The Zugspitze radiative closure experiment for quantifying water vapor absorption over the terrestrial and solar infrared. Part III: Quantification of the near-infrared water vapor continuum under atmospheric conditions

We present a first quantification of the near-infrared (NIR) water vapor continuum absorption from an atmospheric radiative closure experiment carried out at Mt. Zugspitze (47.42° N, 10.98° E, 2964 m a.s.l.). Continuum quantification is achieved via radiative closure using radiometrically calibrated solar FTIR absorption spectra covering the 2500 to 7800 cm -1 10 spectral range. The dry atmospheric conditions at the Zugspitze site (IWV 1.4 to 3.3 mm) enable continuum quantification even within water vapor absorption bands, while upper limits for continuum absorption can be provided in the centers of window regions. Throughout 75 % of the 2500 to 7800 cm -1 spectral range, the Zugspitze results are agree within our estimated uncertainty with the widely used MT_CKD 2.5.2-model (Mlawer et al., 2012). Notable exceptions are the 2800 to 3000 cm -1 and 4100 to 4200 cm -1 spectral ranges, where our measurements indicate about 5 times stronger continuum 15 absorption than MT_CKD. The measurements are consistent with the laboratory measurements of Mondelain et al. (2015), which rely on cavity ring-down spectroscopy (CDRS), and the calorimetric-interferometric measurements of Bicknell et al. (2006). Compared to the recent FTIR laboratory studies of Ptashnik et al. (2012) and (2013), our measurements indicate 2–5 times weaker continuum absorption under atmospheric conditions in the wings of water vapor absorption bands, namely in the 3200 to 3400 cm -1 , 4050 to 4200 cm -1 , and 6950 to 7050 cm -1 spectral regions. 20


Introduction
Atmospheric water vapor is the most important contributor to the absorption of incoming solar radiation in the near infrared (NIR) (Kiehl and Trenberth, 1997).Water vapor absorption comprises both the effect of spectral line absorption and the broadband so-called continuum absorption (e.g.Shine et al., 2012).Depending on the atmospheric state and the choice of continuum model, up to 6 % of the clear-sky water vapor absorption can be attributed to the continuum (Paynter and 25 Ramaswamy, 2011).Consequently, exact quantitative knowledge of this contribution is a prerequisite for realistic atmospheric radiative transfer calculations employed e.g. in climate models (Paynter and Ramaswamy, 2014;Rädel et al., 2015;Turner et al., 2012).
However, the NIR atmospheric water vapor continuum currently still lacks sufficient experimental constraints.Recently, a number of laboratory studies based on different experimental techniques investigated this open question.Several efforts were 30 made to quantify both contributions to the continuum absorption, self and foreign continuum.Several studies made use of cell measurements with FTIR (Fourier Transform Infrared) spectrometers (e.g.Burch 1982;1985;Burch and Alt, 1984;Baranov et al., 2008;Baranov and Lafferty, 2011;Paynter et al., 2009;Ptashnik et al., 2011;2012;2013;2015). Furthermore, a number of spectral regions were covered by cavity ring-down spectroscopy (CRDS) measurements (Mondelain et al., 2014;2015) and by calorimetric-interferometric measurements (Bicknell et al., 2006).However, no consensus has been reached 35 among these studies.As noted e.g. by Mondelain et al. (2014) and Ptashnik et al. (2013), the individual results feature differences far beyond the respective uncertainty estimates whose attribution to causative processes remains tentative.A further challenge for laboratory studies is that they are normally not feasible at representative atmospheric conditions since Atmos.Chem.Phys. Discuss., doi:10.5194/acp-2016-323, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 April 2016 c Author(s) 2016.CC-BY 3.0 License.they typically rely e.g. on higher temperatures.To date, the temperature dependence of the self-continuum still lacks observational constraints (e.g.Paynter and Ramaswamy, 2011).This makes application of the laboratory results on atmospheric radiative transfer calculations non-straightforward.
A validation of continuum absorption strength under atmospheric conditions is therefore highly desirable to address these shortcomings.To this aim, we conducted a radiative closure experiment with the objective of quantifying the NIR water 5 vapor continuum absorption from atmospheric measurements.The study is carried out at the high-altitude Zugspitze site and relies on the solar FTIR measurements implemented at this site (Sussmann and Schäfer, 1997) in the framework of the Network of the Detection of Atmospheric Composition Change (NDACC; www.ndacc.org).
This paper is part of a three-paper series about different aspects of the Zugspitze radiative closure experiment.The first paper, thereafter referred to as Part I (Sussmann et al., 2016, same issue), describes the instrumental setup, evaluates the sensitivity 10 of the closure experiment in the far infrared (FIR), the mid-infrared (MIR), and the NIR, and provides results on the FIR water vapor continuum.A novel radiometric calibration method for solar FTIR spectra in the NIR is presented in a second paper, referred to as Part II (Reichert et al., 2016, same issue).Part III (this paper) contains the NIR continuum quantification method and results.Continuum quantification in the NIR is achieved comparing calibrated radiance spectra, obtained with the method presented in Part II, to radiative transfer model calculations.The results derived from our data set are presented 15 and compared to results from laboratory studies as well as the widely used MT_CKD 2.5.2 continuum model (Mlawer et al., 2012).This paper is structured as follows: Section 2 contains an overview of the instrumental setup used in the closure experiment.
Section 3 outlines the method for water vapor continuum quantification.In Sect.4, the results obtained with this method are presented and compared to previous studies.Finally, Sect. 5 contains a summary and conclusions.20

Setup of the closure experiment
The closure experiment relies on a quantitative comparison of measurements of spectral radiance with synthetic spectra calculated using the line-by-line radiative transfer model (LBLRTM, Clough et al., 2005).Water vapor continuum absorption is then quantified via the spectral residuals, i.e. the difference between simulated and measured spectra.
The instruments used in the Zugspitze radiative closure experiment und the related uncertainties are described in detail in 25 Part I.In summary, spectral radiances in the NIR are measured using a solar FTIR spectrometer setup at the Zugspitze (47.42°N, 10.98° E, 2964 m a.s.l.) summit observatory (Sussmann and Schäfer, 1997).Radiative calibration of measured spectra is achieved via a novel calibration method presented in Part II, which relies on a combination of the Langley method and measurements of a medium-temperature blackbody source.
The atmospheric state at the time of the radiance measurements is required as an input to the LBLRTM radiative transfer 30 calculations.To enable accurate quantification of the water vapor continuum from spectral residuals, the atmospheric state has to be constrained precisely using a number of additional measurements listed in the following.
Vertically integrated water vapor (IWV) constitutes the key input parameter and is derived directly from the solar FTIR spectra (e.g.Sussmann et al. 2009;Schneider et al., 2012).Temperature and pressure profiles are taken from four-times-daily National Center for Environmental Prediction (NCEP) resimulation data.NCEP resimulation data is also used to constrain 35 the shape of the water vapor profile.Column-averaged mixing ratios of CO 2 , CH 4 , and N 2 O are measured using the nearby Garmisch TCCON (Total Carbon Column Observing Network) solar FTIR instrument (Sussmann and Rettinger, 2014).O 3 columns were constrained by combined Brewer/Dobson measurements made at the nearby Hohenpeissenberg observatory (Köhler, 1995).Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-323, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 April 2016 c Author(s) 2016.CC-BY 3.0 License.
Aerosol optical depth (AOD) has to be constrained precisely in order to enable continuum quantification in the wings of the strong NIR water vapor bands and within window regions.A great advantage of the Zugspitze site is that AOD is typically very low, i.e. the AOD for the Zugspitze dataset is about a factor of 10 lower than at typical lowland mid-latitude sites.The AOD levels encountered in our closure data set (data set description and selection criteria see Sect.3.3) are in the range 0.0005 -0.00075 at 2500 cm -1 and in the range 0.0024 -0.0032 at 7800 cm -1 at airmass 1. AOD was measured using the SSARA-Z (Sun-Sky Automatic Radiometer -Zugspitze) sun photometer (Toledano et al., 2009) developed by the Meteorological Institute of the University of Munich and set up at Schneefernerhaus (2675 m a.s.l., 680 m horizontal distance to the Zugspitze solar FTIR).The instrument disposes of 13 spectral channels from 340 to 1640 nm.Only information from 5 channels centered at 439.6 to 781.1 nm was used in the analysis.The exact filter wavelengths and full width at half maximum (FWHM) values of these channels are listed in Table 1.The reason for the channel selection is that in the ultra-violet (UV) to visible range, water vapor continuum absorption can be considered negligible compared to AOD, whereas for the NIR channels continuum absorption will lead to biased AOD results.The channels below 440 nm were excluded since the high influence of Rayleigh scattering in the UV leads to increased AOD uncertainties.
The data analysis of the SSARA-Z measurements was implemented similar to the approach outlined by Toledano et al. (2009).In detail, we used standard Langley calibration for cloud-free periods.Rayleigh scattering was accounted for using the formula given by Bodhaine et al. (1999).In the analysis, a Gaussian shape was assumed for the filter transmissivity curves.The influence of absorption by O 3 was subtracted as outlined in Guyemard et al. (1995).NIR AOD was then deduced by assuming AOD wavelength dependence according to the Ångstrom relation: where τ designates AOD.The Ångstrom exponent α and scaling b are determined by a fit to the UV/visible AOD measurements.More sophisticated descriptions of the AOD wavelength dependence such as the relation given by Molineaux et al. (1998) may be used instead of Eq. (1).However, the number of sun photometer wavelength channels included in our analysis is not sufficient to place tight constraints on the higher number of parameters used in such models.Furthermore, the very low AOD at Zugspitze leads to high relative errors in the sun photometer measurements, which removes the benefits of more advanced models compared to Eq. ( 1).
The following contributions were included in the calculation of the AOD uncertainty: the sun photometer measurements are affected by uncertainty in the radiance measurements.This uncertainty contribution was set according to the 2-σ radiance measurement noise.The calibration uncertainty ensues from the uncertainty of the Langley fit.Additional uncertainty arises from the Rayleigh scattering correction, where central wavelength and FWHM errors of optical filters and atmospheric pressure errors contribute.The treatment of O 3 absorption is also prone to additional errors, due to filter parameter and O 3 column errors.In addition to these contributions, further uncertainty is induced by the fit to Eq. ( 1) that enables constraining the NIR AOD from the UV/visible measurements.The overall AOD uncertainty that ensues from these contributions for our data set at air mass 1 is <0.0015 at 2500 cm -1 and <0.0025 at 7800 cm -1 .

Method overview
The aim of this study is to constrain the NIR water vapor continuum absorption under atmospheric conditions.We make use of the radiative closure experiment setup at the Zugspitze observatory that is described in detail in Part I. Generally, radiative closure experiments comprise a quantitative comparison of spectral radiance measurements to synthetic spectra.The strategy for water vapor continuum quantification employed in this study relies on radiometrically calibrated solar FTIR spectra in the 2500 to 7800 cm -1 -range.Spectra were recorded with the solar FTIR instrument described in Sect. 2 and Part I, using no optical filter, a spectral resolution of 0.02 cm -1 (resolution is defined as 0.9/optical path difference), and averaging over 4-8 scans which leads to a 75-150 s repeat cycle per spectrum.The measured spectra are radiometrically calibrated by means of the calibration method outlined in Part II.Briefly, the calibration approach relies on Langley calibration in suitable spectral windows with little atmospheric absorption.In addition to the Langley technique that enables highly accurate calibration in selected windows, the 5 shape of the calibration curve between the windows is constrained using spectral radiance measurements of a high-temperature blackbody source.The calibration uncertainty achieved with this novel method is 1-1.7 % (2 σ) throughout the spectral range considered.Synthetic radiance spectra are generated using the LBLRTM radiative transfer model.The atmospheric state used as an input to the calculations was set based on the measurements described in Sect. 2. Given the calibrated spectral radiance measurements and the synthetic spectra, radiance residuals ∆I can then be calculated for a set of 10 spectra selected according to the criteria presented in Sect.3.3 where I FTIR designates the radiometrically calibrated solar FTIR spectra, I LBLRTM, no continuum the synthetic LBLRTM spectra not including continuum absorption and AOD the aerosol optical depth.Continuum optical depth  cont is calculated from the spectral residuals as follows: 15 (3) After calculation of the continuum optical depth (OD), absorption coefficients were derived from these results.The continuum OD τ cont is linked to continuum absorption coefficient k cont as follows where m designates the relative air mass, h obs the altitude of the observing instrument, n wv the water vapor number density, and 20 n air the dry air number density.k cont can be further decomposed in self-and foreign continuum contributions according to where  s and  f designate the self-and foreign continuum coefficients and  H 2 O ,  air , and  0 are the densities of water vapor, dry air and a reference density, respectively.In detail,  0 =  0 /( b  0 ), where P 0 = 1013 mbar, k b is the Boltzmann 25 constant, and T 0 = 296 K.
The separation of k cont in self-and foreign continuum contributions from atmospheric measurements is challenging.In principle, an assignment to self-and foreign continuum is possible using a large set of measurements covering a wide range of atmospheric conditions, i.e.IWV and temperature.However, the available data does not permit such an assignment given the sensitivity of our setup as discussed in Sect. 4. Therefore, in the following, we characterize continuum strength using the 30 mean continuum absorption coefficient  ̅ cont , defined as follows: Low-uncertainty constraints on  ̅ cont can only be placed in a number of spectral windows.The selection of such suitable windows is outlined in Sect.3.4.The continuum results for each spectrum were computed as the median of  ̅ cont in all selected spectral windows within 10 cm -1 -wide bins.Finally, an error-weighted mean of  ̅ cont was calculated from the set of 35 ii) A further significant contribution to the error budget results from the IWV measurement uncertainty.The IWV precision was set to 0.8 %, the bias to 1.1 % according to Schneider et al. (2012).The uncertainty resulting from NCEP water vapor profile shape errors was estimated using a comparison of NCEP profiles to radiosonde data (see Part I for details).
iii) The OD uncertainty resulting from NCEP temperature profile errors was deduced from a temperature error covariance matrix estimate for the NCEP resimulation profiles.The error covariance matrix estimate was constructed from the comparison of coincident NCEP profiles to a radiosonde campaign conducted at the site (see Part I for details).iv) Column uncertainties for further trace gases (e.g.CO 2 , CH 4 , N 2 O, and O 3 ) are also included in the uncertainty estimate.
The respective column accuracies are listed in Part I (Tab. 2 therein).
v) The AOD uncertainty is of crucial importance for the OD uncertainty budget in the window regions.As outlined in Sect.2, the AOD uncertainty at air mass 1 is < 0.0025 for the closure data set throughout the 2500 to 7800 cm -1 -range.
The uncertainty contributions i) to v) listed above are linked to the accuracy of the atmospheric state input for LBLRTM calculations.Aside from that, an additional group of error contributions stems from the solar FTIR spectral radiance measurements: vi) The radiance uncertainty due to the radiometric calibration is about 1 -1.7 % and is described in detail in Part II.vii) A further uncertainty contribution results from the solar FTIR measurement noise.It is determined directly from solar FTIR spectra and is among the few uncertainty contributions in the closure experiment of strictly statistical character.It is therefore largely reduced by taking mean results from a larger set of spectra.
viii) Ice layer formation on the liquid nitrogen cooled InSb detector can occur in case of leaks in the detector's vacuum enclosure.Ice formation leads to additional absorption in certain spectral regions, most notably in the 3000 to 3400 cm -1 -range.The uncertainty contribution by varying ice absorption was estimated using lamp spectra routinely recorded with the solar FTIR.Variations in ice absorption during the time period covered by the experiment can be detected as a change of the ratio of measured signal outside and inside the ice absorption band.The maximum variation of this ratio detected in the lamp spectra (1.6 %) was taken as an estimate of the error due to ice absorption.ix) Only a fraction of the solar tracker mirrors is covered by the instrument's field of view (FOV).Due to non-ideal alignment of optical elements, the exact location of the area observed by the instrument on the mirror changes depending on the azimuth and elevation of the instrument's line of sight.The reflectivity of the tracker mirrors features spatial inhomogeneity due to dirt and aging effects.In combination with the moving area covered by the FOV, this results in a variation in measured radiance which leads to spurious variations in the measured OD.An estimate of this uncertainty contribution can be gained using an outgoing laser beam aligned with the instrument's optical axis that enables constraining the mirror area covered by the FOV depending on the instrument's azimuth and elevation.A detailed description of this analysis is given in Part II, Sect.4.1.

Spectra selection
We analyzed spectra recorded under cloud-free conditions in the December 2013 -February 2014 period.Due to inaccuracies in the air mass calculation at high solar zenith angle, air mass was required to be below m = 9.0.
In Sect.3.2, we outlined a source of radiance error in the solar FTIR measurements due to the pointing variation on the tracker mirrors and give an estimate of this contribution.For spectra included in the closure data set, this uncertainty contribution was 5 requested to be negligible compared to other sources of uncertainty, in detail the selection threshold was set to a maximum radiance error of 0.1 %.These selection criteria lead to a final dataset of 52 selected solar FTIR spectra covering an IWV range from 1.4 to 3.3 mm for which the continuum results are presented in Sect. 4. The mean atmospheric state of the closure data set is listed in Appendix A.

Micro-window selection 10
To select suitable windows for continuum quantification, a number of selection criteria were applied to the spectra.Several criteria make use of upper or lower envelopes to the spectra, which were constructed as follows: The upper/lower envelope is defined as the linear interpolation between the highest/lowest values encountered within each 10 cm -1 -wide wavenumber bin.
In detail, the following filtering criteria were applied to the spectra: i) To avoid spectral regions affected by line absorption, only the spectral points with the lowest OD compared to the 15 surrounding spectral region were used.In detail, only points for which the OD exceeds the lower envelope by less than the 2-σ OD uncertainty were used.
ii) Regions around solar lines were excluded.This was implemented as an exclusion of all points for which the extra-atmospheric solar radiance is more than 0.5 % below the upper envelope.
iii) Only regions with low OD uncertainty are included.Therefore, we select points less than 10 % above the lower envelope 20 to the uncertainty.iv) In order to avoid biases of the retrieved continuum due to measurement noise, only regions with a signal-to-rms-noise ratio s/n > 5 were included.
The selection thresholds cited above were adjusted in order to provide sufficiently dense coverage with selected points while maintaining optimum selection quality.Different experimental setups may therefore require different selection threshold 25 values.The final continuum OD results were computed as the median value of all selected spectral points within 10 cm -1 -wide bins.

Results
Figure 1 shows the mean continuum absorption coefficient  ̅ cont determined from the Zugspitze dataset in comparison to the MT_CKD 2.5.2 model predictions and several recent laboratory studies.The mean atmospheric state of the closure dataset is 30 listed in Appendix A, while a table with our results for  ̅ cont and the relative scaling of our results vs. the MT_CKD 2.5.2 predictions and associated uncertainties are available as supplementary material.The assignment of the residual OD to water vapor continuum absorption was made based on two arguments: As outlined in Part I, great care was taken to construct a comprehensive uncertainty budget including thorough estimates of all relevant error contributions to the closure experiment.
Therefore, contributions to the residual OD from other processes than water vapor continuum absorption far beyond the 35 indicated error bars seem unlikely.Furthermore, the IWV dependence of the measured residual OD is consistent with that expected from water vapor continuum absorption.As  ̅ cont includes contributions due to both foreign-and self continuum, it is expected to scale as the sum of a constant and a linear term with respect to water vapor density and therefore also with respect to IWV.The closure dataset covers an IWV range of 1.4 mm < IWV < 3.5 mm, which enables investigation of the Atmos.Chem.Phys. Discuss., doi:10.5194/acp-2016-323, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 April 2016 c Author(s) 2016.CC-BY 3.0 License.
IWV dependence of  ̅ cont .Due to the narrow range of atmospheric temperatures covered in the data set, temperature dependence of the self continuum can be neglected in this analysis.A fraction of 98.6 % of all measured continuum absorption coefficients in the Zugspitze data set are consistent with a combination of constant and linear scaling with respect to IWV, i.e. with being caused by a combination of foreign-and self water vapor continuum.However, 94.2 % of the data are also consistent with a purely constant scaling, i.e. with being solely due to foreign continuum absorption.This is due to the fact that at the atmospheric conditions covered by the data, in all spectral regions where continuum absorption is detectable beyond the experiment's sensitivity, the foreign continuum constitutes by far the dominant contribution, assuming that the partitioning in self-and foreign continuum given by the MT_CKD model is approximately correct.
This analysis shows that the contribution of the self continuum is not unambiguously detectable due to the limited sensitivity of our experiment.We therefore provide values of the mean continuum absorption coefficient  ̅ cont as defined by Eq. ( 6), including contributions from both self-and foreign continuum instead of the more commonly used continuum coefficients c s and c f .The prediction of the MT_CKD 2.5.2 model is shown alongside our results for  ̅ cont in Fig. 1.The MT_CKD 2.5.2-values of  ̅ cont were computed in an analogous way as the values derived from our dataset, i.e.  ̅ cont was calculated according to Eq. ( 6) for the set of atmospheric states encountered in the data set.The results shown in Fig. 1 represent the mean of the MT_CKD predictions for the set of selected measurements.Overall, there is good agreement of our results with the MT_CKD values.Consistency within a 2-σ range is observed for 75 % of the spectral range covered by our measurements.The most apparent discrepancy between MT_CKD and our results occurs in the 2800 to 3000 cm -1 -range, where our results are about a factor of 5 higher than the MT_CKD predictions.However, care has to be taken in the interpretation of this discrepancy since the 2800 to 3000 cm -1 spectral range coincides with a methane absorption band.
Therefore, the accuracy of the continuum result in this range depends on whether the HITRAN error estimate for methane line parameters is correct and whether line coupling effects where treated in sufficiently realistic way in the LBLRTM model.Further significant discrepancies ensue in the 4100 to 4200 cm -1 wavenumber region.The higher measurement results from the Zugspitze data indicate that the MT_CKD-model underestimates the continuum absorption in the wings of the 4000 to 5000 cm -1 window region.In the centers of water vapor absorption bands (i.e.~5200-5400 cm -1 and ~7100-7300 cm -1 ), our results are significantly lower than the MT_CKD-predictions for a number of spectral points.However, the continuum results in these regions are highly sensitive to accurate input and uncertainty estimates for IWV and water vapor line parameters.
Therefore, the slight differences found in the band centers do not provide robust evidence for necessary adjustments of the MT_CKD model.
Figure 1 also includes a comparison of our results to several current laboratory studies using different experimental approaches for continuum quantification.For the comparison,  ̅ cont -values were calculated for the same set of atmospheric states as our results using the continuum coefficients given in the respective studies.For the Mondelain et al. (2015) and Bicknell et al. (2006) results, we used the MT_CKD temperature dependence.For the Ptashnik et al. (2012) and ( 2013) results, we employ the more sophisticated alternative temperature dependence proposed by Rädel et al. (2015), which was deduced from the measurements of Ptashnik et al. (2011).Note, however, that the importance of the continuum temperature dependence is limited (5 to 20 %, see below) for our dataset.This is due to the fact that no temperature dependence is assumed for the foreign continuum, which is by far dominant for most spectral regions given the dry atmospheric conditions encountered in our data set.
Our results show very good agreement with the CDRS-based measurements of Mondelain et al. (2015).For the spectral range examined by Bicknell et al. (2006) with calorimetric-interferometric measurements, only the upper limit of the continuum absorption is constrained by our data, which is consistent with all laboratory studies cited here.In the window regions, our results show good consistency with the BPS-MTCKD 2.0-continuum proposed by Paynter and Ramaswamy (2014).The FTIR-based results of Ptashnik et al. (2012) and(2013) in combination with the temperature dependence proposed by Rädel et al. (2015) lead to higher absorption coefficients than our data in several spectral regions.Significant inconsistencies Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-323, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 April 2016 c Author(s) 2016.CC-BY 3.0 License.
beyond the uncertainty range occur mostly in the wings of water vapor absorption bands, e.g. in the 3200 to 3400 cm -1 , and 4000 to 4200 cm -1 ranges as visible in Fig. 1.In these ranges the absorption coefficients provided by the FTIR laboratory measurements are typically a factor of 2-5 higher compared to our data.
As noted e.g. by Ptashnik et al. (2015), weak lines not included in the line list used for the synthetic spectra calculation may bias the retrieved continuum results.This effect is largely reduced in our analysis due to the spectral selection criteria applied, namely the selection of low-OD windows as outlined in Sect.3.4, criterion i).A drawback of this selection scheme is that our analysis is not well suited for the detection of narrow line-like features in the continuum as reported e.g. by Ptashnik et al. (2011).
An issue not accounted for in our analysis is the uncertainty of the continuum temperature dependence, since an uncertainty estimate is provided neither for the MT_CKD nor the Rädel et al. (2015) relations.However, under the atmospheric conditions covered by our data set and assuming the MT_CKD self-to-foreign ratio, the self continuum contributes only 10 to 30 % to the total continuum absorption at the spectral points for which we detect significant continuum absorption.While no temperature dependence is assumed for the dominant foreign contribution, the temperature dependence of the self continuum changes the mean continuum absorption coefficient by 5 to 20 % within the spectral range considered here and assuming the Rädel et al. (2015) relation.Therefore, it seems unlikely that the differences between the results of Ptashnik et al. ( 2012) and ( 2013) and our data are solely due to inaccuracies in the continuum temperature dependence.

Summary and conclusions
We presented a quantification of the water vapor continuum absorption in the NIR spectral range (2500 to 7800 cm -1 ) from an atmospheric radiative closure experiment.To our knowledge, prior to this study no precise constraints on the continuum absorption under atmospheric conditions were available in this spectral range.The mean continuum absorption coefficient was determined from a set of 52 solar FTIR spectra.The method to achieve continuum quantification relies on the use of radiometrically calibrated spectra obtained by the method presented in Part II.Continuum constraints are presented in the wings and some spectral windows in the centers of water vapor absorption bands.Due to the low IWV encountered throughout our measurement period, only the upper boundary of the continuum can be constrained in the centers of atmospheric windows.
The results show good consistency with the widely used MT_CKD 2.5.2 model, although they indicate a need for increasing the absorption strength compared to the model in some spectral regions such as the wings of water vapor absorption bands.
Our results were compared to a number of recent laboratory studies using different experimental techniques.A first group of studies relies on FTIR cell measurements.Our data generally indicate lower continuum absorption than implicated by the studies of Ptashnik et al. (2012) and ( 2013) in combination with the self continuum temperature dependence given in Rädel at al. (2015).However, significant deviations from these studies only occur in the wings of water vapor absorption bands.
Further experimental techniques used for continuum quantification in laboratory experiments include CRDS.A comparison to the CDRS results of Mondelain et al. (2015) in the spectral region around 4250 cm -1 shows very good agreement to our findings.Bicknell et al. (2006) quantified continuum absorption using a calorimetric-spectrometric technique.While our results agree to the findings of Bicknell et al. (2006), their measurements cover spectral regions where only an upper limit for the continuum absorption can be deduced from our data.
An assignment of the detected continuum absorption to self-and foreign continuum requires improvements of the experimental sensitivity or a data set covering a broader range of IWV values.The same is true for a detection of the continuum beyond the uncertainty limit in window regions which requires improved sensitivity or a data set covering higher IWV values.Aside from these limitations, our results provide a valuable foundation for an improved quantification of the NIR water vapor continuum under atmospheric conditions.Most notably, our analysis provides a tool for atmospheric Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-323,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 25 April 2016 c Author(s) 2016.CC-BY 3.0 License.
52 spectra selected according to the criteria listed in Sect.3.3.The uncertainty estimate of the continuum results is presented in Sect.3.2.Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-323,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 25 April 2016 c Author(s) 2016.CC-BY 3.0 License.3.2.Uncertainty estimateAn interpretation of the residual OD and assignment to causative absorption processes requires a comprehensive uncertainty budget of the closure experiment.The uncertainty estimate of our experimental setup is described in detail in Part I except for contributions only relevant for the NIR closure measurements.The total residual uncertainty and its various contributions are also shown in Part I, Fig. 5.A description of the NIR-specific contributions and a brief outline of the remaining sources of uncertainty are given below.All uncertainty values are quoted on 2-σ confidence level.i) Absorption line parameter uncertainties of water vapor and other absorbing species.These uncertainties were set to the mean value of the uncertainty range specified by the error codes provided in the line parameter file (aer_v3.2) provided alongside the LBLRTM model.Line parameter uncertainties are the dominant contribution to the error budget within absorption bands.
Mean atmospheric state of the closure data set.The data set was selected from Zugspitze solar FTIR spectra measured from Dec 2013 -Feb 2014 and contains 52 spectra.Spectra selection criteria are listed in Sect.3.3.