We report new determinations of total ozone obtained by reanalyzing a unique set of astronomical observations that were made in the mid-20th century at observatories in France (Haute-Provence) and Switzerland (Jungfraujoch) for the purpose of calculating nightly atmospheric extinction coefficients in the UV (Rayleigh scattering and total ozone) as part of a program to measure absolute stellar fluxes. Only a small fraction of the original ozone results, corresponding to data obtained during 1958–1959, are in the public domain at the World Ozone and Ultraviolet Data Centre; the rest were on handwritten sheets and were stored at Haute-Provence. Both astronomical sites are close enough geographically to Arosa (Switzerland) that the respective ozone values can be compared directly. The comparison reveals a generally very close resemblance, even down to the pattern of daily variations, with a correlation coefficient of 0.78, but an overall negative bias of 6–7% in the stellar results. The bias appears to be slightly larger prior to 1958.
 Global column ozone (O3) has been decreasing since the 1980s and requires continued monitoring, even though the increase in chlorofluorocarbons (CFCs) has now abated [e.g., World Meteorological Organization, 2003]. Since global coverage of O3 abundance only began with the era of satellite observations in the 1970s, knowledge of column O3 prior to the buildup of CFCs is limited to a few Dobson sites. There is therefore a sparsity of data available for testing models in the pre-CFC release era. Any other reliable and independent data set will consequently be useful for assessing O3 variability and trends, and to that end an investigation is under way to extract column O3 from historic observations of certain stellar spectra. In this paper we reintroduce a specific O3 data set derived for astronomical research but whose usefulness for atmospheric science has not previously been exploited. In this initial study we redetermine the O3 columns by applying modern absorption coefficients and a revised stratospheric temperature to the original measurements.
 The longest continuous archive of column ozone measurements, at Arosa in Switzerland, commenced in 1926 and is still ongoing, the data deriving from measurements of the Huggins bands (∼305–340 nm) made as frequently as conditions have permitted. Much is owed to the enthusiasm, foresight and persuasive powers of its founder, F. W. P. Götz, in establishing a facility which would continue and even grow in value decades ahead. The number of O3 measurements prior to the IGY in 1957–1958 was limited. However, there are some unexplored astronomical resources (heritage photographic spectra exposed at various sites) that could be investigated as a possible complement, especially in regard to the earlier years, to routine ground-based O3 monitoring. To that end, a CFCAS-funded project at York University has been seeking out and analyzing archived stellar spectra with a view to (1) extracting O3 columns that predate the first ground-based records and (2) examining the natural behavior of O3 at sites far removed from Arosa. In the course of that work a complete subset of astronomical atmospheric extinction measurements, originally unknown to us, was uncovered in France. Those data, which form the subject of this paper, have not hitherto been used for any geophysical purposes and only a small fraction of them has to date been lodged with the World Ozone and Ultraviolet Data Centre (WOUDC). A full appraisal of the original results extracted from those data will be given in a later paper in this series.
1.1. Purpose-Built Ozone Instruments
 The early service instrument employed in ground-based O3 measurements was the Dobson spectrometer, designed and constructed in the early 1920s by G. M. B. Dobson in Oxford, UK, and subsequently copied as the standard for ozone observing stations worldwide. Such an instrument was used at Arosa, and is still one of the current workhorses.
 Because of a growing awareness that column O3 data reflected weather conditions in the upper troposphere and lower stratosphere a global network of sites and sonding experiments to monitor both O3 columns and the O3 height profile was initiated during or following the IGY. Since 1970 dedicated instruments have also operated from space, the longest run of satellite records being the observations by TOMS from 1978–1993. Intercomparisons have thus made possible the identification of instrumental drifts in at least the recent operating Dobson spectrometers [Staehelin et al., 1998], while the ground network has also proved particularly valuable for comparing with space measurements.
 The interpretation of measurements by a Dobson spectrometer incorporates laboratory O3 absorption coefficients. The earliest series were processed, or reprocessed, with laboratory data based on the work of Ny and Choong , and have since undergone a somewhat chequered history. First the adoption of newer values published by Vigroux  necessitated a rescaling of previous values by a factor of 1.36. It was later realized that the Vigroux coefficients gave inconsistent O3 values (as much as 10% was acknowledged) for different band pairs, so a specific pair (AD) was adopted as standard. In 1968 and again in 1992 revised sets of absorption coefficients were then recommended by the International Ozone Commission, following a redetermination of the laboratory values [Vigroux, 1967; Komhyr et al., 1993] and an examination of new O3 observations.
 Homogenization of long runs of data as are maintained in the Arosa and other atmospheric science archives demands great care and patient research, and much credit is due to the Staëhelin-Brönnimann group at ETH-Zürich for their monumental efforts [Staehelin et al., 1998; Brönnimann et al., 2003a, 2003b] in appraising and reanalyzing many of the older O3 data collected by various Dobson instruments in the world, including those operating at Arosa.
1.2. Archives of Astronomical Spectra
 As mentioned above, this project was initiated to investigate the extent and quality of O3 column information that can be extracted from historic stellar spectrograms; a “proof-of-concept” study, which compares stellar O3 results with TOMS data, has been described by Griffin . During our pursuit of archived stellar spectra we learned of a particularly pertinent set in France. The original project was designed to measure absolute stellar fluxes photographically from the UV atmospheric cutoff to the redward limit of sensitivity of a photographic emulsion, and was carried out during the period 1953–1972 by astronomers of l'Observatoire de Paris under the direction of D. Chalonge. The project required nightly determinations of the atmospheric extinction in order to make the necessary wavelength-dependent corrections to the observed stellar fluxes. The plates have now been archived by Chalonge's chief research assistant, L. Divan, at l'Observatoire de Haute-Provence (OHP), though not as part of that observatory's own archive.
 In the following sections we outline the methods employed to extract total O3 from the stellar observations within their original astronomical context, and describe our reanalysis using more recent laboratory data and a revised stratospheric temperature. Since the primary photographic density measurements were made in routine fashion by a fully trained and competent team using equipment which is no longer extant, we have commenced our reanalysis by adopting those numerical measurements.
2. Stellar Observations and the Chalonge Spectrograph
2.1. Stellar Photometry
 The spectral energy distribution of a star furnishes us with fundamental astronomical information from which the star's principal physical characteristics (temperature, radius, mass) can be determined, and its age and evolution divined. The shape of the full spectral energy curve from far UV to far IR, when approximated by a black body, yields the star's effective temperature, Te; the height of the discontinuity in the continuum levels between 400 and 364 nm (where the Balmer continuum appears) indicates the size of the star (i.e., whether a giant or dwarf), while an integration of the observed fluxes provides its absolute magnitude or luminosity. Measurements of the gradient in the stellar continuum in the blue (400–460 nm) can also be adapted to yield Te. The astrophysical application of observations of stellar energy distribution curves has been described by (inter alia) Chalonge and Divan ; we refer preferentially to that description since the measurements which form the basis of the present paper were designed with the objective of supplying material for the Chalonge-Divan program.
 Although simple in principle, those photometric observations are tricky to perform accurately and precisely. Ground-based astronomical photometry is not readily adapted for absolute measurements, and has to be calibrated by the observation of (1) one or more “standard” stars whose spectral energy distributions are assumed known and can be reliably modelled, plus (2) a terrestrial source whose temperature and blackbody curve can be readily measured in a laboratory. The latter requirement has caused astronomers untold grief in their attempts to construct, observe and analyze the output from a mere tungsten filament lamp as a comparison for stars of some 10,000–25,000°K, while the former has sparked a volley of arguments [e.g., Hayes and Latham, 1975, and references therein] over treatments of the observed and modelled energy distribution of the bright standard star Vega. (Furthermore, like all bright stars it seems, the enigmatic Vega has proved to be anything but a normal standard, though probably not so as to undermine seriously the decades of its use as such in these experiments. Ground-based photometry has also to be corrected for absorption by the Earth's atmosphere, and it is the determination of those corrections which is germane to the present paper.)
 All observations of celestial objects made from the ground receive light that has necessarily traversed the layers of absorbing gases surrounding the Earth. The composition of those layers may vary from place to place, with altitude, temperature and season (or weather), while the amount of absorption detected will also depend on the elevation of the observed object (path length). The practice normally adopted by photometrists is to measure, in sequence with the target star, one or more standard stars that are as close to it as possible in the sky; the measurements of the relative fluxes of the standard star(s), presumed known, thus calibrate the photometry of the target star, and at the same time eliminate the overall telluric absorption (which is presumed the same for both target and standard) in one step.
 Most stellar spectrographs are not suitable for the observation of stellar fluxes. The use of a narrow slit inevitably excludes an unknown fraction of the image that spreads over its jaws as a result of atmospheric scintillation, and the need to observe the whole spectral range in one exposure (to avoid changes in instrumental settings) demands a wide-field detector with uniform response, something that is still difficult to achieve. A photographic plate could span the physical extent of a spectrogram of the modest resolution that is sufficient for this task, but its wavelength-dependent response and limited dynamic range called for specific measures to prevent the plate from overexposure in the visible region (which would render it useless for photometry) while at the same time recording the spectrum adequately down to the violet limit. Like many others in the field both before and since, Chalonge planned the construction of a purpose-designed spectrograph which incorporated special features to mitigate, if not fully compensate for, most of those problems. Designed to be mounted at the Cassegrain focus of a modest telescope and using only quartz optics, the Chalonge prism spectrograph was constructed in triplicate at the Paris Observatory in the early 1950s [Baillet et al., 1952]; it provided a dispersion in the focal plane of 22 nm mm−1 at Hγ (434 nm), or 8.0 nm mm−1 at 320 nm. A unique feature was a rotating sector which spread out the received spectrum in proportion to wavelength, thus avoiding overexposure in the blue while achieving a satisfactory exposure in the UV.
 During 1953–1972 the spectrographs were operated from time to time at four different sites: OHP, Jungfraujoch Observatory (JO) in Switzerland, the European Southern Observatory (ESO) in Chile, and McDonald Observatory, Texas, though this last was for one observing run only. Almost all of the plates were analyzed at l'Observatoire de Paris by L. Divan or by assistants working under her supervision. Relevant details of the reduction process are summarized below (see section 2.4); they are described more fully by L. Divan (manuscript in preparation, 2006).
2.3. Observing Procedures
 In order to convert their ground-based stellar spectrophotometry into extraterrestrial fluxes, the Chalonge team designed observing routines to extract unambiguously the individual wavelength-dependent components of telluric absorption (Rayleigh scattering and ozone absorption) for each night, whenever conditions permitted. The atmospheric extinction is expressed as
where m is the air mass traversed, αλ the Rayleigh scattering coefficient at wavelength λ, z the mean zenith distance of the star under observation, ɛλ the effective ozone column and κλ the O3 absorption coefficient, so κλ.ɛλ is the O3 absorption at wavelength λ, at minimum air mass, that is, at the zenith. Aerosol scattering was ignored on the grounds that it had relatively little effect at UV wavelengths. By making two observations of the same object at zenith distances z1 and z2, corresponding to air masses m1 and m2, one could then work differentially and avoid certain absolute quantities:
where Aλ represents the difference in atmospheric extinction between the two positions of the star.
 The accuracy and precision of the results can in principle be improved by increasing (m2 − m1). Pairs of observations of standard stars were therefore made at both high and low elevations, within limits ranging from 0° to 60°–65° zenith angle (the latter enforced by engineering constraints on most telescopes). An extra refinement, sometimes achieved at ESO, involved observing a standard star both low in the east and (as it set) low in the west, so the atmospheric transparency could be monitored for at least 6 hours. The recorded metadata included the celestial coordinates of the star, its hour angle (angular distance from the meridian) at midexposure, the time of the observation, and the local barometric pressure. The standards selected for the observing program were hot stars. Exposures were characteristically a few minutes in duration, and covered 315–610 nm.
2.4. Original Data Reductions
 In total, several hundred plates were exposed during observing runs with the Chalonge equipment. Not all were capable of yielding independent measurements of column O3, owing largely to vagaries of the weather, and not all of the plates exposed, notably those of the latest runs and including that at McDonald Observatory, were ever reduced by the Paris team. The great majority of the observations were, however, fully reduced; the workings are handwritten, and include all the relevant intermediate steps.
 The Chalonge team traced their spectra with an analogue microdensitometer at the Paris Observatory. Wavelength reference points were marked on the output chart recordings, and intensities in each spectrum were measured at 50 specific wavelengths. Half of those were in the UV between 311 and 334 nm, where they corresponded alternately to the maximum of an O3 feature or to an adjacent cross-section minimum, as already described elsewhere [e.g., Chalonge and Divan, 1952]. Unfortunately all the scanning and chart-recording equipment and handmade aids employed by the Chalonge team have long since been discarded, so it is not possible even in principle to repeat those procedures today. Moreover, the strong likelihood of “personal equations” in the reduction procedures means that even if appropriate digital technology were now made available, the complete set of plates would have to be reprocessed, a project that is beyond the scope of this initial investigation of those data.
 The first step in calculating atmospheric extinction was to eliminate the Rayleigh scattering. The logarithm of the ratio of the measured intensities corresponding to high- and low-elevation observations of the same (standard) star for each wavelength point was evaluated as log (I/I0), scaled by the difference in air mass, and plotted against (μ2 − 1)2λ−4, where λ is the wavelength in cm, μ the refractive index of air. Between 350–450 nm the relationship was practically linear, with gradient αλ (the Rayleigh scattering coefficient). Other sources of absorption were detected as deviations from a linear extrapolation, those at the short-wavelength end being due to the Huggins O3 absorption and equivalent to κλ.ελ. When the measured deviations, Dλ, of those scaled logarithmic intensities were plotted against the laboratory O3 absorption coefficients for the corresponding wavelengths a linear relationship with a gradient of ελ was obtained. No specific reference to aerosol scattering was made by the Chalonge team, beyond a verbal expectation that it would not contribute at UV wavelengths. In fact, since Mie scattering varies only very slowly with wavelength any contribution will in effect have been removed from the observations through the empirical determination of αλ.
 The absorption coefficients adopted by the Paris team were those of Vigroux , and later their subsequent update [Vigroux, 1967]. Although Vigroux's observations of the O3 absorption spectrum were of high enough resolution to detect some of the band fine structure, he only tabulated values of the absorption coefficient at points of inflexion corresponding to maxima or minima, or secondary maxima and minima if they could be distinguished. His laboratory experiments were conducted at a range of selected temperatures (T) from −92°C through −75°C, −59°C, −44°C, −30°C to +18°C and beyond, thus encompassing the temperature of the stratosphere. The Chalonge team adopted for the stratosphere a temperature of −30°C, as recommended by Barbier and Chalonge .
3. Reanalysis of the Chalonge Ozone Measurements
3.1. Selection of Data
 From a set of 306 nights on which useful data were obtained and originally reduced, we selected a subset of 240 nights involving 35 observing runs; most of the observing runs were at OHP, 11 were at JO and three of the later ones at ESO. Table 1 gives the geographical coordinates of the three sites. We limited our selection to the better quality, independent determinations of κλ.ελ, discarding measurements made on any night in which an impurity or variability of the atmosphere was noted or demonstrated since it placed doubt on the photometric quality of the night in question and hence on the reliability of the effective O3 columns that were determined. Data were also discarded if the intended sequence of observations of standard spectra was interrupted for some reason. Table 2 summarizes the material used in this reanalysis. From an astronomical perspective, even if the criteria for a fully independent determination of the atmospheric extinction could not be met during any one night it was still possible to salvage something of value from the stellar observations, either by accepting a less well determined value of the atmospheric extinction or by assuming a value for αλ and adopting the O3 columns published by Arosa for the adjacent days. However, for the present work the components of atmospheric extinction derived for such “nonphotometric” nights were clearly not independent determinations.
Table 1. Coordinates of the Observing Sites and of Arosa
 A digital database of the relevant sections of the handwritten observations and reductions was created. In it was copied, for each night, the determined αλ, and the measured excess absorption, equal to κλ.ɛλ, which remained after the Rayleigh scattering had been subtracted. For reference, it also recorded the values of ɛλ (in cm) determined by the Paris team. The reductions carried out by the latter received the greatest attention to detail which the data and the available equipment would permit, and (more importantly) ensured a remarkable degree of homogeneity in the results throughout the entire interval spanned by the observations. It was therefore considered neither necessary nor desirable at this point to try to remeasure the original plates, and our redetermination of ελ proceeded from the κλ.ελ values already recorded.
3.2. O3 Absorption Coefficients
 In this reanalysis of the Paris measurements we adopted the Bass-Paur absorption coefficients tabulated by Komhyr et al. , and interpolated them in steps of 0.005 nm for T = −52°C; they are plotted in Figure 1. On the same graph, for comparison, are plotted the absorption cross sections measured at the same T by the Bremen group [Burrows et al., 1999] after rescaling by Löschmidt's number. (Although we later choose a representative stratospheric T of −44°C, our choice at this juncture was limited to the values for which the Bremen data were published). We have included the difference spectrum (Bremen–Bass-Paur) in Figure 1; for clarity, the differences have been multiplied by a factor of 5. There is a systematic discrepancy between the two data sets of some 2%, which can be annulled if the temperature assigned to the Bremen absorption cross sections is about 22° lower than that used to calculate the Bass-Paur coefficients. We note that the Bass-Paur absorption coefficients are the ones employed in calculating O3 columns from measurements made with modern-day Dobson and Brewer instruments and with TOMS. The T dependence of O3 has already been discussed elsewhere [e.g., Barbier and Chalonge, 1942; Voigt et al., 2001]; such sensitivity to T, even if not large, will evidently be a limiting factor in the eventual accuracy of all column O3 measured via the Huggins bands since it is necessary to represent a range of stratospheric T along the line of sight by a mean value. The height profile of O3 also alters with season and with meteorological conditions, so without comprehensive measurements of the prevailing conditions over each observatory some additional uncertainty is inevitable on that account as well.
3.3. Redetermining ελ
 The values of κλ.ελ listed for each wavelength point for each night were plotted against the Bass-Paur κλ calculated for a mean stratospheric temperature of −44°C. There is justification for adopting a mean stratospheric T for these midlatitude data: recent investigations by Bernhard et al.  have shown that T-related errors in the Dobson data are typically within 2% unless the solar zenith angle is substantially larger than was the case for any of these stellar measurements. It was assumed that the relationship was linear and that it passes through the origin. The gradient of each set was determined as the line giving the best fit to the observed points by minimizing their perpendicular distances from it; each gradient gives ελ for the night in question. Sample plots are shown in Figure 2 for JO, Figure 3 for OHP and Figure 4 for ESO.
3.4. Seasonal Variations
 For this purpose we combined the data from the two northern astronomical sites, since they are at similar latitudes (see Table 1). Owing to their considerably different elevations dissimilar degrees of aerosol concentrations and tropospheric O3 variations could have been sampled, though a difference of only ∼0.5% (see section 4) was found in their respective overall O3 averages.
 In Figure 5 the columns determined for each night for OHP and JO are plotted against calendar month. The seasonal variation is obvious, and in the manner expected: a maximum in spring and a minimum in the fall; the dotted curve (depicting the Arosa monthly means for 1953–1972) suggests its form.
3.5. Possible Sources of Error
 Errors or uncertainties will have been introduced into the original analysis and into our reanalysis via the graphs, similar to those in Figures 2, 3, and 4, which were constructed for each observing run. They can be summarized as (1) photometric errors in the stellar measurements, (2) errors in the intensity readings from microdensitometer tracings of the stellar spectra, and (3) errors in the laboratory O3 data.
 Errors could also have been introduced through (4) changes in the intrinsic O3 levels during the interval of (typically) 3–4 hours between a pair of observations of a particular standard star, and (5) excluding contributions from aerosol scattering. With regard to source 4, the routine inclusion of more than one standard star per night ensured that the photometric quality of the night was adequately monitored. If variations were suspected the basis of the extinction calculations was then rendered invalid; such nights were excluded during our selection procedure (section 3.1). Source 5 was not quantified at the time and is therefore potentially of some concern. Industrial pollution from Marseille, ∼70 km to the SW of OHP, was a conceivable source if it penetrated to the rural areas where the stellar observations were made. SO2 absorption is only significant shortward of ∼315 nm; the band near 313.2 nm could depress the “minimum” O3 intensity reading at 313.5 nm near the short-wavelength limit of the Chalonge program, though probably tending to add to the scatter in Figures 2 and 3 rather than affecting the gradients. On the other hand any Mie scattering, which has only a slow wavelength dependence, will have been removed from the intensity measurements along with αλ, as described in section 2.4. A study of the presence of SO2 absorption and of Mie scattering as deduced from these historic data would constitute an interesting project on its own.
 One indicator of a problem is discernible in Figures2–4, where the open circles, representing the intensity measurements made within a Huggins feature and assigned odd numbers in the French reductions, tend to fall below the mean line while the solid points, representing measurements made in the nearby apparent stellar continuum (O3 minima) and assigned even numbers, are mostly on the higher side; this “odd-even” effect is particularly noticeable in Figure 2. Sources 2 and 3 of the errors list above (and discussed in detail below) could both have been responsible.
3.5.1. Photographic Photometry
 The nonlinear response of a photographic emulsion to light can be a source of error in the inferred intensities if the necessary intensity calibration is not applied rigorously enough. For high-dispersion photographic spectroscopy one may be obliged to determine a fresh calibration curve whenever its characteristics (which are wavelength sensitive) alter by more than a specified amount. In the near UV the wavelength sensitivity of an astronomical emulsion is usually not very acute, and a common procedure is to adopt a single calibration relation for the entire spectral span. However, the effect is most noticeable when comparing measurements of rather sharp lines of substantial strength, whereas in relative measurements of features that are all of a similar (and fairly weak) intensity the error can be expected to be rather small.
3.5.2. Intensity Readings
 Reading microdensitometer charts can be problematic, particularly where they correspond to rather high or low photographic densities, and the readings will always contain a personal equation. The tracing of a weak photographic exposure is dominated by noise arising from the grain in the emulsion, which manifests itself as a “jitter” whose amplitude depends on the size of the grain (i.e., on emulsion type) and on the signal-to-noise ratio. Just where one defines the mean level through the noise in the “continuum” corresponding here to the O3 minima is partly a matter of personal equation, but can also be influenced to some extent by the slope of the tracing in the region in question, and by the presence of weak but real stellar features. If the continuum level is slightly misplaced the measurements of the O3 minima will be affected more than those of the maxima, and could thus contribute to the “odd-even” effect described above. Since the effect in Figures 2–4 is predominantly in the same sense, it seems more likely to have been due to a general overallowance for stellar features than to a random personal equation.
 Reciprocity failure (the inability of the grains in a photographic emulsion to turn black below a certain threshold of incident light) is another relevant factor. A weak exposure could thus result in intensities, especially of the minima, that are too small where the emulsion sensitivity is relatively poor, which for these spectra is at the shortest wavelengths and is also incidentally where the strongest Huggins features occur; the effect would therefore be in contrast to the observed “odd-even” effect. However, the fact that intensity levels may be near those of the photographic noise also risks a bias that has been well demonstrated among observational stellar spectroscopists, in that the noise on low-dispersion tracings tends to be misinterpreted as actual stellar lines and the continuum is accordingly placed higher (and the intensities therefore measured as too strong) than when the same features are measured on better quality material where their profiles are more cleanly defined.
3.5.3. Laboratory Data
 The laboratory data are highly sensitive to T, both in absolute values and in the contrast between maximum and minimum, more noticeably so for the stronger features (see Figure 1), so a small deviation of the selected T from the “best” representative value could affect the gradient in Figures 2–4, as mentioned in section 2.4. The “odd-even” effect can readily be increased or lowered by respectively reducing or raising the value of T, even by as little as 14°C from the value of −44°C used here to the −30°C applied in the original French analyses.
 Investigations of the effects described above proved interesting but inconclusive. In order to estimate the magnitude of a possible continuum error we applied a downward shift to the values of Dλ corresponding to the even points (equivalent to readjusting the stellar continuum levels) to bring them in line with the odd points. For example, the alignment of the intensities for JO in March 1954 can be improved by a continuum correction of −3.5% to the relevant ordinates; the corresponding gradients were then reduced by 3.5–4.5%. We also tried selecting T other than −44°C; using T = −30°C instead of −44°C we again induced a decrease in the derived gradients, of 2.5–3%. However, without proof that the causes we have thus treated were actually dominant in these stellar observations no conclusions can be drawn.
4. Comparing Stellar and Arosa Data
 We limited this aspect of the work to the 204 observations obtained at OHP and JO. Since the observatory sites in France and Switzerland are not far distant (OHP-Arosa = 448 km, Jungfraujoch-Arosa = 130 km, OHP-Jungfraujoch = 338 km) a strong correlation between all three sets of O3 measurements is anticipated. For example, in their assessment of the performance of the world ground-based total ozone network from comparisons with satellite data Fioletov et al.  showed that ozone variations have a large spatial scale; the standard deviation of the difference between O3 values measured at sites ∼500 km apart was just 4%. A close correlation was also assumed by the Chalonge team, who adopted Arosa O3 values for their stellar program on nights when it had not been possible to derive good-quality extinction values from their stellar measurements. Considering (1) that the Arosa columns are means of daytime measurements whereas the stellar measurements were performed during the night, and (2) the geographical distance involved and the prevalence of low-pressure weather systems advancing from approximately SW to NE, that is, from OHP toward Arosa, we compared each stellar value with the mean published for Arosa for the following day, though thereby limiting the data set to 155 observations. Figure 6 is a scatterplot of the stellar observations against Arosa values, while the mean difference and its s.e. for each observing run is displayed in Figure 7. We have rather arbitrarily rejected the OHP measurements from 1956; their mean differences of the order of −22% seemed exceptionally low and the data quality was somewhat borderline. Figure 8 compares each stellar column against the daily Arosa means (drawn as a continuous line) for the entire period.
 In Figure 6 the stellar data are correlated to the Arosa ones by a coefficient of 0.78, the separate coefficients being 0.73 for OHP (103 nights) and 0.89 for JO (52 nights). Factors affecting those correlations include the quality and quantity of observations and the distance between the stations, so a higher correlation coefficient could be expected for the JO data. The data show a bias of 7% that is the same for both sites, though it reduces to about 4% for the more recent (and more sparse) data. That tendency can also be detected in the distribution of mean differences in Figure 7; when, for example, the differences are segmented into pre-1958 and post-1958 so as to divide the sample in half, they reveal a step which is statistically significant at the 95% level, the pre-1958 values giving −8.0% ± 0.9 and the later ones −4.6% ± 0.9. Since very little of that may be attributable to distance it suggests the influence of an additional agent during the earlier years investigated here. However, more data are required before that result can be confirmed or investigated in depth.
 The overall (weighted) mean difference between the stellar and corresponding Arosa columns is −6.32% ± 0.67. The mean differences for the individual sites are almost the same: −6.1 (OHP) and −6.7 (JO), though it should be noted that observations were never made simultaneously at the two sites. A corresponding adjustment has been applied in Figures 5 and 8, but no adjustments have been made for seasonal variations. No correction has been made for tropospheric contributions to total O3; in view of the substantial difference in altitude between OHP and JO (see Table 1) the JO columns might be expected to show a systematic deficiency of up to 2% relative to OHP.
 A systematic difference between the stellar and solar results is not unexpected, given the quite different basic concepts of analysis and laboratory data upon which they have depended, but the repetition of many of the features illustrated in Figure 8 is quite impressive. The detailed similarity in the pattern is particularly remarkable in 1954.2, 1960.0, 1960.8, 1961.0, 1961.8, 1964.1, 1965.2, 1965.8, etc. However, that frequent closeness of the stellar and solar columns raises the question as to why the match is not uniformly as good, and the next section discusses possible reasons.
5.1. Quality of the Stellar Data
 The stellar measurements are much smaller in volume and have much larger random errors than the solar ones. Because of the nature of astronomical observing and telescope time allocation they are also rather severely bunched, too much so for the construction of meaningful running averages. Moreover, the stellar observations were made in simple pairs at large and small zenith angles, rather than in sequences at different zenith angles as were the solar ones. On the other hand the exposures were only a few minutes long, and since the local barometric information was carefully recorded the differences in calculated air mass were known rather precisely. The same equipment appears to have been used throughout (though unspecified “improvements” are occasionally mentioned in the literature), and the same team was responsible for both the observing and the data reductions. We performed an external check on the homogeneity of the stellar data by examining the calculations of the Rayleigh extinction coefficient (αλ), as recorded by the French team, for evidence of equipment problems or a transient bias in the reductions (a positive bias would result in O3 being too low). The absence of any correlation provided some reassurance concerning the overall homogeneity of the stellar data set.
Section 3.5 rehearsed the errors that are likely to have affected stellar observations of this nature and their analysis. When a photographic exposure is rather weak the detection of O3 features can be challenging, and their measurements on tracings which already show a substantial slope, as any of the descriptive papers from the French group [e.g., Chalonge and Divan, 1952, Figure 1] illustrate, may well contain systematic errors. The scattering of light that bedevils observations of a source as bright as the Sun, especially at very large zenith angles where the UV signal is greatly attenuated, is less of a problem with stellar observations than is the detection and measurement of a signal that is weak. Although a hot star emits relatively strongly in the UV, the increasing atmospheric extinction, and a possible fall in efficiency of the spectrograph, combine to reduce the amount of light received at the shortest wavelengths; even though that is where the stronger Huggins features occur, it is the observed intensity of the stellar continuum that is the important factor. Our reason for rejecting the particularly low OHP values for 1956 was based on a general weakness of those observations. However, the observations from which Figure 2 was derived were evidently rather well exposed, and yet that observing run at the JO in 1954 yielded a mean percentage difference of −13.2, which is the biggest discrepancy of all the pre-1958 runs. The JO run earlier in 1954 also shows a rather low value (−9.7).
 The choice of O3 laboratory data does not appear to have had much influence on the detected bias, since our plots of Dλ/(m1 − m2) against κλ.ελ for different runs show very similar patterns to those constructed by the Chalonge team, albeit on a different vertical scale owing to the different scaling of the laboratory values used by them.
 The temperature dependence of the O3 absorption coefficients (section 3.2) affects all total O3 determinations, and may represent a universal limitation to the precision of total O3 measurements. Section 3.5 showed how changing T can induce a systematic difference in the results. We cannot conclude simply that the temperature of the stratosphere in the 1950s was higher than the mean value adopted in our reductions, because the change in the bias is a differential effect and the same value of T must apply to Arosa as well as to both stellar sites. It is possible that the value of −44°C, though currently the widely recommended mean for this type of work, is not in fact the most representative one. However, if (as is probable) a change of T affects the different methodologies in different ways, it remains plausible that the representative temperature of the stratosphere has undergone changes that have not been reflected adequately in either the stellar or the solar reductions.
5.2. Quality of the Solar Data
 Accounts of the pioneering O3 monitoring in Oxford (UK) and its parallel developments in Arosa [e.g., Dobson, 1968] provide an excellent background history of the instrumental techniques and their variants which were central to the then embryonic field of O3 monitoring. As the need for a broader network of independent stations (already Dobson's vision) became more obvious, and was explicitly planned for the IGY, requirements for better standardization of the different contributors became apparent. Examination of the accrued data sets for long-term trend analyses demanded even more stringent calibrations of ongoing data sets, or reworking the more historic ones that were no longer active. Relatively short-lived apparent anomalies, such as the period of strong O3 in 1940–1942 [Brönnimann et al., 2004], need to be well documented so as not to bias evidence of broader decadal trends. The size and scope of the effort required and the statistical reliability of the results for that purpose are fully described by Brönnimann et al. [2003a, 2003b]. The resulting homogenization, both of the Arosa data and of the whole ground-based network linked to it (all archived at WOUDC), has created an invaluable tool for long-term trend analyses.
 The same standards of excellence presently achievable cannot be extended backward at quite the same high level to the period before the IGY, when independent measurements were rarely available for providing checks on instrument stability. Developments in detector technology (from photographic plate via photoelectric recorder to photomultiplier) and revisions in the choice of wavelength band pairs were (rightly) recognized as progress in the field, so a superseded instrument became merely a source of spare parts for the upgraded one that replaced it. However, as a consequence one finds that (for example) when the Arosa Dobson spectrograph known as D2 was replaced by D15 in 1950 the archive records only about 20 truly overlapping measurements, and those during the space of just a few weeks. That history has left open the small possibility that the various data recalibrations, exacerbated by new spectrometers having different characteristics and sensitivities from the ones they were replacing, may have unwittingly introduced different biases in the data, for short or long periods, so the availability of independent series of total O3 which are homogeneous, or at least exhibit a different set of biases, is likely to be of substantial value. The Arosa observing station actually occupied three different sites, relocating in 1954 and 1972, but as all were less than 500m apart and at closely similar altitudes it was reckoned that the various sets of measurements could be combined without requiring adjustments on that account.
 A reanalysis of atmospheric extinction measurements (Rayleigh scattering and total O3) made at sites in France and Switzerland between 1953–1972 has yielded total O3 values which, for the most part, agree remarkably well with the daily means from Arosa. We detected an overall bias of −6.3%, regardless of site, in the stellar data and an indication that the bias was more negative before O3 monitoring burgeoned following the IGY in 1957–1958. We have reviewed various possible causes of error in the stellar measurements, but cannot yet identify unambiguously the reason for such a change in the bias.
 On the one hand, the stellar database is remarkably homogeneous; the measurements were treated uniformly throughout, effectively by the same personnel. On the other hand, the stellar measurements are grouped into short, concentrated runs and as such are not optimal for interseries comparisons. Short-term instrumental variations cannot be ruled out, but the fact that both OHP and JO exhibit the same trend reduces the likelihood of undetected transient problems in both stellar spectrographs.
 Because of the nature and the history of the stellar work it is not now feasible to reconstruct precisely all the different reduction steps that have been carried out on those stellar data, just as it would be very difficult to reconstruct all the recalibrations that were ever performed on the original Arosa measurements. The most promising pursuit for more, and more reliable, independent information is to increase the size and scope of the stellar database, and the next intention for this research is to analyze quantities of historic stellar spectra exposed at other midlatitude observatories (though mostly not in Europe) during the period 1930–1960, using the method described by Griffin .
 It is worth pointing out that the stellar data analyzed in this paper represent only a very modest, though a particularly well angled, subset of all the stellar data which can in principle be reworked for total O3. It is hoped that this demonstration of their interdisciplinary usefulness, in concert with the historic results described by Griffin , will provide a strong impetus to continue and expand this aspect of ozone research.
 First and foremost we would like to thank L. Divan, formerly of the IAP, Paris, and now retired to OHP, for her help, cooperation, and patience in allowing full access to her excellent plate and paper archive of the substantial volume of Chalonge material which she had personally rescued and installed at OHP. The added value which she graciously supplied to our project through many personal reminiscences of the people, conditions, and conversations that were germane to the original data reductions that we adopted was an unexpected and much appreciated bonus. This project has been supported by the Canadian Foundation for Climate and Atmospheric Sciences, grant GR-338: Measuring Ozone Columns from Astronomical Archives. We would like to thank our many colleagues for stimulating and fruitful discussions, in particular, David Wardle (MRC, Canada), Johannes Staehelin and Stefan Brönnimann (ETH-Zurich), Philippe Keckhut (Paris and OHP), Ian McDade (York University), and Paul Feldman (HIA). This research was carried out at the Dominion Astrophysical Observatory, HIA, Victoria, Canada, where REMG enjoys Guest-Worker privileges.