Field measurements of water vapor continuum absorption in the visible and near-infrared

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 109,, doi: /2003jd003586, 2004 Field measurements of water vapor continuum absorption in the visible and near-infrared B. Sierk, 1,2,3 S. Solomon, 1 J. S. Daniel, 1 R. W. Portmann, 1 S. I. Gutman, 4 A. O. Langford, 1 C. S. Eubank, 1,2 E. G. Dutton, 5 and K. H. Holub 4 Received 11 March 2003; revised 11 December 2003; accepted 24 December 2003; published 27 April [1] We carried out a spectroscopic field experiment designed to measure water vapor continuum absorption in the visible and near-infrared spectral regions. Atmospheric spectra at 1 nm resolution were recorded using direct sunlight at high solar zenith angles during sunrise. Simultaneously radiosonde soundings and a network of geodetic Global Positioning System (GPS) receivers were deployed to constrain the water vapor amount along the absorption path. The solar spectra were analyzed using the Differential Optical Absorption Spectroscopy technique, while the GPS and radiosonde observations were used as input data to a line-by-line radiative transfer model to compute theoretical differential absorption spectra. The difference between the measurements and the simulated spectra provides information regarding the additional absorption owing to the H 2 O continuum. The data are compared to predictions of the widely used Clough- Kneizys-Davies continuum model as well as with theoretically derived spectra of water dimer. The results show that continuum absorption contributes significantly to solar absorption even in highly saturated H 2 O bands. The comparisons provide the needed observations to improve future continuum parameterizations. INDEX TERMS: 3359 Meteorology and Atmospheric Dynamics: Radiative processes; 3360 Meteorology and Atmospheric Dynamics: Remote sensing; 0365 Atmospheric Composition and Structure: Troposphere composition and chemistry; KEYWORDS: water vapor continuum, radiative transfer, DOAS Citation: Sierk, B., S. Solomon, J. S. Daniel, R. W. Portmann, S. I. Gutman, A. O. Langford, C. S. Eubank, E. G. Dutton, and K. H. Holub (2004), Field measurements of water vapor continuum absorption in the visible and near-infrared, J. Geophys. Res., 109,, doi: /2003jd Introduction [2] Water vapor in the atmosphere plays a key role for the energy budget of our planet. It is not only involved in the transport and conversion of energy but is also the strongest atmospheric absorber of solar radiation in the short-wave spectral region, where the Sun has its emission maximum. Thus understanding the highly complex absorption spectrum of H 2 O is essential for atmospheric radiative transfer (RT) calculations, which in turn are critical for climate models. Given the high degree of nonlinearity in atmospheric processes, even small absorption effects can have an impact on model predictions and therefore have to be understood. There are several sources of uncertainty in 1 Aeronomy Laboratory, National Oceanic and Atmospheric Administration, Boulder, Colorado, USA. 2 Cooperative Institute of Environmental Sciences, University of Colorado, Boulder, Colorado, USA. 3 Institute of Environmental Physics, University of Bremen, Bremen, Germany. 4 Forecast Systems Laboratory, National Oceanic and Atmospheric Administration, Boulder, Colorado, USA. 5 Climate Monitoring and Diagnostics Laboratory, National Oceanic and Atmospheric Administration, Boulder, Colorado, USA. Copyright 2004 by the American Geophysical Union /04/2003JD modeling the atmospheric water vapor spectrum in the visible (Vis) and near-infrared (NIR) spectral regions. One problem arises from errors in the spectral line parameters (e.g., spectral position, intensity, half width, etc.), which characterize the several thousand individual rotation-vibrational H 2 O transitions. The reliability of spectroscopic databases such as High-Resolution Transmission Molecular Absorption Database (HITRAN) [Rothman et al., 1998] or European Space Agency (ESA) water vapor (WV) database for the red (R) region [Schermaul et al., 2001]. However, the reliability of these databases has recently been questioned and problems have been identified which affect the results of RT codes in the Vis/NIR [e.g., Belmiloud et al., 2000; Giver et al., 2000]. [3] Another absorption effect which represents a key uncertainty is the so-called water vapor continuum absorption. Hettner [1918, p. 495] presented the first evidence of this low-frequency absorption component which underlies H 2 O line absorption in the infrared (IR) window region between 8 and 14 mm. Using laboratory measurements, he observed that the absorption beyond the 6.3 mm band never fades completely. This result was difficult to reconcile with physical understanding, and when Elsasser [1938a, p. 504] reviewed these data in light of his theoretical analysis of the rotational water vapor spectrum, he concluded that the absorption found by Hettner must be 1of20

2 ascribed to some other effect, which cannot yet be identified. However, in the same year, Elsasser suggested that the explanation for continuous absorption in the infrared window regions might be the accumulated far-wing contribution of strong lines in the water vapor bands on either side of a spectral window [Elsasser, 1938b; Elsasser, 1938c]. This hypothesis was supported by Bignell et al. [1962] and generally accepted until Penner and Varanasi [1967] and Varanasi et al. [1968] suggested that hydrogen bonded water dimers could contribute to the continuum. They emphasized an anomalously strong water pressure and exponential negative temperature dependence of the window absorption as arguments in favor of the dimer hypothesis. However, large uncertainties in the magnitude and spectral structure of water dimer absorption still exist, and the issue of the physical mechanisms causing the continuum has not yet been fully resolved. To date, most efforts to model the continuum absorption component are based on line shape theories rather than a dimer or other complexes [Clough et al., 1989; Tipping and Ma, 1993; Ma and Tipping, 2000, 2002]. [4] Owing to the fundamental impact on the atmosphere s energy budget, extensive studies of continuum absorption have been carried out in the atmospheric window regions of the thermal infrared. However, H 2 O continuum can also have a significant impact within the strong pure rotational and vibrational bands, sometimes referred to as in-band continuum [Tobin et al., 1996]. Since the discovery that radiative transfer models (RTM) systematically underestimate the total solar flux observed at the ground [Stephens and Tsay, 1990; Cess et al., 1995; Ramanathan et al., 1995], the short-wave spectral region of the atmospheric spectrum has also attracted broad attention. Among other factors, the water vapor continuum in the Vis/NIR has been proposed to contribute to this so-called anomalous absorption since it is often neglected in the short-wave region. A search for hidden absorbers such as water dimers [Daniel et al., 1999; Hill and Jones, 2000] has not yet resulted in sufficiently high absorption to make up the observed differences. Vogelmann et al. [1998] performed a field study using a Sun photometer in order to estimate the impact of neglecting H 2 O continuum in models of solar absorption. They found that the continuum is an unlikely explanation for the potential anomalous absorption of W/m 2. [5] However, there are many reasons why it is important to quantify and understand this weak absorption in the short-wave region. The line shape continuum models currently in use for the climatologically important window regions have been derived from laboratory measurements largely in the thermal infrared [Burch, 1981; Burch and Alt, 1984; Burch, 1985; Tobin et al., 1996] but are frequently applied in all wavelength regions. Measurements of the continuous absorption underlying the complex rovibrational line spectrum in the Vis/NIR therefore represent a challenge for these models. To our knowledge the magnitude and spectral structure of the continuum in the Vis/NIR H 2 O bands have never been assessed using field measurements. [6] In this paper we describe a spectroscopic field experiment, designed to study the in-band water vapor continuum within two H 2 O polyads centered at 720 and 940 nm, respectively. We performed medium-resolution (1 nm) spectroscopic measurements of direct solar radiation during sunrise. The spectra are analyzed using the Differential Optical Absorption Spectroscopy (DOAS) [e.g., Platt and Perner, 1983] technique, which is very sensitive to structured absorption features. We will demonstrate that DOAS, in combination with independent remote sensing techniques, e.g., Global Positioning System (GPS) meteorology, is capable of providing spectral information about the water vapor continuum. The results are compared to two versions of the widely used Clough-Kneizys-Davies (CKD) continuum model, which is based on spectral line shape theory, CKD [Han et al., 1997], and its most recent compilation, MT_CKD_1.0 (M. J. Mlawer et al., manuscript in preparation, 2003). We also examine our data set for evidence of possible water dimer absorption postulated by some studies. Finally, we explore the possibility of finding indications on the physical origin of H 2 O continuum absorption. 2. Continuum Retrieval and Modeling [7] We carried out a field experiment in Boulder, Colorado, over a 1 month period in summer 2001 with the goal of studying various aspects of atmospheric absorption by tropospheric water vapor in the Vis and NIR spectral regions. The experiment has been briefly described in a paper investigating the accuracy of spectral line parameters from molecular databases [Sierk et al., 2003]. Here we will describe how the data set from the same field experiment was used to study water vapor continuum absorption DOAS Approach [8] The DOAS technique, as indicated by its name, is directed at obtaining a differential signal of molecular absorption with respect to a baseline or reference spectrum for the purpose of determining the atmospheric abundance of absorbing molecules. Various implementations of this technique differ not only in instrumental specifications like spectral resolution but also in the viewing geometry. The approach used in this study is schematically depicted in Figure 1. A solar telescope, located on the roof of the David Skaggs Research Center (DSRC) in Boulder, continuously tracked the Sun from sunrise to noon. Via split optical fibers, the collected sunlight was transmitted into three medium-resolution (1 nm) grating spectrometers (type Ocean Optics S2000) located inside the laboratory. The instruments, covering a spectral range from 500 to 1000 nm were cooled to 20 C to minimize thermal noise of the CCD detectors. Spectra near the horizon with solar zenith angles (SZA) up to 89 were recorded to take advantage of long propagation paths, maximizing the attenuation signal and enabling the detection of even weak absorption features. The measurements of these foreground spectra are divided by a low zenith angle spectrum acquired around noon. This division eliminates instrumental effects like the response function of the spectrometers and largely removes solar Fraunhofer lines. A disadvantage of this approach is that the differential spectra do not contain absolute information about the absorber and spectral features. Instead they reflect with great precision any changes in radiative transfer with respect to the conditions present at the time when the background was recorded. 2of20

3 Figure 1. Schematic illustration of the DOAS setup deployed to retrieve water vapor continuum. A series of foreground spectra is recorded during sunrise using a Sun tracking telescope feeding three spectrometers. The sunrise spectra are divided by a high-sun background spectrum acquired around noon to eliminate instrumental effects and Fraunhofer features. [9] A simple radiative transfer model (RTM) was used to compute the theoretical differential spectra. It computes the fraction of sunlight transmitted along the propagation path, taking into account refraction. The RTM models the extinction owing to molecular absorption as well as Rayleigh and Mie scattering along the direct refracted path. Given the 1 field of view of the telescope, the contribution of diffuse light to the detected absorption signal can be neglected [Halthore et al., 1997]. The monochromatic intensity of the downwelling direct solar radiation is calculated from Beer s law and the superposition of all single rotation-vibrational transitions in the investigated H 2 O band. Denoting the extraterrestrial solar intensity on top of the atmosphere by I 0, we write the intensity of direct solar radiation reaching the telescope as IðlÞ ¼ I 0 ðlþexp S g t g ðlþ þ t aero ðlþþt ray ðlþ ; where t g are the optical depths owing to gaseous absorption at wavelength l, and the summation runs over the number of absorbing species. With t aero and t ray we denote the optical depths owing to aerosol and Rayleigh scattering, respectively. The latter can be directly computed from the pressure profile using the wavelength dependence of the cross section given by Nicolet et al. [1982]. The aerosol optical depth was measured in our experiment by a direct beam Sun photometer (SP02), which was colocated with the solar telescope. The instrument performs measurements of solar irradiance at the wavelengths 411.8, 499.9, 675.3, and nm. From the observations, aerosol optical depths t aero for the above wavelengths can be derived [Dutton et al., 1994]. These are in turn used to determine the Angstrom exponent a in a least squares fit to the equation t aero ¼ b l a AMF aero ; with a constant extinction coefficient b and the air mass factor AMF aero. The measured Angstrom exponent accounts ð1þ ð2þ for the wavelength-dependent part of the extinction signal that is introduced in the presence of aerosols and cannot be accounted for by the offset and slope parameters C 0 and C 1 (see equation (5) below). Typical values of a range between 0.8 and 1.4. The impact of the aerosol absorption on our continuum retrieval is relatively small because the DOAS technique is only sensitive to structured absorption features within the investigated wavelength window. The spectral variation of the differential (ratioed) aerosol optical depth is smooth over the width of an observed absorption band. [10] In the following we consider only H 2 O as a molecular absorber, reducing the sum in equation (1) to a single term and computing the optical depth for water vapor by Z t w ðlþ ¼ N w ðþs s l S l ðts ðþþf l ðl; Ts ðþ; Ps ðþþds; ð3þ S where P and T denote the pressure and temperature along the propagation path s, and N w the number density of the water molecules. S l is the line intensity of transition l, whose temperature dependence is modeled as described by Rothman et al. [1998], and f l a factor describing the spectral shape of the individual absorption lines. We used the Voigt profile for f l which results from convolving the line shapes owing to Doppler and collision broadening. In case of water vapor the latter component is dominant throughout the troposphere, where most of the absorbing water molecules reside. While the Doppler component can be computed from the temperature profile, which is measured by the radiosondes, the collision broadening is described by the Lorentz profile: f c;l ðnþ ¼ g l pn ð n 0 Þ 2 þg 2 l : ð4þ [11] Here g l is the pressure broadened or Lorentz half width and n 0 the line center position in wave numbers. For 3of20

4 the derivation of equation (4), instantaneous molecular collisions have to be assumed. In the line shape hypothesis of the water vapor continuum, interactions between the collision partners cause deviations from the Lorentz shape in the far wings of the true line profiles, and the accumulated super-lorentzian absorption constitutes the continuum. A a more detailed discussion of continuum models is given in section 2.3. Our RTM evaluates equations (1) (4) at a spectral resolution of 0.01 cm 1 out to 300 cm 1 from the line center for each H 2 O transition. Simulation calculations show that the errors in transmission introduced by applying this cutoff are less than 0.01% and can safely be neglected. The line-by-line calculation requires knowledge of spectral parameters, such as intensity and half width at norm conditions, which we extracted from version 11.0 of the HITRAN database, released in December The impact of the spectral a priori information on our retrieval is discussed in section 3.1. Using refractivity profiles computed from the radiosonde observations (RAOB), we calculated correction tables relating the true and apparent solar angles following the method of Auer and Standish [2000]. The absorption path is then computed by ray tracing through a spherically layered atmosphere. For the DOAS analysis we form the ratio of the foreground and background spectrum T d measured by our instruments, as in good approximation this eliminates I 0 (l) in equation (1). (The error introduced by the assumption, that the application of Beer-Lambert s law and the convolution with the instrument function are interchangeable, the so called I 0 effect, was shown to be negligible for our experiment [Sierk et al., 2003].) In the RTM we simulate the measurement T d by computing the two spectra for the corresponding conditions at high spectral resolution (0.01 cm 1 ) and forming the ratio T d ðlþ ¼ Z Z Dl Dl I fg ðl 0 ÞAðl l 0 Þdl 0 exp½ ðc 0 þ lc 1 ÞŠ: ð5þ I bg ðl 0 ÞAðl l 0 Þdl 0 [12] In the above equation the monochromatic intensities of both the foreground and the background spectrum (denoted by the subscripts fg and bg) are convolved with the instrument function A(l) of width Dl, which accounts for the finite resolving power of the spectrometer. This function is determined from emission line spectra, which are acquired using neon and argon calibration lamps. The width of emission lines produced by these light sources is typically narrower than the pixel bandwidth of our spectrometers by about an order of magnitude. Therefore their light can be regarded as monochromatic, and the measured and normalized line profiles yield A(l). The coefficients C 0 and C 1 in the last two terms of equation (5) are parameters estimated in the DOAS analysis accounting for residual differences between model and measurement, which can be approximated as offset and slope in optical depth. If we assume that the meteorological profiles N(s), P(s), and T(s) along the absorption paths are known for both the foreground and background spectrum (e.g., from radiosonde observations), we can compute T d (l) using equations (1) (5). We define this forward calculated differential intensity as the baseline spectrum. By forming the difference with the measurements we obtain a quantity, which reflects any additional absorption not accounted for in the RTM (plus measurement and model errors). In our analysis we refer to this quantity as the retrieved differential continuum. [13] In practice, the retrieval from the field spectra is done in two steps: First, we carry out a DOAS analysis performing a nonlinear least squares fit of the RTM results to the measured differential spectra. The parameters determined in the fit are the foreground slant column of water vapor Z SCW fg ¼ N w ðþds; s fg as well as the coefficients C 0 and C 1. Two further parameters, the so-called shift d and stretch e are estimated, which correct for wavelength misalignment between the background and the foreground spectrum. The background value SCW bg is computed from a radiosonde sounding, coinciding with the acquisition of the background spectrum and is fixed in the least squares fit. The DOAS analysis of this first retrieval step compensates for any additional absorption not accounted for in the RTM by estimating higher absorber amounts. Therefore, since the continuum is not included in the RTM, we expect the retrieved values of SCW fg to be higher than the true values at solar zenith angles near the horizon, where it contributes significantly to the total observed absorption (see also section 2.3). In order to retrieve the continuum in the second step, we recompute the theoretical foreground spectrum using an independent value SCW fg obtained from the combined RAOB and GPS analysis described in the subsequent section. In this forward model calculation the values of the fit parameters C 0, C 1, d, and e obtained in step 1 are reused. The result is a theoretical spectrum simulating the observations in absence of continuum absorption. Forming the difference with the ratioed measured spectrum yields the retrieved differential continuum Monitoring the Humidity Field [14] One source of uncertainty in the approach outlined above is the assumption that the absorber profile along the ray path is known. In order to minimize the error introduced by this assumption, we measured the horizontal and vertical distribution of water vapor within the part of the troposphere that was intersected by the ray path during sunrise. Two radiosondes were simultaneously launched every morning 30 min after sunrise for in situ measurement of the vertical profiles N w (s), P(s), and T(s). One sounding was colocated with the spectrometers in Boulder (DSRC), the other in Ft. Morgan (FTMC), 150 km further to the eastnortheast, roughly in the direction of the sunrise. The two launch sites are indicated in Figure 2, which depicts a map of the experiment area. In addition to the two balloon soundings in the morning, a third radiosonde was launched around noon at the Boulder site in order to provide the atmospheric profiles for the computation of the background spectrum I(l) bg in equation (5). ð6þ 4of20

5 Figure 2. Map of the experimental GPS network used in this study. The six GPS sites are indicated by the red circles (along with their site identification code). They are located roughly along the sunrise direction to provide accurate estimates of horizontal water vapor gradients. The range of sunrise directions during the experiment is indicated by the black lines. Radiosondes were simultaneously launched in Boulder (DSRC) and Ft. Morgan (FTMC), providing vertical profile information. [15] For monitoring of horizontal gradients as well as the temporal evolution of the humidity field, we deployed a remote sensing technique on the basis of the Global Positioning System (GPS). The so-called GPS meteorology, a method which has been developed over the past decade [Bevis et al., 1992; Rocken et al., 1995], utilizes the refraction effect of tropospheric water vapor on microwave signals to estimate the propagation delay of the satellite signals to the receiver. The technique requires a continuously operating network of receivers with precisely known site coordinates. In such a network the path delays obtained from a geodetic analysis can be transformed into an estimate of the water vapor column above the receiver if accurate pressure and temperature data are available at the GPS site. We set up a chain of six geodetic GPS receivers from the Boulder measurement site to Sterling, 188 km to the eastnortheast, which is depicted in Figure 2. Each GPS site was additionally equipped with meteorological ground sensors for pressure and temperature, which enabled the separation of the wet and the dry component of the refraction delay. Further details on the processing of the GPS data are given by Sierk et al. [2003]. The GPS technique has been proven to provide estimates of zenith column water with an absolute accuracy of 1 kg/m 2 [e.g., Kruse et al., 1999]. Most of this uncertainty can be attributed to error sources like satellite orbit precision and residual ionospheric refraction, which affect our experimental GPS sites in the same way, owing to their short baseline distances. The relative accuracy of the GPS retrievals is therefore likely to be much higher, making the receiver chain very sensitive to horizontal gradients of column water vapor. Furthermore, its time resolution of 30 min is sufficient to monitor temporal trends in the humidity field. In combination with the radiosonde measurements on either end of the GPS receiver chain, this enables us to characterize the water vapor distribution in the experiment area and select periods with suitable conditions for the continuum retrieval. The DOAS data set acquired over the 2 weeks of the experiment was filtered by screening the RAOB and GPS data for periods with spatial homogeneity and temporal stability of the humidity field. The following selection criteria were applied: [16] 1. Low horizontal gradients: The difference in vertical columns in Boulder and Ft. Morgan (measured from the Boulder altitude 1670 m) was required to be within 0.5 kg/m 2. [17] 2. Similarity of vertical humidity profiles: The maximum difference in absolute humidity measured by the RAOBs was limited to 1 g/m 3. [18] 3. Temporal stability of the humidity field: No strong variations were allowed during the first hour after sunrise. [19] 4. Consistency of humidity measurements: The difference between GPS and RAOB derived vertical column of water (VCW) had to be within 1 kg/m 2. [20] Four days have been selected for the analysis, which fulfilled these criteria. Figure 3 shows the water vapor profiles measured by the radiosondes in Boulder and Ft. Morgan for the selected days. The water column from integrating the humidity profiles starting from the Boulder 5of20

6 Figure 3. Selected radiosonde profiles of absolute humidity for Boulder and Ft. Morgan. The corresponding vertical columns integrated from the Boulder altitude (1670 m) are indicated in the legends. altitude (1670 m) are also indicated in the plots. From the figures it is obvious that the largest differences between the humidity profiles of the selected days occur below the altitude of the Boulder station in the boundary layer of the Ft. Morgan site located at 1386 m elevation. This part of the humidity profile becomes important only at SZA > 90. We therefore limit our analysis to zenith angles above the horizon. 6of20

7 Figure 4. Vertical column of water vapor (VCW fg )asa function of solar zenith angle (SZA). Results from the two investigated H 2 O bands are plotted along with the GPS estimates from the Boulder station and the integrated radiosonde soundings launched in the morning. Consistent with Figure 7, the DOAS results show a water vapor increase with SZA, while both the GPS as well as the radiosondes suggest a fairly constant humidity field during sunrise. The retrievals from simulated data show that systematic errors in the line intensity parameters do not result in the observed sharp increase in excess water column, while the simulation with the latest CKD model shows a SZA dependence similar to the one observed. [21] Vertical water column estimates from GPS and DOAS are shown as a function of SZA in Figure 4. The columns from integrating the humidity profiles measured by the radiosondes launched in Boulder and Ft. Morgan are indicated by the triangles in the plot. In addition to the measured columns Figure 4 shows results from tests with simulated spectra, which will be explained in section 3.1. At SZA > 80 we see a sharp increase in VCW from DOAS measurements. Since the two RAOBs and the GPS results do not indicate a horizontal gradient in the water vapor field, we attribute the retrieved excess water amount to the increasing continuum absorption at large SZA. At solar angles less than 80, a good agreement between GPS and DOAS can be observed, with the exception of a 1 hour period in the SZA range Such deviations within the accuracy limits of the GPS technique are frequently observed in comparisons with remote sensing techniques providing measurements of slant water column, like microwave radiometry [Kruse et al., 1999] or DOAS [Sierk et al., 2003]. They can be attributed to the spatial and temporal smoothing of the GPS technique, which results from averaging over the 30 min session lengths (estimation intervals) and the lines of sight toward the observed satellites, distributed over the hemisphere. These limitations of the GPS technique demonstrate the importance of the radiosonde soundings to provide an independent absolute measurement of the water column. In the computation of the forward model for continuum retrieval (step 2 of the procedure described in the previous section), we used the radiosonde soundings to define the absolute value of vertical water column. The corresponding GPS time series were adjusted by a correction offset to match the RAOB values. These corrected GPS time series were used to monitor temporal variations of the water vapor amount during sunrise Continuum Models [22] We have chosen two water vapor bands for our study of the continuum, the 3 n and the 4 n polyads centered at 940 and 720 nm, respectively. Simulated absorption spectra computed using the RTM described in section 2.1 are depicted in Figures 5 and 6. They show the components of the spectrum at instrument resolution for a solar zenith angle of 85 and a model atmosphere on the basis of the radiosonde soundings of 7 August A particular difficulty with the 4 n polyad at 720 nm is the overlap with the oxygen B band centered around 690 nm (see Figure 6). In order to avoid misinterpretations owing to higher fit residuals in this region, the O 2 B band is excluded from the analysis. This does not introduce additional errors since the DOAS technique exploits only the spectral signature of an absorption feature to extract information on the absorber, and the strongest features reside outside the overlap region. The disadvantage of limiting the fit window to nm is that we will not be able to characterize the continuum on the short-wavelength shoulder of the band. Another strong H 2 O band located between the two selected ones, the 3 n + d polyad centered around 820 nm, will not be considered here because of the evidence of large errors in line strength data for this band [Sierk et al., 2003]. [23] Along with the H 2 O and O 2 monomer contributions, Figures 5 and 6 also show the corresponding continuum absorption components as computed by the widely used Figure 5. Theoretical transmission for the 3 n polyad centered at 940 nm, omitting attenuation by scattering. The plot shows the water vapor band at instrument resolution (grey) as well as the expected continuum component based on the CKD (blue) and dimer models (green). The dimer absorption has been multiplied by a factor of 10 in this figure. Although the continuum shows a substantial absorption in this region, the impact on the total transmission (black) is relatively small, owing to the high degree of saturation within this strong H 2 O band. 7of20

8 Figure 6. Theoretical transmission for the 4 n polyad centered at 720 nm, omitting attenuation by scattering. The plot shows the water vapor band at instrument resolution (red) as well as the expected continuum component based on the CKD (blue) and dimer models (green). The dimer absorption has been multiplied by a factor of 10 in this figure. The oxygen B band visible on the left side of the plotted interval is excluded from the analysis in order to avoid misinterpretations owing to residual O 2 absorption in the overlap region. Clough-Kneizys-Davies model (CKD) version [Clough et al., 1989; Han et al., 1997]. The CKD model is based on the assumption that noninstantaneous molecular collisions give rise to a super-lorentzian line shape. To account for the effects of collision durations, so-called c functions are introduced which describe the deviation of the true line shapes in the far wings of the H 2 O transitions from the Lorentz profile (equation (4)). The analytical forms of the c functions for H 2 O-H 2 O collisions (self continuum) and water-air molecular broadening (foreign continuum) are postulated, yielding a super-lorentzian line shape which decays exponentially to become sub-lorentzian about 100 cm 1 from the line center. The parameters of the c functions are determined in a least squares fit to laboratory spectra in the thermal infrared windows. In order to separate the high-frequency line absorption from the broadband continuum, the CKD model defines a spectral window of 25 cm 1 width around the line center, in which the Lorentzian line shape is computed. (The latter is approximated by the van Vleck-Weisskopf-profile [van Vleck and Huber, 1977], which does not differ significantly from equation (4) in the wavelength regions investigated here.) The continuum contribution of the single line is then obtained by subtracting this center part, reduced by its value at the 25 cm 1 interval limits, from the CKD absorption line profile. What remains is the line shape outside the 25 cm 1 window (including the c functions) plus the so-called pedestal, the constant residual absorption given by equations (3) and (4) evaluated 25 cm 1 from the line center n 0. The continuum is composed of these remaining absorption features from all transitions in the band. The temperature dependence of the self continuum is described by a logarithmic interpolation between the continua determined for several measured temperatures. The absorption coefficients of the self continuum are higher than for foreign broadening by about an order of magnitude. However, under atmospheric conditions the foreign continuum dominates within the H 2 O bands, while the self continuum is most important in the window regions of the water vapor spectrum. [24] In addition to version of the CKD model, we also tested its most recent release, which is referred to as MT_CKD_1.0 (M. J. Mlawer et al., manuscript in preparation, 2003). It represents the first complete recomputation of the entire self and foreign broadened continuum since the original model was published [Clough et al., 1989]. MT_CKD_1.0 is based on a new formulation consisting of two components: One describes the additional (super- Lorentzian) absorption in the intermediate line wings as a result of an additional collision-induced dipole moment of the water molecule. The second component models the line wings associated with the normal transitions as Lorentzian in the near wing and sub-lorentzian in the far wing. Unlike the previous releases the latest continuum model is based predominantly on atmospheric spectral measurements. [25] It should be noted that the inferred continuum absorption depends on the above outlined formulation of the CKD model, which yields a nonzero continuum component even in case of purely Lorentzian collision line shapes (without enhanced absorption in the far line wings). If it is included in the RTM as in our comparisons in section 4.1, this has to be accounted for by applying the same 25 cm 1 cutoff to each H 2 O transition in the line-by-line calculation and subtracting the pedestals before adding the continuum from the CKD model. Whenever it is not included, as for the computation of the baseline spectrum used in the continuum retrieval, no pedestal is subtracted and the line shapes are computed out to 300 cm 1 from n 0. It is obvious that the CKD models depend on database values for spectral line parameters, as its derivation requires line-by-line RTM calculations. Since both versions of the CKD model investigated in this study are based on the HITRAN database, it is the natural choice for the spectroscopic a priori information in our analysis. However, it should be mentioned that other databases for spectral line parameters exist, e.g., the more recently compiled ESA-WVR [Schermaul et al., 2001]. [26] For completeness in our discussion of continuum absorption, theoretical spectra of the water dimer molecule are also plotted in Figures 6 and 7. They have been computed using the equilibrium constants given by Vaida et al. [2001], as well as cross section data from Low and Kjaergaard [1999]. Unlike the two models based on line shape theory, water dimer absorption theory is not specifically intended to account for the continuous component of water vapor monomer absorption, but rather represents a different stable molecule. However, it has been speculated in the past that hydrogen-bonded water dimers contribute partly to the phenomenon and therefore will be investigated here. Our calculations suggest that the expected magnitude of the dimer absorption is lower than the CKD model by an order of magnitude in the considered spectral regions. However, Vaida et al. [2001] estimate the uncertainty in the equilibrium constants by a factor of 3 and that of the cross sections by a factor of 2. 8of20

9 Figure 7. Slant columns of water vapor (SCW fg ) inferred from DOAS measurements and simulations of the 940 nm band for the conditions on 7 August The blue curve represents the theoretical values from integrating the radiosonde sounding and multiplying with the air mass factor (see equations (7) and (8)). The red and green curves represent retrievals from observed and simulated spectra, respectively. The latter were computed by including the MT_CKD_1.0 continuum model in the RTM. The slant columns from simulated and real data are in good agreement, both showing steeper increase with SZA than the theoretical air mass factor. The plot demonstrates the impact of a water vapor continuum on DOAS retrievals of H 2 O slant columns at large SZA. A more recent study by Goldman et al. [2001] suggests a significantly stronger temperature dependence of the equilibrium constant for water dimerization, supporting an enhancement in the expected dimer abundance. The large uncertainties in spectral properties and abundance of the water dimer leave open the possibility of stronger absorption by an order of magnitude. We have therefore arbitrarily multiplied the dimer cross sections by a factor of 10, indicated by the green curves in the figures, to explore whether or not our technique could differentiate the continuum from possible dimer structures either spectrally or by the zenith angle dependence of the band-averaged absorption. 3. Testing the Approach [27] As will be shown in section 4.1, the expected absorption signal of the differential continuum is on the order of a few percent. Given these relatively weak signals it is essential to assess the accuracy of our retrieval technique. In this section we present several tests we performed to verify the capability of our DOAS approach to retrieve continuum absorption using spectroscopic measurements during sunrise. The tests deal with different aspects of our approach, which represent potential error sources: The use of spectroscopic a priori information and the computation of the absorption path using ray tracing. Both tests put constraints on the accuracy of our retrievals and can therefore be regarded as part of the error analysis, which is summarized in section Tests With Simulated Data [28] In order to test the sensitivity of our approach to continuum absorption, we investigated the impact of a super-lorentzian component as predicted by the CKD models on DOAS retrievals of slant water vapor column. Theoretical spectra of the 940 nm band were computed using the RTM described in section 2.1, adding the optical depths for the continuum as calculated from the MT_CKD_1.0 model (see section 2.3). The ratios of theoretical spectra (from equation (5)) for realistic foreground and background conditions (measured by the radiosondes) were then used as simulated data, from which the slant water columns SCW fg,sim were retrieved with the DOAS algorithm outlined in section 2.1 (step 1). The theoretical air mass factor (unaffected by the continuum) is given by Z 1 AMF w;raob ¼ VCW fg;raob fg N w;raob ðþds; s where the integration is carried out along the computed ray path s, and the vertical column of water vapor VCW fg,raob is obtained from integrating the radiosonde measurement of the number density N w,raob over height. We can compare the corresponding theoretical slant column of the RAOB sounding SCW fg;raob ¼ VCW fg;raob AMF w;raob ; with the simulated DOAS retrievals SCW fg,sim and the observations from real data SCW fg,doas. The comparison is shown in Figure 7, where the observed, simulated, and theoretical slant columns for 7 August 2001, are plotted versus SZA. The green line in the figure represents the retrievals from simulated spectra SCW fg,sim with underlying continuum component. At large SZA the simulated retrievals deviate increasingly from the theoretical values SCW fg,raob (plotted in blue) which are computed from the direct light path and the water vapor profile using equations (7) and (8). The effect becomes substantial only at large zenith angles, but for SZA > 80 the deviation increases rapidly with SZA, reaching up to 20% near the horizon. The discrepancy between the slant columns from processing simulated spectra and SCW fg,raob reflects the impact of a super-lorentzian continuum component on the air mass factor, yielding higher values than AMF w,raob and consequently larger slant columns. This shows the necessity to incorporate a water vapor continuum model into the RTM for accurate prediction of H 2 O slant columns at large SZA. It is this impact of continuum absorption on DOAS retrievals which we exploit to infer the differential continuum, as we use the theoretical slant columns SCW fg,raob to calculate the baseline spectrum (see section 2.1). The red line in Figure 7 represents observed slant columns SCW fg,doas retrieved from measured spectra. As can be seen in the plot, they show a good agreement with the retrievals from simulated data. This indicates that the continuum model used in the computation of the simulated spectra predicts the observed excess columns (or the enhancement in the air mass factor) fairly ð7þ ð8þ 9of20

10 well. The continuum component retrieved in the analysis is therefore expected to be on the same order of magnitude as the super-lorentzian model. [29] As outlined in section 2.1, the DOAS technique relies on accurate H 2 O line parameters. Several studies have been carried out to assess the accuracy of spectroscopic a priori information from molecular databases, and systematic differences with laboratory and field measurements have been identified [Learner et al., 1999; Giver et al., 2000; Belmiloud et al., 2000; Chagas et al., 2001]. The present data set with simultaneous colocated measurements of DOAS, RAOB, and GPS was used to assess the consistency of five H 2 O bands [Sierk et al., 2003]. Owing to the accuracy limits of the GPS technique, we used the strong 3 n polyad of water, which is also under investigation here, as a reference band to compute band-averaged correction factors for the other four bands. We found an almost perfect agreement between the HITRAN parameters of the two bands investigated here (within 1 2% band-averaged intensity). The uncertainty of the GPS estimates did not allow for an assessment of the absolute accuracy of the line intensities from HITRAN. Nevertheless, we can use the identified bias between GPS and DOAS measurements of vertical water column to estimate the band-averaged error in line intensities. During the entire observation period of 12 days we determined a positive offset of the GPS estimates of 0.4 kg/m 2 with respect to the DOAS retrievals from the 940 nm band with a day-to-day standard deviation of 0.13 kg/m 2. This result suggests that the HITRAN line intensities of this band are too high by 6% on average. Since the absolute accuracy of the GPS is only within 1 kg/m 2,it is also possible that the lines in the band are underestimated by roughly the same amount. We therefore assume an uncertainty in the HITRAN values of ±6% and analyze the impact of line intensity errors of this magnitude on our DOAS retrievals. This was done by simulating data in the same way as described above, but instead of including CKD continuum absorption, we calculated differential spectra with 6% increased and decreased line intensities. The DOAS analysis of these simulated spectra was performed using the unchanged HITRAN data as described in section 2.1. The results are shown in Figure 4, where the vertical water columns VCW fg from the simulations are plotted against SZA, along with the retrievals from simulated spectra including the MT_CKD_1.0 model (see above). The figure also shows the DOAS retrievals from the real observations as well as the GPS estimates and integrated RAOB soundings. We can see from the plot that the strong increase in the observed VCW fg is not produced by the assumed systematic errors in the spectral database. Increasing the line intensities in the simulated data results in a much slower rise of retrieved vertical column with SZA, which reaches a maximum of 0.9 kg/m 2 at 89. In contrast, the inclusion of the latest CKD continuum yields almost five times higher values of excess zenith column, which agrees well with the observed values from both investigated H 2 O bands. The explanation for the strong impact of the continuum component is that the broadband absorption feature becomes more important with increasing absorption path, as it fills the nonsaturated spectral regions between the heavily saturated vibrational transitions. Unrealistically high corrections of the line strengths would be required to reproduce the observed increase in absorption since under saturated conditions they only affect the line wings. Any such large database errors would also show up in DOAS retrievals at low SZA, yielding a significant offset between the model and the measurement. This would be in contradiction to the good agreement found in our long-term comparison between GPS and DOAS. The simulated data generated with reduced line intensities yield columns that are smaller than the theoretically true RAOB value. Under such conditions, our retrieval technique would infer a negative continuum. [30] While the results shown in Figure 4 indicate that erroneous spectral line parameters are unlikely to be the source of the observed increase in absorption at high SZA, they can still be present and affect the magnitude of the retrieved continuum. Similar to the simulation calculations presented above, we can perform retrievals with modified HITRAN line parameters to estimate their contribution to the error in continuum retrieval. This will be done in the error analysis reported in section Tests With O 2 and O 4 Absorption [31] The performance of the RTM can be assessed by testing our approach with absorption bands of oxygen and its collision complex O 4. Atmospheric oxygen is an ideal test species for our retrieval technique because like water vapor, O 2 exhibits sharp absorption features caused by saturated and unsaturated lines within our observable spectral range, while the collision complex O 4 displays a continuous absorption. Since oxygen, unlike water vapor, is a well-mixed gas in the atmosphere, the abundances of both O 2 and the O 4 are directly determined from ground pressure. This in combination with accurately known spectroscopic properties [Greenblatt et al., 1990] enables us to compute O 2 and O 4 spectra with high precision. [32] The first aspect of the RTM to be tested using oxygen absorption is the ray tracing algorithm. It is used to compute the absorption path along which equation (3) is integrated, taking into account the ray bending owing to refraction at large zenith angles near the horizon. DOAS retrievals of oxygen can be compared with theoretical values, analogously to H 2 O in the previous section and Figure 7. With known absorber amount in the absence of H 2 O continuum, deviations between retrieval and theoretical values reveal imperfections of the ray tracing algorithm. We analyzed the strongest bands of O 2 and O 4 in the observed spectral window, which are depicted in Figures 8 and 9. The oxygen A band centered at 760 nm consists of strong (mostly saturated) O 2 transitions giving rise to a sharp and quite narrow absorption signal at our instrument resolution. The DOAS results from these two bands for 30 July 2001, are presented in Figure 10. The plots show retrieved and theoretical slant columns of O 2 and O 4. The excellent agreement between modeled and retrieved O 2 columns confirms the good performance of the RTM in modeling refracted absorption paths near the horizon. It also demonstrates the good quality of the instrument functions determined from emission lines, an example of which is also depicted in Figure 8. The O 4 band centered at 575 nm overlaps with the 5 n polyad of H 2 O (see Figure 9), but the excellent agreement shown in Figure 10 suggests that it is well separated by the retrieval algorithm. Figure 11 shows 10 of 20

11 Figure 8. Oxygen A band as measured by the DOAS spectrometer. The light blue line indicates the RTM calculation of the spectrum at high resolution. The model spectrum at instrument resolution is obtained by convolving with the normalized emission line profile (dark blue). The slant O 2 column is obtained in a least squares fit (grey squares) of the model to the measured spectrum (black). the O 4 vertical column densities corresponding to the slant columns in Figure 10 to better identify differences between theoretical and retrieved values. The applied air mass factors were computed using the ray tracing algorithm accounting for refraction in a spherical atmosphere (in analogy to equation (7)). Again we see excellent agreement between the measurements and theoretical values within the accuracy limits of the spectroscopic technique throughout the entire range of zenith angles. The results from both the O 2 and the O 4 analysis confirm the correctness of the absorption paths Figure 9. O 4 band at 575 nm and overlapping 5 n water vapor polyad centered at 590 nm. Despite the overlap of the two absorber species, the retrieved O 4 slant columns match the theoretical values and show the characteristic zenith angle dependence (see Figure 10). Figure 10. Slant column of (top) O 2 and (bottom) O 4 for 30 July The results retrieved from the O 2 A band at 760 nm and the O 4 band centered at 590 nm are plotted along with the theoretical values on the basis of ground pressure data and the ray tracing algorithm which computes the absorption path. The good agreement between theoretical and retrieved absorber amounts demonstrates the ability of the RTM to model refracted absorption paths near the horizon. computed and give us confidence that the errors introduced by the ray tracing algorithm are negligibly small. [33] In addition to the aspects of instrument function and absorption path calculation in the RTM, oxygen absorption can also be used for a direct test of the continuum retrieval technique outlined in section 2.1. A spectral region well suited for testing our approach is the O 2 g band centered at 630 nm. It features absorption line transitions which are spectrally colocated with an underlying O 4 continuum, as can be seen in Figure 12, which shows the g band and its components at our instrument s resolution. In order to test our technique, we retrieved the differential continuum of the g band owing to the collision complex and compared the result to the theoretically expected O 4 spectrum. The overlap of both oxygen features with the 4 n + d polyad of H 2 O (see Figure 12) makes the g band an even more stringent test for our approach. Figure 13 depicts the retrieved differential O 4 continuum compared with the corresponding RTM calculations at two different solar zenith angles. It is obvious from the plot that our measurements agree with the true O 4 continuum in magnitude, spectral shape, and 11 of 20