High-precision atmospheric oxygen measurement comparisons between a newly built 1 CRDS analyzer and existing measurement techniques 2 3

10 Carbon dioxide and oxygen are tightly coupled in land-biospheres CO2 O2 exchange 11 processes, while they are not coupled in oceanic exchange. For this reason, atmospheric 12 oxygen measurements can be used to constrain the global carbon cycle, especially oceanic 13 uptake. However, accurately quantifying the small (~1-100 ppm) variations in O2 is 14 analytically challenging due to the very large atmospheric background which constitutes 15 about 20.9 % (~209500 ppm) of atmospheric air. Here we present detailed description of the 16 analyzer and its operating principles as well as comprehensive laboratory and field studies for 17 a newly developed high-precision oxygen mixing ratio and isotopic composition analyzer 18 (Picarro G-2207) that is based on cavity ring-down spectroscopy (CRDS). From the 19 laboratory tests, we have calculated a short-term precision (standard error of one-minute O2 20 mixing ratio measurements) of < 1 ppm for this analyzer based on measurements of eight 21 standard gases analyzed for two hours consecutively. In contrast to the currently existing 22 techniques, the instrument has an excellent long-term stability and therefore a calibration 23 every 12 hours is sufficient to get an overall uncertainty of < 5 ppm. Measurements of 24


32
Atmospheric oxygen comprises about 20.9 % of the global atmosphere and in the past (1) 40 In contrast, the global average atmospheric CO2 mixing ratio increased to 405.0 ppm 41 averaged over 2017 since its preindustrial value of 280 ppm (Le Quéré et al., 2017). As the shape modeling. Ultimately, the usefulness of the spectral model is to be evaluated by the 144 precision and stability of the O2 measurements when compared with established techniques. 145 Ultimately, the usefulness of the spectral model is to be evaluated by the precision and stability of 146 the O2 measurements when compared with established techniques.For spectral model 147 development, this spectrometer has the drawback that the cavity FSR, is too large to reveal 148 much detail of the absorption line shape, even with the simplifying assumption of a Galatry 149 line shape. We therefore acquired a set of four interleaved spectra, with the PZT-actuated where ν0 is the transition frequency, kB is Boltzmann's constant (J. K -1 ), T is the sample 166 temperature (K), M is the molecular mass (amu), and c is the speed of light (m/s). Figure 2 shows the values of y and z obtained from spectra acquired in the same way as Figure  to the literature value of 0.233 cm 2 s -1 for O2 in air at 45 °C (Marrero and Mason, 1972). 177 Although the anticipated use of the analyzer is for ambient air samples having a very small 178 range of O2 concentrations, we did investigate the variation of the line shape in binary 179 mixtures of O2 and N2 shown in Figure 3. The error bars are taken from the output of the 180 Levenberg-Marquardt fitting routine (Press et al., 1992). The dependence of the collisional 181 broadening parameter z on O2 mole fraction was considered too small to be significant, but 182 the variation in y was used in the subsequent analysis of the air samples. Note that Wójtewicz 183 et al. (Wójtewicz et al., 2014) also found collisional broadening coefficients for nitrogen to be 184 slightly larger than for oxygen in measurements of one O2 line in the B-band. 185 The primary goal in designing the analyzer was to achieve high enough precision to 186 make meaningful measurements of O2 in clean atmospheric samples. Although the current 187 best practice for such high-precision measurements is to work with dried samples, we decided values is 10 % to 20 %, this level of agreement seems reasonable. 208 We also looked at absorption from water near the Q13Q13 absorption line of O2.

209
These spectra were measured in a background of pure nitrogen to reveal the very weak lines 210 interfering with the O2 measurement. Without the strong O2 lines, it was impossible to 211 interleave FSR-spaced spectra, so in this case the frequency axis comes from the analyzer's 212 wavelength monitor. The upper panel of Figure 5 shows the spectrum of saturated water vapor 213 in nitrogen, together with a fit to a Voigt model of the molecular lines. abbreviated HDO). The lower panel of Figure 5 shows the lines tabulated in Hitran.

218
Immediately after the data in Figure 5 were acquired, measurements were also made at 219 7816.85210 cm -1 , to establish the relationship between the absorption strengths in the two here. In the case of CO2, the dilution of oxygen due to 400 ppm of CO2 is significant, and 227 larger than any direct spectral interference. 228 Finally, we investigated the influence of water vapor on the shape of the O2 Q13Q13 229 line. Switching between the two lasers sources, we acquired FSR-spaced spectra of 230 humidified synthetic air, alternately covering the 7817 cm -1 and 7878 cm -1 regions. Individual 231 spectra were acquired in less than 2 s, so changes in water vapor concentration between 232 spectra were small. These spectra, with frequency resolution of 0.0206 cm -1 , were analyzed by 233 nonlinear least-squares fitting with the following spectral models: the 7817 cm -1 spectra were 234 modeled as the sum of an empty-cavity baseline having an adjustable offset level and slope 235 and three water peaks and the two weak perturbing peaks. The molecular absorption of the 236 main peak was expressed as an adjustable amplitude, Aw, multiplying a dimensionless, area-237 normalized Galatry function (Varghese and Hanson, 1984). The weak perturbers were modeled by Voigt profiles with amplitudes and line widths that constrained to be in fixed 239 ratios to the strong line, and therefore added no new degrees of freedom to the fitting 240 procedure. Since the amplitude Aw multiplies an area-normalized shape function, it is 241 essentially equivalent to the area of the absorption line, to the extent that the Galatry model 242 provides a valid description of the line shape. The Doppler width of the Galatry function was 243 fixed based on the measured cell temperature, the y-parameter was allowed to vary, and the z-  The alternating measurements at 7817 cm -1 and 7878 cm -1 also calibrated the  contribution from pressure variations that the pressure sensor is unable to detect. The ratio 299 AO2/y can be determined from the fit much more precisely than AO2 alone and so gives a more 300 sensitive measurement of molecular absorption. It also has the advantage of being 301 independent of sample pressure, to the extent that the Galatry model applies ( Figure 2).

302
However, using the ratio AO2/y as a metric for absorption adds more complications if 303 measurements are to be made over a range of O2 and water concentrations, because the O2/ N2 304 ratio and water concentration affect the line width independently of pressure and O2 305 concentration alone. To minimize systematic errors due to these broadening effects, we define 306 a nominal y-parameter based on the measured amplitudes of the O2 and water lines and the 307 line broadening dependences shown in Figures 3 and 4. The measured ratio AO2/y is 308 normalized by the nominal y to obtain a quantity that is ideally independent of pressure and 309 water concentration, and this is the quantity that is multiplied by a calibration constant to give the reported O2 fraction. In addition, a dry mole fraction is reported for O2, defined as the 311 directly measured mole fraction corrected for water dilution.

312
The main goal in developing this instrument was to make high precision 313 measurements of O2 mole fraction, based on absorption by the dominant 16 O2 isotopologue.

314
The absorption lines of the rarer isotopologues are also present nearby, so a mode of operation 315 was included in which one laser is scanned over neighboring lines of 16 O2 and 16 O 18 O and the 316 ratio of amplitudes is used to derive an isotopic ratio, reported in the usual delta notation. In that for the O2 fraction measurement, so it will not be described in detail, only the main 321 differences will be noted. One is that in determining an isotopic ratio there is no advantage to 322 be obtained from normalizing absorption amplitudes to line widths, instead we simply take   From these experiments, we determined a temperature sensitivity of -2.1x10 -4 K -1 and 344 a pressure sensitivity of +9.8x10 -6 hPa -1 . The temperature sensitivity is somewhat larger than  program. Possible sources of error include: temperature drifts due to sensor drift or gradients; 365 pressure errors due to sensor drift; optical artifacts such as parasitic reflections, higher order 366 cavity mode excitation, and/or loss nonlinearity that can distort the reported oxygen spectrum.

367
More work is required to identify and eliminate these small drifts.

368
The Allan standard deviation of the reported O2 fraction is shown in Figure 9. The   The performance of the instrument was tested by analyzing eight standard gases with 381 precisely known CO2 and O2 compositions (Table 1) Figure 11. Standard gases were directly connected to the 387 pressure controlling unit, and a multi-port valve (V2) was used to select among the standard 388 gases. The flow from each cylinder was adjusted to about 120 ml min -1 which was eventually 389 directed to a selection valve (V1), allowing switching between ambient air and standard gases.

408
In the cases of the smaller splitting ratios (1:1, 1:4 and 4:1), which are relevant for the 409 results presented in this study, only minor differences in the measured O2 mixing ratios were 410 observed when compared to case ii (i.e. without a Tee-junction). For these two cylinders 411 measured, the average differences in these cases were about 0.5 ppm, calculated as the mean  Figure 12 shows the standard gas measurements for the seven cylinders with known 421 CO2 and O2 mixing ratios (Table 1)   where fH2O is the measured water mole fraction.

497
The efficiency of water correction by this function was assessed in two ways: i) by comparing 498 the water vapor content in standard air measured by this analyzer with similar measurements 499 by the NDIR analyzer and ii) by comparing the oxygen mixing ratios between non-dried ambient air measured and corrected for water dilution by the CRDS analyzer with dried air 501 measured using a paramagnetic analyzer.
502 Figure 14 shows the water vapor content for standard gases measured continuously for 503 two days by the CRDS and the NDIR analyzers. Note that the two data sets are manually 504 fitted to each other as the measured water values by the NDIR analyzer are not calibrated.

505
Based on these plots, the two analyzers are in very good agreement although there are small 506 differences during very dry conditions (low water content).

507
The water correction test was conducted by measuring dried ambient air (Figure 15a measurements of the two analyzers (Fig. A.1). Note that the CRDS measurements were 514 corrected for the observed drift using the polynomial fit to the two standard gas measurements 515 stated above.

516
In the first period of the measurement when both analyzers measured dried ambient 517 air, the absolute differences between the 1-minute averages measured over two days by the 518 two analyzers were mostly within 15 ppm and symmetrically distributed around zero.   the rare isotopologue in isotopic mode Hence, we have conducted the remaining test 598 measurements in concentration mode.

599
As this tower has five sampling height levels, we first followed three minutes of 600 switching per inlet level, which enables four measurements per hour at a given level.

601
However, we noticed hardly any difference among the different levels due to strong short 602 term variability in O2 mixing ratios between the consecutive heights. Hence, we switched to a 603 longer sampling period of six-minutes per height. Figure 17 shows the diurnal CO2 and O2  (Keeling, 1988b).
618 Table 2 shows the oxidation ratios derived as the slopes of the linear regression 619 between CO2 and O2 mixing ratios at the different height levels measured on 25 February 2017. Accordingly, height dependent slopes were observed with a slope of -0.98 ± 0.06 at the 621 lowest level, close to the biological processes induced slope but slightly lower than its mean 622 value. For the highest level, we calculated a slope of -1.60 ± 0.07 a value close to fossil fuel 623 combustion oxidation ratio. Note that depending on fossil fuel type the oxidation ratio can 624 range between -1.17 and -1.95 for coal and natural gas, respectively (Keeling, 1988b). While

662
The T-split tests for the current measurement setup based on the measurements of two  Figure 1.