifts.

Fig. 1 .
Fig. 1.Time series of total O 3 as measured by the site standard Brewer #157, the traveling standard Brewer #185, and the FTIR instrument.

Fig. 2 .
Fig. 2. Correlation between total column amounts of Brewer #157 and Brewer #185.Black circles are individual measurements, red line represents linear regression line of least squares fit.

Fig. 3 .
Fig.3.Time series for the differences of total O 3 measured by the Brewer #157 and the Brewer #185((#185-#157)/#157).Upper panels: for all coinciding measurements; bottom panels: for measurements with low slant columns (<400 DU).The left panels show all individual coincidences and the right panels the statistics of three month averages: the black shaded area covers the range within which the mean value is situated with a probability of 95%; the grey-shaded area indicates the standard deviation.The scale of the y-axis of the right panels is expanded by a factor 2.

Fig. 4 .
Fig. 4. Differences of total O 3 measured by the Brewer #157 and the Brewer #185 ((#185-#157)/#157) versus slant column amount.Upper panel: for measurements between April and November 2005; bottom panel: for measurements between December 2005 and January 2007.The left panels show all individual coincidences and the right averages over values within a radius of 12.5% of the slant column amount: the black shaded area covers the range within which the mean value is situated with a probability of 95%; the grey-shaded area indicates the standard deviation.The scale of the y-axis of the right panels is expanded by a factor 2.

Fig. 3 )
are separated for the years 2005 and 2006: we observe a slight systematic difference in the Brewers' performance between 2005 and 2006.

Fig. 8 .
Fig. 8. Correlation between total column amounts of Brewer #157 and FTIR.Black circles are individual measurements, red lines linear regression lines of least squares fits.Left panel: correlation between Brewer and BA02 FTIR retrieval; right panel: correlation between Brewer and SH08 FTIR retrieval.

Fig. 10 .
Fig.10.Time series for the differences of total O 3 measured by the Brewer #157 and the FTIR.((#157-FTIR)/FTIR).Content of panels, symbols, and colours is the same as in Fig.3.

Fig. 11 .
Fig. 11.Time series for the differences of total O 3 measured by the Brewer #185 and the FTIR.((#185-FTIR)/FTIR).Content of panels, symbols, and colours is the same as in Fig. 3.

Table 1 .
Main differences between Brewer and FTIR experiments.
• 18 N, 16 • 29 W at 2370 m a.s.l.The Izaña Observatory is run by the Spanish Weather Service (Agencia Estatal de Meteorología).It is a World Meteorological Organization (WMO) Global Atmospheric Watch (GAW) station of global importance, and there are many different institutes from different countries involved in its manifold measurement program: in-situ measurements of O 3 , CO 2 , CO, CH 4 , N 2 O, NO x , SO 2 , SF 6 , . . ., different in-situ analysers and filter radiometer to determine optical, physical and chemical properties of aerosols.In March 2001 Izaña's ECC-sonde, DOAS, Brewer and FTIR activities were accepted by the NDACC (Network for Detection of Atmospheric Composition Change, formerly called NDSC: Network for Detection of Stratospheric Change tween the years 2005 and 2006/2007.First, the year 2005 is much more variable, which may indicate that the ILS of the FTIR instrument is less stable than in 2006/2007.Second, the difference between both approaches is larger in 2005 than in 2006/2007 (1.5% compared to 0.8%).Assuming that the amplitude of the difference follows the amplitude of the remaining ILS error means that in 2005 the ILS is less well characterised than in 2006/2007.The bottom right panel of Fig.6shows the situation for slant column amounts below 400 DU.Here the differences are especially large in October 2005 (up to 2.2%).Furthermore, we observe that the difference increases from April to October 2005.We think that these observations indicate a gradual decrease of the ILS performance during 2005.
The upper panels show the situation for all 106 coincidences www.atmos-chem-phys.net/8/5535/2008/Atmos.Chem.Phys., 8, 5535-5550, 2008 between January 2005 and November 2005.For increasing slant column amounts we observe a clear increase of the Brewer #157 O 3 amounts with respect to the FTIR amounts.At low slant columns the Brewer measures amounts which are 5.6% lower than the FTIR amounts.For slant column amounts at 700 DU this difference reduces to 4.4%.For the 2006 period the dependence on the slant column amount is less pronounced: at 700 DU the difference is around 4.5% and below 450 around 4.8%.This means that the FTIR and Brewer #157 data are more consistent in 2006 than in 2005.The upper right panel of Fig. 12 has significant similarities with the upper right panel of Fig. 7.This suggests that in 2005 the inconsistencies between the FTIR and Brewer #157 are mainly due to errors in the FTIR data.