We have developed an automated observatory for measuring atmospheric column abundances of CO2 and O2 using near-infrared spectra of the Sun obtained with a high spectral resolution Fourier Transform Spectrometer (FTS). This is the first dedicated laboratory in a new network of ground-based observatories named the Total Carbon Column Observing Network. This network will be used for carbon cycle studies and validation of spaceborne column measurements of greenhouse gases. The observatory was assembled in Pasadena, California, and then permanently deployed to northern Wisconsin during May 2004. It is located in the heavily forested Chequamegon National Forest at the WLEF Tall Tower site, 12 km east of Park Falls, Wisconsin. Under clear sky conditions, ∼0.1% measurement precision is demonstrated for the retrieved column CO2 abundances. During the Intercontinental Chemical Transport Experiment–North America and CO2 Boundary Layer Regional Airborne Experiment campaigns in summer 2004, the DC-8 and King Air aircraft recorded eight in situ CO2 profiles over the WLEF site. Comparison of the integrated aircraft profiles and CO2 column abundances shows a small bias (∼2%) but an excellent correlation.
 In the last two decades, numerous studies [e.g., Gurney et al., 2002; Rayner et al., 1999; Tans et al., 1990] have combined in situ measurements of CO2 obtained from a global network of surface sites [GLOBALVIEW-CO2, 2005] with global transport models to estimate regional-scale surface exchange of CO2. Although the surface measurements are highly accurate, their limited spatial coverage and proximity to local sources and sinks make these estimates quite sensitive to the errors in the transport model (e.g., vertical mixing), particularly for sites located in the continental interior. In particular, because the surface fluxes and vertical transport are correlated on diurnal and seasonal timescales, errors in transport fields are aliased into the inferred exchange term as so-called “rectifier” effects [Denning et al., 1996; Gurney et al., 2002].
 Precise and accurate CO2 column measurements can complement the existing in situ network and provide information about CO2 exchange on larger geographic scales. Unlike the near-surface volume mixing ratio (VMR), the column integral of the CO2 profile is not altered by diurnal variations in the height of the boundary layer. As a result, it exhibits less spatial and temporal variability than near-surface in situ data, while retaining information about surface fluxes [Gloor et al., 2000]. Because few CO2 column measurements are available, understanding of their potential information content has been largely limited to simulations [Rayner and O'Brien, 2001; Olsen and Randerson, 2004]. These studies show that CO2 column measurements at carefully selected sites could be effective in constraining global-scale carbon budgets [Rayner and O'Brien, 2001].
 Three recent analyses of near-infrared solar spectra obtained by Fourier Transform Spectrometers (FTS) demonstrate that column-averaged CO2 VMRs can be retrieved with high precision [Yang et al., 2002; Dufour et al., 2004; Warneke et al., 2005]. The near-infrared spectral region is an appropriate observational choice for several reasons: (1) It is near the peak of the solar Planck function, expressed in units of photons/s/m2/sr/cm−1, maximizing signal-to-noise; (2) retrievals from O2 absorption lines in the near-infrared spectral region provide an internal standard; and (3) highly sensitive uncooled detectors are available for this region. For these reasons, the near-infrared region has also been chosen by several space-based column observatories, including the Orbiting Carbon Observatory (OCO), the Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY), and the Greenhouse Gases Observing Satellite (GOSAT).
 Most of the existing FTS instruments from the Network for the Detection of Stratospheric Change (NDSC) [Kurylo and Solomon, 1990] are not well suited for measurement of CO2 and other greenhouse gases. Most NDSC sites are located at high altitude to facilitate stratospheric measurements. To understand the sources and sinks of greenhouse gases, however, observatories should be located at low altitude. In addition, the existing NDSC sites are optimized for observations of the midinfrared spectral region, with KBr beamsplitters, aluminum optics, and midinfrared detectors. Although many trace atmospheric constituents have fundamental vibrational-rotational absorptions in the midinfrared spectral region, the near-infrared spectral region is a better choice for measuring CO2 and other greenhouse gases.
 The Total Carbon Column Observing Network is a new network of ground-based FTS sites. We describe the first dedicated laboratory in this network. This is an automated FTS observatory developed for highly precise, ground-based solar absorption spectrometry in the near-infrared spectral region. Atmospheric column abundances of CO2, CH4, CO, N2O, H2O, HDO, O2, and HF can be retrieved from the observed near-infrared spectral region. The observatory was assembled in Pasadena, California, and then deployed to Park Falls, Wisconsin, during May 2004. We compare the column CO2 results with integrated in situ aircraft profiles, and present the CO2 column values for May 2004 to October 2005. Readers interested in these results may wish to skip the detailed instrumental description in section 2 and proceed directly to section 3.
2.1. Bruker 125HR Spectrometer
 Solar spectra are acquired at high spectral resolution using a Bruker 125HR FTS housed in a custom laboratory (Figure 1). The Bruker 125HR has been substantially improved over its predecessor, the Bruker 120HR. One important improvement is the implementation of the interferogram sampling method described by Brault , that takes advantage of modern 24-bit delta-sigma analog-digital converters to improve the signal-to-noise ratio.
 The spectrometer described here has been optimized for measurements in the near-infrared spectral region, with gold-coated optics and a CaF2 beamsplitter with broadband coating. Interferograms are simultaneously recorded by two uncooled detectors. Complete spectral coverage from 3800 to 15,500 cm−1 is obtained by simultaneous use of an InGaAs detector (3800–12,000 cm−1) and Si-diode detector (9500–30,000 cm−1) in dual-acquisition mode, with a dichroic optic (Omega Optical, 10,000 cm−1 cut-on). A filter (Oriel Instruments 59523; 15,500 cm−1 cut-off) prior to the Si diode detector blocks visible light, which would otherwise be aliased into near-infrared spectral domain. The observed spectral region includes absorption bands of CO2, CH4, CO, N2O, H2O, HDO, O2, and HF. Spectra from the Si-diode detector are not analyzed in this work, but are useful for comparison with OCO and other future satellite instruments measuring the b1Σg+(v = 0) ← X3Σg−(v = 0) O2 transition (A-band) between 12,950 and 13,170 cm−1. For the spectra obtained here, we use a 45 cm optical path difference and a 2.4 mrad field of view, yielding an instrument line shape that has a full-width at half-maximum of 0.014 cm−1. This is sufficient to fully resolve individual absorption lines in the near-infrared. The input optics use an off-axis parabolic mirror that is the same type as the collimating mirror. Hence the external field of view is also 2.4 mrad, and the instrument accepts only a small fraction of the 9.4 mrad solar disk. The beam diameter is stopped down to 3.5 cm to reduce the saturation of the detectors and signal amplifiers. Figure 2a shows a typical spectrum, acquired in 110 s, with signal-to-noise ratios of ∼900:1 and ∼500:1 for the InGaAs and Si diode detectors, respectively. The observed intensity is the product of the solar Planck function with the instrument response.
 To maintain stability of the optical alignment, the internal temperature of the spectrometer is controlled between 28 and 30°C. To reduce acoustic noise and eliminate refractive inhomogeneities, the internal pressure is maintained at less than 2 hPa using a Varian TriScroll 300 scroll pump. The spectrometer is evacuated once per day, before sunrise, and has a leak rate less than 2 hPa day−1. The instrument line shape is monitored using narrow HCl lines in the first overtone band (ν0 = 5882 cm−1). A 10 cm cell with 30′ wedged Infrasil windows containing 5.1 hPa of HCl gas is permanently mounted in the source compartment, prior to the entrance field stop wheel, as shown in Figure 1b. Because of space constraints, the sample compartment typically supplied with the 125HR is not used.
2.2. Laboratory and Other Instrumentation
 The 125HR spectrometer is mounted inside a modified 6.1 × 2.4 × 2.6 m steel shipping container. The container is equipped with an air conditioning and heating wall unit, power, lights, and telephone connection. The interior of the container is insulated with 9.0 cm of R19 fiberglass covered with 0.32 cm thick aluminum sheet. These materials were chosen to minimize outgassing that may otherwise interfere with spectral observations.
 The optical assembly (solar tracker) that transfers the direct solar beam from the roof of the container to the FTS was purchased from Bruker Optics. It consists of a servo-controlled assembly with two gold-coated mirrors that rotate in azimuth (0–310°) and elevation (−5 to +90°). The solar tracker has two operational modes: pointing to the calculated solar position and active servo-controlled tracking. Initially, the solar tracker is pointed toward the calculated solar ephemeris. This position is typically within 0.3° of the actual solar position, with error attributed to the alignment and leveling of the solar tracker. The solar beam is directed down through a hole in the laboratory roof, which is sealed with an 11.5 cm diameter wedged CaF2 window. Inside the laboratory, a small fraction of the incoming solar beam is focused onto a Si quadrant detector located at the entrance to the spectrometer. The solar tracker then uses the quadrant detector signal for active tracking of the Sun, with a manufacturer-specified tracking accuracy of 0.6 mrad. Three heaters on the base of the solar tracker activate when the temperature drops below 5°C, to prevent damage to the optics and electronics.
 The solar tracker is housed in a fiberglass telescope dome, manufactured by Technical Innovations, Inc. in Barnesville, Maryland. The dome is constructed of industrial grade fiberglass, with a 1.0 m × 1.3 m oval base. The wide shutter opening allows unobstructed views from the horizon to 5° beyond zenith. The dome is bolted to the container roof, which is reinforced with eight 6.4 cm thick steel tubes welded to the frame, and covered with a 0.64 cm thick steel plate. This stabilizes the solar tracker and prevents vibrations that may degrade spectral quality and flexing of the container roof which may degrade the solar tracker alignment.
 A Setra Systems, Inc. Model 270 pressure transducer (±0.3 hPa), is mounted inside the container, with an input tube at ∼2 m outside. Accurate knowledge of the pressure is important for evaluating the accuracy of the retrieved O2 columns. In addition, synoptic surface pressure variations of ±10 hPa (±1%) would overwhelm the changes in the total CO2 column that we wish to observe. The calibration of the pressure sensor is checked periodically by comparison to a Fortin mercury manometer (Princo Instruments, Model 453) mounted in the laboratory as an absolute standard. In addition, the temperature of the Setra pressure transducer is monitored for evidence of bias. A weather station mounted at ∼5 m includes sensors for air temperature (±0.3°C), relative humidity (±3%), solar radiation (±5% under daylight spectrum conditions), wind speed (±0.5 m s−1), wind direction (±5°), and the presence of rain.
 A small network camera (Stardot Technologies) with a fisheye lens (2.6 mm focal length) is positioned on the roof of the laboratory. The dome, solar tracker, weather station, and a wide view of the sky are visible within the field of view of the camera. This allows us to remotely monitor the operation of the equipment and verify weather conditions.
 Accurate knowledge of the time is critical in calculating the solar zenith angle (SZA), which is necessary to convert retrieved atmospheric slant column abundances into vertical column abundances. We use a high-precision GPS satellite receiver with a network time server (Masterclock NTP100-GPS) to maintain time synchronization of the Bruker 125HR.
2.3. Data Acquisition and Instrumental Automation
 The laboratory equipment consists of the 125HR spectrometer, scroll pump, solar tracker, dome, weather station, NTP-GPS satellite time receiver, network camera, heaters (for 125HR, solar tracker, and scroll pump), temperature sensors, current and voltage sensors, and uninterruptible power supply (UPS). Each of these components is monitored and/or controlled with an integrated CPU board (Hercules, Diamond Systems) and an additional custom-built control board. The Hercules board includes four serial ports, used for communication with the solar tracker, dome, weather station, and modem. The Hercules board also includes 32 wide-range analog inputs for monitoring temperatures, voltage, currents, and the pressure of the scroll pump. Five digital I/O lines of the Hercules board are used to command power to the solar tracker, dome, modem, FTS, and the FTS reset line. The FTS, network camera, NTP-GPS satellite time receiver, and UPS are IP-addressable and are commanded within the local area network.
 The operating system chosen for the Hercules computer is QNX (QNX, Kanata, Ontario), a realtime, multitasking, multiuser, POSIX-compliant operating system for the Intel family of microprocessors. QNX was selected because of its stability and because its simple message-passing method of interprocess communication allows the acquisition and control functions of the data acquisition software to be separated into a number of logically discrete processes.
 Throughout the night, the acquisition software records weather and housekeeping data. When the calculated solar elevation angle reaches 0°, the scroll pump is commanded on and the FTS is evacuated to 0.5 hPa. Following the pumping sequence, the dome opens and the solar tracker points to the calculated solar ephemeris. If the solar intensity is sufficient (45 W m−2), the solar tracker begins active tracking of the Sun and the FTS begins acquisition of solar interferograms. The specific acquisition parameters, including the field stop diameter, detector gains, scanner velocity, and optical path difference, are set in software. Typically, each scan requires 110 s to complete and consists of a single-sided interferogram with 45 cm optical path difference recorded at 7.5 kHz laser fringe rate. Forward and reverse interferograms (with the moving mirror traveling away from and toward the fixed mirror) are acquired in sequence. Throughout each scan, the solar intensity measured by the solar tracker quadrant sensor is recorded at 0.5 Hz. Since only spectra acquired under stable solar intensity are suitable for atmospheric retrievals, the standard deviation of the solar intensity is later used to evaluate spectral quality. Forward and reverse interferograms are analyzed separately to maximize the number of unobstructed scans. Acquisition of solar interferograms continues as long as the solar intensity is sufficient for active tracking of the Sun. If the weather station detects rain, then the dome closes and spectral acquisition ceases until weather conditions improve. When the calculated solar elevation reaches 0° at the end of the day, the dome is closed.
 Each night, interferograms recorded during the day are copied onto a removable hard disk. Overnight analysis software performs a Fourier transform to produce spectra from the interferograms, and performs preliminary atmospheric column retrievals. These results are then emailed to Pasadena to monitor performance. At two month intervals, the removable hard disk is manually replaced with an empty one. The full disk is mailed to Pasadena for analysis and archiving. This is necessary because only dial-up Internet access is available at the WLEF site. The operational data rate is ∼50 GB month−1.
3. Measurement Site
 The FTS observatory was assembled and tested in Pasadena, California, and deployed to northern Wisconsin during May 2004. The laboratory is located 25 m south of the WLEF television tower site (45.945°N, 90.273°W, 442 m above sea level) in the Chequamegon National Forest, 12 km east of Park Falls, Wisconsin (pop. 2800). The region is heavily forested with low relief, and consists of mixed northern hardwoods, aspen, and wetlands. Boreal lowland and wetland forests surround the immediate research area. The Chequamegon National Forest was extensively logged between 1860 and 1920, but has since regrown.
 This site was chosen because the National Oceanic and Atmospheric Administration Earth Systems Research Laboratory (NOAA ESRL) and other organizations conduct extensive in situ measurements at the WLEF tower, facilitating intercomparison between the column and boundary layer measurements. Monitoring began in October 1994, when WLEF was added as the second site in the Tall Tower program. CO2 concentrations are measured continuously at six levels on the 447 m tower [Zhao et al., 1997; Bakwin et al., 1995]. Fluxes of CO2, water vapor, virtual temperature, and momentum are monitored at three levels [Berger et al., 2001; Davis et al., 2003]. In addition, NOAA ESRL conducts weekly flask sampling [Komhyr et al., 1985] and monthly aircraft profiles which collect flask samples between 0.5 km and 4 km [Bakwin et al., 2003].
4. Data Analysis
 In this work, spectra are analyzed using a nonlinear least squares spectral fitting algorithm (GFIT) developed at the Jet Propulsion Laboratory. Atmospheric absorption coefficients are calculated line-by-line for each gas in a chosen spectral window, and are used together with the assumed temperature, pressure, and VMR profile in the forward model to calculate the atmospheric transmittance spectrum. This is compared with the measured spectrum and the VMR profiles are iteratively scaled to minimize the RMS differences between the calculated and measured spectra. The theoretical instrument line shape, verified from fits to low-pressure HCl gas cell lines, is used in calculating the forward model. Figure 2b shows a measured spectrum and the fitted result, for a region with strong CO2 lines.
 The atmosphere is represented by 70 levels in the forward model calculation. Pressure- and temperature-dependent absorption coefficients are computed for each absorption line at each level. Profiles of temperature and geopotential height are obtained from the NOAA Climate Diagnostics Center (CDC), with 17 pressure levels from 1000 to 10 hPa and 1° × 1° geographic resolution. At pressures less than 10 hPa, climatological profiles of temperature and geopotential height are used. Measured surface pressure is used to define the lowest model level.
 We retrieve CO2 and O2 in three bands: O2 0–0 a1Δg−X3Σg− (ν0 = 7882 cm−1); CO2 (14°1) – (00°0) (ν0 = 6228 cm−1); and CO2 (21°2) – (00°0) (ν0 = 6348 cm−1). These will be referred to as the O2 7882 cm−1, CO2 6228 cm−1, and CO2 6348 cm−1 bands. Retrievals in these three bands require accurate spectroscopic parameters for O2, CO2, H2O, and solar lines. The HITRAN 2004 line list parameters [Rothman et al., 2005] were found to be deficient at the high accuracies that we require. In HITRAN 2004, the O2 7882 cm−1 band has severe errors in strengths for low J lines and errors in widths for high J lines; the CO2 6228 cm−1 and 6348 cm−1 bands have errors in line positions, air-broadened widths, and pressure shifts.
 We have adopted improved line parameters for the O2 7882 cm−1 retrievals, including line strengths from PGOPHER model results [Newman et al., 2000], air-broadened widths [Yang et al., 2005], and temperature-dependent air-broadened widths [Yang et al., 2005]. In addition, we have made two empirical corrections to minimize temperature and air mass dependence of the O2 retrieval: (1) The air-broadened width values [Yang et al., 2005] have been increased by 1.5%. (2) The temperature dependence of the air-broadened width values [Yang et al., 2005] have been increased by 10% to bring them into better agreement with measurements by Newman et al. . Both of these empirical corrections are within the reported measurements uncertainties. Four recent laboratory studies report the integrated O16O16 7882 cm−1 band strength as 3.166 ± 0.069 × 10−24 cm molecule−1 [Lafferty et al., 1998], 3.10 ± 0.10 × 10−24 cm molecule−1 [Newman et al., 1999] (all O2 isotopes), 3.247 ± 0.080 × 10−24 cm molecule−1 [Cheah et al., 2000], and 3.210 ± 0.015 × 10−24 cm molecule−1 [Newman et al., 2000]. Because the Newman et al.  PGOPHER model shows good agreement with our atmospheric fitting retrievals, we have also adopted the Newman et al.  integrated band strength.
 In addition to the discrete lines of the O2 7882 cm−1 band, there is an underlying continuum absorption caused by collision-induced absorption. On the basis of laboratory measurements [Smith and Newnham, 2000; Smith et al., 2001], we generated a pseudo-linelist representing collision-induced absorption which includes separate contributions from O2-O2 and O2-N2 collisions. Although the collision-induced absorption is included in the line-by-line calculation to improve estimation of the continuum, only the discrete 7882 cm−1 O2 lines are used in the computation of the O2 column amount.
 We have used updated line parameters for the CO2 6228 cm−1 and 6348 cm−1 band line strengths, air-broadened widths, and pressure shifts based on recent work by Toth et al. [2006a, 2006b]. We have also adopted updated H2O line parameters for the 5000–7973 cm−1 region (B. Toth, private communication, 2005). These new line lists were found to give superior spectral fits to our atmospheric spectra. The solar line list for all near-infrared spectral retrievals is derived from disk-center solar spectra recorded at Kitt Peak (31.9°N, 116°W, 2.07 km).
 For O2, the assumed a priori VMR profiles are constant with altitude. For CO2, the assumed a priori VMR profiles vary seasonally in approximate agreement to model output from Olsen and Randerson . We have examined the sensitivity of the column CO2 retrieval to different reasonable a priori functions, including a profile which is constant with altitude, and found that the effect on retrieved column CO2 is ≤ 0.1%.
5. Column O2 and CO2
 The consistency between retrieved column O2 and measured surface pressure is an important test of instrumental stability. O2 is well mixed in the atmosphere, with a dry-air VMR of 0.2095. This provides an internal standard that can be used to check the short-term and long-term precision of the FTS column retrievals. As described in section 3, surface pressure at the Park Falls site is recorded at 1 Hz using a calibrated Setra 270 pressure sensor. The calibrated accuracy of this sensor is ∼0.3 mb, which corresponds to an uncertainty of ∼0.03% in the surface pressure. For the May 2004 to October 2005 spectra, retrieved column O2 is consistently 2.27 ± 0.25% higher than the dry pressure column (where the dry pressure column is equal to the observed surface pressure converted to a column density minus the retrieved H2O column). This error exceeds both the uncertainty in the dry pressure column and the reported 0.5% uncertainty in the integrated O16O16 7882 cm−1 band strength of 3.21 × 10−24 cm molecule−1 ± 0.015 × 10−24 cm molecule−1 [Newman et al., 2000]. However, the ∼4% spread in recent measurements of the integrated O16O16 7882 cm−1 band strength (section 4) suggests that this discrepancy may fall within the uncertainty of the laboratory measurements. In this analysis, the retrieved O2 columns have been reduced by 2.27% to bring the retrievals into agreement with the known atmospheric concentration of O2. Figure 3a shows O2 retrievals for air masses between 2 and 3 (SZA 60–70°) plotted as a function of the dry pressure column. Throughout this work, “air mass” refers to the ratio of the slant column to the vertical column and is approximately equal to the secant of the SZA; when the Sun is directly overhead, the SZA is 0° and the air mass is 1.0. The residuals are shown in the upper panel of Figure 3a.
Figure 3b shows the time series of O2 VMR, calculated from column O2/dry pressure column. Results are not shown for 8 May 2005 to 14 July 2005, because of an instrumental error in solar pointing. Much of the scatter in Figure 3b can be attributed to error in the line strengths and the air-broadened widths that cause the O2 retrievals to vary with temperature and air mass. However, the systematic increase in O2 VMR over time (∼0.3%) is larger than (and of opposite sign to) the seasonal changes in O2 VMR. Coincident changes in HCl concentration retrieved from the calibration cell are also observed. During May 2004 to October 2005, the reflectivity of the gold-coated solar tracker mirrors slowly degraded because of a manufacturing flaw. This reduced the measured solar intensity by approximately 60% in the near-infrared spectral region. It is possible that the errors observed in the O2 and HCl retrievals may be caused by this signal loss, coupled with nonlinearity in the response of the InGaAs detector. Studies are underway to quantify this error and remove its influence on the retrievals.
 Column retrievals of CO2 from the 6228 cm−1 and 6348 cm−1 bands show high precision and repeatability. Observations of column CO2 during one clear day and one partly cloudy day in August 2004 are shown in Figure 4a. Figure 4b shows the column O2 retrievals during the same time period. Spectra have been discarded as obstructed by clouds if the solar intensity measured by the quadrant detector fluctuated by more than 5% RMS during the recording of an interferogram. The mean and standard deviation of the CO2 columns measured during a 1-hour clear observation period around local noon (24 individual spectra) on 14 August 2004 is 7.7235 ± 0.0078 × 1021 molecules cm−2 and 7.7406 ± 0.0074 × 1021 molecules cm−2 respectively for the 6228 cm−1 and 6348 cm−1 CO2 bands. This precision of ∼0.1% is typical for column CO2 obtained under clear sky conditions in Park Falls. However, the 6228 cm−1 and 6348 cm−1 band CO2 retrievals differ by ∼0.2% in absolute column CO2. This is attributed to errors in spectroscopic parameters for line strengths and air-broadened line widths. Observations on 15 August 2004 during partly cloudy conditions show greater variability in Figures 4a and 4b, even after filtering for the standard deviation of the solar radiance to remove spectra that are significantly affected by clouds. Column-average CO2 VMR can be calculated from retrieved CO2 column, according to
There are two methods for calculating the total dry column:
where Ps is surface pressure, m is mean molecular mass, and g is the density-weighted gravitational acceleration. In Park Falls, the column H2O correction in (2) is a maximum of 0.6%.
 Using (3) will improve the precision of the column-average CO2 VMR (fCO2) if scatter in the column abundances is common to both the CO2 and O2. Common scatter could arise from errors in the spectra, such as instrumental line shape or detector nonlinearity, or from errors in the calculated slant path due to uncertainty in the surface pressure or SZA. However, dividing by column O2 will increase the random scatter (since column O2 is typically noisier than Ps) and will introduce spectroscopic line list errors from the O2 region, such as temperature and air mass dependence, into the column-average CO2 VMR. In addition, the systematic changes in column O2 observed over time in Figure 3b are likely due to detector nonlinearity. However, this systematic error is expected to affect the CO2 and O2 column retrievals similarly, and can be eliminated from the column-average CO2 VMR by using (1) and (3).
 Column-average CO2 VMR calculated via (2) and (3) is shown in Figures 4c and 4d. Comparing Figures 4c and 4d, the greatest improvement in scatter is seen on the partly cloudy day (15 August 2004). The major sources of scatter on cloudy days are error in the solar pointing and variation in intensity during the scan, which affect the CO2 and O2 retrievals similarly. Dividing column CO2 by column O2, rather than dry pressure column, therefore improves the precision, especially on partly cloudy days. A comparison of the results is shown in Table 1.
Table 1. Mean and Standard Deviation During 1-Hour Observational Period Around Local Noon
Column CO2, 1021 molecules cm−2
Column CO2/Dry Surface Pressure, ppmv
0.2095 × Column CO2/Column O2, ppmv
Clear day (14 Aug 2004) 6228 cm−1 band
7.7235 ± 0.0078
376.46 ± 0.30
376.55 ± 0.26
Clear day (14 Aug 2004) 6348 cm−1 band
7.7406 ± 0.0074
377.29 ± 0.28
377.38 ± 0.22
Cloudy day (15 Aug 2004) 6228 cm−1 band
7.707 ± 0.058
375.8 ± 2.8
375.48 ± 0.82
Cloudy day (15 Aug 2004) 6348 cm−1 band
7.724 ± 0.055
376.7 ± 2.7
376.35 ± 0.68
6. Comparison of FTS Column and Integrated Aircraft Profiles
 The Intercontinental Chemical Transport Experiment–North America (INTEX-NA) and CO2 Boundary Layer Regional Airborne Experiment (COBRA) campaigns provided an opportunity to calibrate the column CO2 measurements on an absolute scale relative to the standardized network of in situ measurements. As the difference between CO2 6228 cm−1 and 6348 cm−1 column retrievals in Figure 4a demonstrates, results from each of the CO2 bands are precise, but not sufficiently accurate. This is attributed to remaining limitations in the available spectroscopic parameters.
 The NASA DC-8 and University of Wyoming King Air measured in situ CO2 during profiles over the WLEF Tall Tower site during summer 2004, using well-calibrated, mature in situ CO2 sensors. On board the DC-8, dry CO2 VMR was measured at 1 Hz using a modified LI-COR model 6252 infrared gas analyzer [Vay et al., 2003; Anderson et al., 1996]. In-flight calibrations were performed at 15 min intervals using standards traceable to the WMO Central CO2 Laboratory. On board the King Air, similar 1 Hz measurements were performed using a modified LI-COR model 6251 [Daube et al., 2002]. In-flight calibrations were performed with standards traceable to the Carbon Dioxide Research Group at the Scripps Institute of Oceanography and NOAA ESRL. In-flight calibrations show that the typical long-term flight-to-flight precision of this technique is better than ±0.1 ppmv [Daube et al., 2002].
 The aircraft CO2 profiles can be integrated with respect to pressure for direct comparison with FTS column CO2. Mathematically, this is found by combining the definition of the column integral
with the hydrostatic equation
where f is the atmospheric mixing ratio, g is gravitational acceleration, mair is the mean molecular mass, n is the number density, p is pressure, Ps is surface pressure, z is height, and Zs is the surface height. The atmospheric mixing ratio of CO2 is defined as
where fCO2,dry is the dry-air CO2 vmr, measured by in situ instruments. Combining equations (6) and (7), together with the relationship
Integrated column CO2 from (9) can be divided by the total column from (2) to yield the column-average CO2 VMR.
 Eight unique aircraft profiles were measured on five dates during 2004: 12 July, 14 July, 15 July, 14 August, and 15 August. The first profile of the series, shown in Figure 5a, was a descending spiral by the NASA DC-8 from 10.0 km to 0.7 km. Because the aircraft has a limited altitude range, it is necessary to make assumptions about CO2 and H2O in the upper troposphere and stratosphere when using (9) to find integrated column CO2. The tropopause pressure is determined from the NOAA CDC assimilated temperature profile. The median CO2 value measured in the free troposphere is assumed to extend from the aircraft ceiling to the tropopause. Above the tropopause, the assumed CO2 profile is taken from an in situ balloon profile (35°N, 104°W) recorded over Fort Sumner, New Mexico, during September 2004. The balloon profile of CO2 as a function of altitude is coordinate-transformed into CO2 as a function of potential temperature (θ), using simultaneous temperature and pressure measurements. For the aircraft profile, θ is calculated from the NOAA CDC assimilated temperature data. CO2 is assumed to be well mixed in the planetary boundary layer between the surface and the 0.7 km floor of the aircraft profile. This is confirmed by in situ measurements on the Tall Tower. The CO2 profile shown in Figure 5b is integrated with respect to pressure to find column CO2. The assumed CO2 profile above the aircraft ceiling contributes the greatest uncertainty to the integration, and we have attributed a generous uncertainty of ±2 ppmv to this portion of the profile.
Figure 5d shows the FTS column-average CO2 recorded during a 2-hour period which brackets the aircraft profile. These profiles were performed at air mass 1.1–2.0 (SZA 25–60°). The column data are not continuous because intermittent cloud prevented the acquisition of solar spectra. The 45-min period of the aircraft profile is indicated. The averaging kernel for the FTS CO2 retrievals during this period is shown in Figure 5c. The shape of the averaging kernel is typical for a uniformly mixed, moderately strong absorber fitted by a nonlinear least squares profile-scaling retrieval. To accurately compare the FTS column-average CO2 and integrated aircraft profile, it is necessary to weight the aircraft profile by the FTS averaging kernel [Rodgers and Connor, 2003]. Because the averaging kernel varies slightly with air mass, a separate averaging kernel is calculated for each aircraft overpass.
 Comparison of the eight integrated aircraft profiles with the FTS CO2 columns is shown in Figure 6. There is a linear relationship between the integrated aircraft columns and the retrieved FTS columns. The slope relationships differ for the two CO2 bands, with values of 1.0216 and 1.0240 for CO2 6228 cm−1 and CO2 6348 cm−1 respectively. The standard deviation of the fitting residuals is 0.39 ppmv and 0.42 ppmv for the two bands. The slope relationships from Figure 6 can be used to correct the FTS CO2 columns, bringing them into absolute agreement with the calibrated in situ network.
7. Error Analysis for Column-Average CO2 VMR
 The column-average CO2 VMR calculated according to 0.2095 × column CO2/column O2 is affected by three main sources of error:
 1. The first source of error is measurement precision. As discussed in section 5, the standard deviation of column CO2/column O2 during a 1-hour period is better than 0.1% under clear sky conditions and ∼0.2% under partly cloudy conditions. Repeatability of the measurement is not a significant source of error.
 2. The second source of error is spectroscopic errors. As discussed in section 5, the retrieved O2 columns were reduced by 2.27% to bring them into agreement with the dry surface pressure. Although this correction falls outside the reported uncertainty of the 16O16O 7882 cm−1 integrated band strength, we believe that it is likely attributed to an error in the line strengths or air-broadened width parameters.
 The absolute accuracy of the CO2 retrievals was calibrated by comparison to integrated aircraft profiles, resulting in a correction of 1.0216 and 1.0240 for the CO2 6228 cm−1 and 6348 cm−1 bands. The standard deviation of the fitting residuals is 0.39 ppmv and 0.42 ppmv, or approximately 0.1%. The aircraft profiles were performed with the Sun at air mass 1.1–2.0 (SZA 25–0°), and the column-average CO2 VMR is now well calibrated for these values. However, this does not calibrate the column-average CO2 VMRs at higher air mass. A 1% change in the air-broadened widths results in a CO2 VMR change of ∼0.2% (±0.8 ppmv) at air mass 3 and ∼0.6% (±2.3 ppmv) at air mass 12. These parameters are not sufficiently constrained by current spectroscopic line lists, leaving this as a significant source of systematic error which can be correlated with air mass, time of day, and temperature.
 3. The third source of error is systematic instrumental changes over time. As described in section 5, retrieved O2 VMR increased by ∼0.3% during the observation period. Because this increase is seen for O2 retrievals from the InGaAs detector in the 7882 cm−1 band, and not for O2 retrievals from the Si diode detector in the 13095 cm−1 A-band, we believe that this is due to detector nonlinearity and can be corrected. We expect that this error affects the CO2 and O2 retrievals similarly, but for now assume that the column CO2/column O2 ratio may also have a systematic error of 0.3% over the observation period.
 The measurement precision of ∼0.1% under clear sky and ∼0.2% under partly cloudy conditions does not affect the accuracy of the measurements. However, spectroscopic errors introduce a systematic bias which depends on air mass. We have calibrated the FTS column retrievals at air mass 1.1–2.0 during July to August 2004, and expect that the absolute accuracy at these air masses has been maintained within 0.3% throughout the subsequent data record.
8. Column-Average CO2 VMR During May 2004 to October 2005
 Applying the slope corrections from section 6 allows the FTS column-average CO2 VMR to be compared directly to in situ CO2 measurements. Column-average CO2 VMR, corrected in this manner, is shown in Figure 7a, together with in situ CO2 measurements from 30 m and 396 m on the Tall Tower. The in situ CO2 measurements are influenced by the diurnal rectifier effect, which is caused by the overnight decrease in the height of the planetary boundary layer. During the day, CO2 surface fluxes are diluted within a thicker boundary layer, while CO2 surface fluxes at night are concentrated near the surface. The column-average CO2 VMR is minimally influenced by the diurnal rectifier effect. Summertime drawdown in CO2 is observed in both the in situ and column measurements.
 The seasonal cycle of column-average CO2 VMR observed at Park Falls during May 2004 to October 2005 is shown as daily averages for air masses between 2 and 4 (SZA 60–75°) in Figure 7b. In situ CO2 measurements from the Tall Tower are shown as daily averages between 1600–2000 UT (1000–1400 CST). As expected, the variation of CO2 is muted in the column, as compared to surface measurements, on all timescales. During May 2004 to May 2005, the observed peak-to-peak variation of column-average CO2 VMR is approximately 13 ppmv, with an average value of 376.2 ppmv. Comparing column-average CO2 retrievals observed in September and October during 2004 and 2005, we calculate a secular increase of 1.8 ppmv yr−1. After accounting for this, we infer a peak-to-peak seasonal amplitude of 11 ppmv for Park Falls. These results are higher than model results by Olsen and Randerson , which predict a mean seasonal column CO2 amplitude of 7–8 ppmv in Wisconsin. This difference could potentially arise from an error in the model predictions, because of uncertainty in the specifications of surface fluxes or errors in the parameterization of mixing. Alternatively, the difference could be caused by differences between the assumed meteorology and emission inventories included in the MATCH model.
 We have deployed an automated solar observatory to Park Falls, Wisconsin. Near-infrared solar absorption spectra have been acquired continuously since May 2004. Short-term and long-term precision are evaluated by the repeatability of column retrievals within a day and by the comparison of column O2 with surface pressure measurements. The precision of retrieved column CO2 under clear-sky conditions is ∼0.1%, as determined by the 1σ variability of retrievals recorded within 1 hour. Under partly cloudy conditions, the CO2 column precision is much worse, but can be improved by dividing column CO2 by column O2 to calculate column-average CO2 VMR. This calculation eliminates errors which are common to both CO2 and O2 retrievals, such as errors in solar pointing and variation in solar intensity during interferogram acquisition, and allows useful retrievals to be obtained under partly cloudy conditions. Comparison of retrieved column O2 to dry surface pressure during May 2004 to October 2005 shows linear agreement with a 2.27 ± 0.25% bias.
 The column CO2 retrievals were calibrated using aircraft profiles from the INTEX-NA and COBRA campaigns during summer 2004. The CO2 6228 cm−1 and CO2 6348 cm−1 band retrievals overestimate the integrated aircraft profiles by factors of 1.0216 and 1.0240 respectively, with standard deviation of the fitting residuals of 0.39 ppmv and 0.42 ppmv. The systematic differences are attributed to known uncertainty in the CO2 spectroscopic line strengths and air-broadened width parameters. The comparison to aircraft integrated columns allows the CO2 6228 cm−1 and CO2 6348 cm−1 retrievals to be corrected to the accepted in situ calibration scale. The aircraft profiles were performed with the Sun at air mass 1.1–2.0, and we are confident that our column-average CO2 VMRs are now well calibrated for these summertime, low air mass values. After calibration of the column retrievals with the integrated aircraft profiles and consideration of the complete error budget, we calculate the uncertainty in retrieved column-average CO2 VMR to be ∼0.3% (±1.1 ppmv) at air masses less than 2 (SZA less than 60°) throughout the measurement time series.
 We thank Jeffrey Ayers for maintaining the ground-based FTS laboratory in Park Falls, Wisconsin. We thank Arlyn Andrews and the NOAA CCGG for providing WLEF Tall Tower CO2 measurements. Bruce Daube thanks Victoria Chow and Bhaswar Sen for their support in obtaining the balloon CO2 profile. We thank Andrew Orr-Ewing for helpful discussions and providing PGOPHER model results for the O2 7882 cm−1 band. R.A.W. acknowledges support from the National Science Foundation and the California Institute of Technology. This work was funded by NASA grant NAG5-12247 and NNG05-GD07G. Research at the Jet Propulsion Laboratory, California Institute of Technology, is performed under contract with NASA.