1. Introduction

Water vapor measurements by sun photometry using the 940-nm water vapor absorption band have been compared to in situ and other remote (e.g., microwave) measurements in several previous publications (e.g., Schmid et al. 2000, 2001, 2003a, b, 2006; Redemann et al. 2003; Livingston et al. 2000, 2003, 2007). Those comparisons were all restricted to sun photometer altitudes <∼6 km, with water vapor columns ∼0.1 to 5 g cm⁻², and water vapor densities ∼0.1 to 17 g m⁻³.

In January–February 2003, the 14-channel National Aeronautics and Space Administration (NASA) Ames airborne tracking sun photometer (AATS-14; hereafter simply AATS) flew on the DC-8 along with the NASA Langley/Ames Diode Laser Hygrometer (DLH). The flights were part of the second Stratospheric Aerosol and Gas Experiment (SAGE) III Ozone Loss and Validation Experiment (SOLVE II). They provided an opportunity to test AATS water vapor measurements in higher, drier environments, including altitudes up to 12 km and water vapor columns ∼0.002 to 0.1 g cm⁻² above 4 km. The availability of DLH measurements on the same aircraft as AATS provided an especially good comparison opportunity because the DLH was designed and built to perform well in environments this high and dry, and previous DLH measurements had been compared to other state-of-the-art water vapor measurements in such regions (e.g., Podolske et al. 2003). To our knowledge there had not been any previous tests of suborbital spectroscopic water vapor measurements using the 940-nm band in conditions so high and dry.

It should be noted, however, that several satellite instruments that view the sun through the earth’s atmospheric limb do use the 940-nm band for water vapor measurements, and they have been validated. These instruments include the SAGE and Polar Ozone and Aerosol Measurement (POAM) families of sensors, validations of which have been published by Nedoluha et al. (2002), Taha et al. (2004), Thomason et al. (2004), and Lumpe et al. (2006), among others. The SAGE and POAM measurements benefit from the long viewing path of their limb-viewing geometry [e.g., ∼200 km in a 1-km-thick atmospheric shell, resulting from a local solar zenith angle (SZA) of ∼90°], which produces measurable absorption in the 940-nm band even for stratospheric concentrations of water vapor (typically 3–8 ppmv). The AATS DC-8 measurements reported in this paper had true solar zenith angles ranging from 68.6° to 89.1°, which includes viewing paths (hence airmass factors) considerably less than those for the SAGE and POAM viewing geometries.

2. Instruments and data analysis techniques

a. Fourteen-channel Ames airborne tracking sun photometer

The 14-channel NASA Ames airborne tracking sun photometer has been described previously in the literature (e.g., Russell et al. 2005, 2007), so we give only a brief synopsis here. The AATS measures the direct beam solar transmission in 14 channels with center wavelengths from 354 to 2138 nm, including a channel centered at 941 nm. Azimuth and elevation motors rotate a tracking head to lock on to the solar beam and maintain detectors normal to it.

The AATS channel wavelengths are chosen to permit separation of aerosol, water vapor, and ozone transmission along the measured slant path. Our methods for data reduction, calibration, and error analysis have been described in detail previously (Russell et al. 1993a, b; Schmid and Wehrli 1995; Schmid et al. 1996, 2001, 2003b). Water vapor analysis methods are briefly reviewed below. Results for AATS aerosol optical depth and ozone measurements from the DC-8 in SOLVE II are described by Russell et al. (2005) and Livingston et al. (2005).

AATS was calibrated by analysis of sunrise measurements acquired at Mauna Loa Observatory (MLO), Hawaii, for six sunrises in November 2002 prior to SOLVE II and for seven sunrises in March 2003 after SOLVE II. Exoatmospheric detector voltages V₀ were derived using the Langley plot technique (e.g., Russell et al. 1993a, b; Schmid and Wehrli 1995) for all channels except 941 nm, for which a modified Langley technique (Reagan et al. 1995; Michalsky et al. 1995; Schmid et al. 1996, 2001) was employed to account for water vapor absorption.

Because absorption by water vapor varies strongly within the 5-nm FWHM bandpass of the AATS channel centered at 941 nm, the usual Beer–Lambert–Bouguer expression must be modified to describe correctly the relationship between the output detector voltage V(941) and the atmospheric attenuators on the sun-to-instrument path. In particular,

View Expanded

where V₀(941 nm) is the exoatmospheric calibration voltage, d is the earth–sun distance in astronomical units at the time of observation, m_i is the airmass factor (ratio of slant path optical depth to vertical optical depth) for attenuating species i (where i represents gas scattering, nonwater vapor gas absorption, or aerosol extinction), τ_i is the optical depth for the ith attenuating species other than water vapor, and T_w is the water vapor transmittance (weighted by absorption strength, source intensity, and filter function). Consistent with the approach followed by Livingston et al. (2007), we used the three-parameter expression of Ingold et al. (2000) to parameterize T_w as a function of the amount of water vapor w_s along the slant path:

View Expanded

In this expression, the coefficients a, b, and c are least squares fitting parameters. In particular, calculations were performed using the radiative transfer code LBLRTM V9.2 (Clough et al. 2005) for a variety of model atmospheres (tropical, midlatitude summer, midlatitude winter, subarctic summer, subarctic winter, and 1976 U.S. Standard Atmosphere) and a range of solar zenith angles to extend Livingston et al.’s (2007) Table 2 results to include all altitudes (maximum of ∼12.5 km) flown by the DC-8 during SOLVE II. These results are shown in Fig. 1 and the complete set of fitting coefficients is given in Table 1. We have calculated one set of fitting parameters to LBLRTM results for the composite set of model atmospheres rather than calculating fitting parameters separately for each model atmosphere because inclusion of all the model atmosphere results yields a wide range of water vapor transmittances and slant path water vapor amounts. These can then be applied to a wide range of sun photometer transmittance measurements obtained at any altitude with no extrapolation to values of water vapor transmittance and slant amount outside the range of results from LBLRTM calculations that use a single model (e.g., the subarctic winter atmosphere).

As noted by Livingston et al. (2007) and illustrated here in Fig. 1, a retrieval that ignores the instrument altitude can result in an incorrect determination of slant path (hence, column) water vapor. In particular, if it is assumed that the instrument is located at sea level, then column water vapor (CWV) would be underestimated for instrument altitudes above sea level, and the errors would be greatest for large SZAs (high airmass values), and high CWV (hence, high slant water vapor) amounts.

Column water vapor is calculated from the slant amount by dividing by an appropriate water vapor airmass factor. These airmass factors were calculated using the methodology reported in Russell et al. (2005) and Livingston et al. (2005) by assuming a water vapor vertical distribution corresponding to the subarctic winter atmospheric model for the 21 January 2003 measurements and the midlatitude winter atmospheric model for the 6 February 2003 data. The uncertainty in CWV is computed following Schmid et al. (1996). The calculated CWV values are averaged within 50-m vertical bins, and a smoothing spline is then fit to the resultant CWV profile. Water vapor density (ρ_w) is obtained as the derivative of the spline fit. This procedure has been described in detail in Schmid et al. (2000) and Livingston et al. (2007).

3. Results

Results from the first comparison profile, which was flown 21 January 2003 on a DC-8 ascent out of Kiruna, Sweden, are presented in Fig. 2. The AATS was able to view the sun at DC-8 altitudes between ∼7.7 and 10 km, thus allowing calculation of CWV at each altitude in that range. The AATS and DLH CWV profiles are overplotted in Fig. 2a, in addition to the atmospheric static temperature profile measured by a probe mounted on the DC-8. Because DLH measures local (not column) water vapor, the DLH CWV value at profile top is set equal to the AATS value there, and DLH CWV values below that altitude are obtained by integrating DLH local values downward. Corresponding AATS and DLH water vapor densities are overplotted in Fig. 2b. As noted in section 2a, AATS ρ_w is obtained as the vertical derivative of a spline fit to the AATS CWV profile that results after averaging the CWV values within 50-m vertical bins.

Scatterplot comparisons of AATS and DLH CWV and ρ_w for the 21 January profile are shown in Figs. 2c and 2d. The statistics shown on each scatterplot quantify the correlation and agreement between the AATS and DLH results. For CWV, the data yield r ² 0.977, rms difference 0.0004 g cm⁻² (13.0%), and bias (AATS − DLH) 0.0002 g cm⁻² (8.1%). Integration of the DLH ρ_w values over the layer bounded by the minimum (7.69 km) and maximum (9.97 km) altitudes for which AATS CWV values were calculated yields layer water vapor (LWV) of 0.0038 g cm⁻², which is 24% drier than the corresponding AATS LWV value, 0.0050 g cm⁻². Here, we note that if the absolute rms difference and bias in CWV (or, equivalently, in LWV) are expressed as percentages of LWV instead of CWV, the relative values increase (rms difference: 31.1%; bias: 19.6%) because LWV is less than CWV. Corresponding values for ρ_w are r ² 0.887, rms difference 92.1%, and bias (AATS − DLH) 37.3%. This agreement is noteworthy in light of the very dry conditions: maximum AATS CWV of 0.007 g cm⁻² and maximum ρ_w of 0.1 g m⁻³. Table 2 summarizes the AATS and DLH results for the two SOLVE II profiles and includes ρ_w results from three other campaigns in which coincident AATS and in situ sensor (not DLH) water vapor measurements were acquired from the same aircraft.

Examination of Figs. 2a and 2b reveals that the AATS – DLH LWV difference of 24% (0.0012 g cm⁻²) is dominated by differences between AATS and DLH water vapor values in the lowest 0.5 km (7.7–8.2 km) of the profile. In an effort to explain these differences, we have examined the sensitivity of the AATS results to the water vapor airmass factors—that is, to the water vapor vertical distribution used to calculate these airmass factors. As noted above in section 2a, the AATS CWV values were derived using water vapor airmass factors calculated from the water vapor number density vertical distribution defined by the subarctic winter atmospheric model. The DC-8 ascent from 7.7 to 10 km covered a horizontal distance of ∼160 km in 0.25 h. During the ascent the true SZA decreased from 89.08° to 88.56°, for which water vapor airmass factors calculated from the subarctic winter atmosphere model decrease from 35.9 to 25.9, although the airmass factor between 7.7 and 8.2 km only decreases from 35.9 to 33.5. In fact, DLH measurements are available between 2.0 km and 10 km and have been combined with the subarctic winter atmospheric model below and above the DLH profile to construct a composite water vapor profile, for which a new set of water vapor airmass factors has been derived. Application of these airmass factors to the AATS water vapor transmittances decreases the AATS LWV from 0.0050 to 0.0043 g cm⁻², in much better agreement with the DLH LWV value of 0.0038 g cm⁻², but the new AATS CWV profile (not shown) still falls within the AATS CWV error band shown in Fig. 2. Use of these air masses yields corresponding decreases in AATS − DLH CWV and ρ_w bias and rms differences. In particular, the AATS − DLH CWV bias decreases from 8.1% to 1.5%, and the rms difference decreases from 13.0% to 7.4%. The AATS − DLH ρ_w bias decreases from 37.3% to 19.1%, and the rms difference decreases from 92.1% to 76.2%. However, the general shape of the AATS ρ_w profile does not change, and the divergence between the AATS and DLH ρ_w values in the altitude range between 7.7 and 8.2 km is only marginally reduced. We are left to conclude that these differences are real and are due to water vapor spatiotemporal inhomogeneity. Although this exercise is useful for providing some measure of the sensitivity of the 21 January AATS CWV retrieval to air mass (i.e., to the water vapor vertical distribution), it is not practical for calculating real time AATS ρ_w profiles because, typically, no a priori information on the vertical distribution of water vapor is available.

Figure 3 shows results from the second comparison profile, which was flown 6 February 2003 as the DC-8 descended into Edwards Air Force Base, California, on its return from Sweden. In this case, AATS was able to view the sun at DC-8 altitudes from ∼12.5 km to 1.1 km. This profile is noteworthy for combining very dry conditions above ∼5.5 km (similar to those in the 21 January case; cf. Fig. 2) with significantly increased CWV and ρ_w as well as strong vertical structure below ∼5.5 km (although values are still small compared to the previous AATS comparisons cited in section 1). Another significant feature of this profile is the availability of water vapor measurements from a Sippican II radiosonde released from Edwards at 1446 UTC (1.75 h before the DC-8 began its descent). Figures 3a and 3b overplot the results for the entire range of measurement altitudes, Figs. 3c and 3d expand the results for altitudes between 6.0 and 12.5 km, and Figs. 3e and 3f show results between 1.1 and 6.0 km. We note that the spline-fitting function was adjusted to smooth the AATS CWV values above 4.5 km (i.e., for relatively low values of CWV) more than those below 4.5 km to minimize the effect on the calculated AATS water vapor density profile due to spatial and/or temporal water vapor variability (as manifested by increasing CWV with altitude for selected altitudes above 8.5 km) along the AATS-to-sun slant path.

For the entire altitude range traversed by the DC-8 during its descent on 6 February, the DLH measures LWV (0.291 g cm⁻²) that is 5% greater than the corresponding AATS LWV value (0.277 g cm⁻²). Corresponding values of AATS and DLH CWV and ρ_w for the 6 February profile are compared in scatterplots in Fig. 4 and are listed in Table 2. Results for CWV are r ² 0.998, rms difference 0.0041 g cm⁻² (6.6% CWV; 7.0% LWV), and bias (AATS − DLH) 0.0005 g cm⁻² (0.8% CWV; 1.0% LWV). Corresponding values for ρ_w are r ² 0.965, rms difference 0.068 g m⁻³ (25.2%), and bias (AATS minus DLH) −0.013 g m⁻³ (−4.8%). As can be seen in Table 2, if the analysis (not plotted separately) of the 6 February data is restricted to altitudes above 5.5 km, the CWV and ρ_w r ² values are similar to the 21 January results. The values of absolute rms difference and bias for the altitude-restricted 6 February analysis are about twice those of the 21 January profile, whereas corresponding values of relative ρ_w rms difference and bias are significantly lower than the 21 January results. For comparisons between AATS and the radiosonde, the overall agreement is good: for AATS values interpolated to sonde-reported altitudes, AATS and sonde CWV have r ² 0.998, rms difference 0.010 g cm⁻², and bias (AATS − sonde) 0.006 g cm⁻². Corresponding values for ρ_w are r ² 0.910, rms difference 0.109 g m⁻³ (34.6%), and bias (AATS − sonde) 0.026 g m⁻³ (8.3%). The maxima of the scatterplots (CWV < 0.3 g cm⁻²; ρ_w < 1.1 g m⁻³), although larger than those shown in Fig. 2, are still noteworthy for their small values. For example, in previous AATS comparisons to in situ and other remote water vapor measurements, CWV has ranged up to ∼5 g cm⁻² and ρ_w up to ∼17 g m⁻³.

As can be seen in Figs. 3d and 3f, the radiosonde measured higher values of ρ_w than did AATS or DLH between 5.5 and 7 km and in the lower stratosphere (LS) between 11 and 12 km, but lower ρ_w between 7 and 9 km and between ∼3.2 and 4.5 km. Previous studies by Wang et al. (2003), Ferrare et al. (2004), and Miloshevich et al. (2006) have found that the Sippican carbon hygristor sensor fails to respond to humidity changes at colder temperatures and thus cannot provide reliable measurements of relative humidity in the upper troposphere (UT). This may explain the higher Edwards sonde ρ_w values measured in the LS and certainly raises doubts about the validity of the data not only in the LS but throughout the UT (i.e., above ∼6 km). Also, because no radiation correction is applied to the Sippican temperature data, heating by solar radiation could lead to an incorrect temperature measurement and thus contribute to an incorrect estimate of water vapor density. However, there is no direct evidence of this effect; examination of Fig. 3a indicates that measurements of static temperature by the DC-8 sensor and by the radiosonde are virtually indistinguishable throughout most of the troposphere, except within the temperature inversion at 3.2 km, where the sonde is warmer by ∼2°C, and in the LS, where the DC-8 sensor is actually 1–2°C warmer than the sonde.

Figure 5 shows scatterplots that combine AATS and DLH results for the profiles on 21 January and 6 February. AATS and DLH CWV (Fig. 5a) have r ² 0.998, rms difference 7.2% (7.7% LWV), and bias (AATS − DLH) 0.9% (1.0% LWV). Corresponding values for ρ_w (5b) are r ² 0.968, rms difference 27.6%, and bias (AATS − DLH) −4.2%. Results for altitudes >5.5 km only are shown for CWV and ρ_w in Figs. 5c and 5d, respectively. For CWV, values are r ² 0.994, rms difference 9.5% (18.0% LWV), and bias (AATS − DLH) 5.0% (9.5% LWV). For ρ_w, values are r ² 0.825, rms difference 49.6%, and bias (AATS − DLH) 9.7%.

4. Summary and conclusions

This paper has compared values of CWV and ρ_w calculated from simultaneous measurements acquired by the AATS and DLH sensors on the DC-8 for two vertical profiles flown during SOLVE II. This study is unique because it includes data acquired at altitudes (above 6 km) where we had previously never taken measurements and where the amount of water vapor in the atmosphere was the lowest we have ever measured. To our knowledge this is the first time suborbital spectroscopic water vapor measurements using the 940-nm water vapor absorption band have been tested in conditions so high and dry.

Measurements were acquired at altitudes between 7.7 and 10 km during the 21 January aircraft ascent and between 1.1 and 12.5 km during the 6 February descent. AATS and DLH CWV values (where the DLH value was set equal to the AATS CWV value at the top of the profile) yielded r ² of 0.997 for 21 January and 0.998 for 6 February. Comparison of AATS and radiosonde CWV values for sonde measurements (normalized to AATS CWV at the top of the profile) acquired 1.75–2.15 h before the AATS measurements on 6 February yielded an r ² of 0.997. For measurements taken at altitudes >5.5 km, the composite AATS and DLH dataset gave an r ² of 0.994, an rms difference of 0.0006 g cm⁻² (9.5% CWV; 18.0% LWV), and a bias (AATS − DLH) of 0.0003 g cm⁻² (5.0% CWV; 9.6% LWV). Corresponding values for ρ_w were r ² of 0.825, an rms difference of 0.012 g m⁻³ (49.6%), and a bias (AATS − DLH) of 0.002 g m⁻³ (9.7%). The large relative rms and bias differences reflect the dry atmosphere with ρ_w < 0.1 g m⁻³ at those altitudes.

The AATS − DLH CWV (LWV) results reported for the two profiles in the current study are not directly comparable to AATS in situ sensor results obtained in previous AATS field campaigns (Schmid et al. 2003b, 2006; Livingston et al. 2007) because each of the previous studies included 26–35 separate aircraft profiles and reported r ², rms difference, and/or bias LWV statistics calculated for the ensemble of profiles. The relative ρ_w statistics may be comparable, but caution must be exercised in drawing any conclusions from such comparisons because of the much larger values of ρ_w measured in the prior studies. Nevertheless, to put the current results in the context of the previous measurements, we include in Table 2 the comparative AATS in situ sensor ρ_w statistics reported in those studies.

Most recently, Livingston et al. (2007) compared LWV and ρ_w measured by AATS and by a Vaisala HMP243 humidity sensor mounted on the same aircraft during 35 vertical profiles acquired over the Gulf of Maine, with maximum LWV ∼ 3.7 g cm⁻² and maximum ρ_w ∼ 16 g m⁻³. For the ensemble of 35 profiles, they found for ρ_w an r ² of 0.98, an rms difference of 20.3% (0.904 g m⁻³), and bias (AATS − Vaisala) −7.1% (−0.313 g m⁻³). For coincident measurements acquired with AATS and with an EdgeTech 137-C3 chilled mirror during 35 vertical profiles over the Atmospheric Radiation Measurement (ARM) southern Great Plains (SGP) site in 2003, Schmid et al. (2006) reported for ρ_w an r ² of 0.958, an rms difference of 19.8% (0.628 g m⁻³), and an AATS − mirror bias of −5%. In each of these studies, AATS was equipped with the same 941-nm interference filter used during SOLVE II. Schmid et al. (2003b) compared coincident measurements acquired with AATS (using a different 941-nm interference filter from that used in the 2003 SGP and 2004 SOLVE II campaigns) and with an EdgeTech Chilled Mirror 137-C3 and a Vaisala HMP243 mounted on the same aircraft during 36 vertical profiles during the 2003 ACE–Asia field campaign. For the AATS − mirror ρ_w, they calculated an r ² of 0.961 and an rms difference of 24.9% (0.633 g m⁻³). The AATS − Vaisala results were almost identical. When the data obtained at all altitudes for 6 February 2003 are included in the analyses, the AATS − DLH r ² values for SOLVE II equal those reported in the previous AATS studies. For data obtained on 21 January and for the composite dataset restricted to those data acquired above 5.5 km on 6 February, the AATS − DLH r ² values (0.82–0.89) are less than the earlier results. The AATS − DLH ρ_w relative rms difference and bias for 21 January are significantly higher than the previous results, as are relative rms differences for the high altitude 6 February and high altitude composite datasets. Again, about all that can be concluded is that the large relative rms and bias differences reflect the dry atmosphere at those altitudes.

We are encouraged that useful water vapor retrievals were derived from the AATS measurements on 21 January when data were acquired at high altitude for relatively high SZAs (89.08° to 88.56°). These results are consistent with successful retrievals of aerosol optical depth and columnar ozone from AATS measurements taken at high altitude and high SZAs during SOLVE II (Russell et al. 2005; Livingston et al. 2005). Although horizontal inhomogeneities are more likely to occur along long slant paths, these same long slant paths can improve signal-to-noise ratio and reduce sun photometer calibration uncertainties to allow extension into an area of low ambient water vapor density.

The comparisons presented here included a very limited set of measurements at altitudes in the 6–12 km range, and additional comparisons between sun photometer and in situ sensors are needed at these altitudes where the amount of water vapor in the atmosphere is so low to permit quantification of the uncertainty in the sun photometer-calculated ρ_w values. Nevertheless, the agreement we have found between the AATS and DLH retrievals of CWV (essentially, LWV) gives us hope that airborne sun photometer measurements can provide useful data for validation of satellite water vapor retrievals not only for the full atmospheric column and for ρ_w in the lowest few km of the troposphere, as has been shown in previous studies, but also at altitudes in the upper troposphere where water vapor is limited.