Description of the MPI-Mainz HO retrieval (Version 5., March ) Thomas Wagner, Steffen Beirle, Kornelia Mies Contact: Thomas Wagner MPI for Chemistry Joh.-Joachim-Becher-Weg 7 D-558 Mainz Germany Phone: +49 ()63 3567 Fax: +49 ()63 35388 email: thomas.wagner@mpic.de Content: Description of the retrieval Choice of the settings for the DOAS fitting Saturation correction Correction for effects of different profile shapes of O ans HO Selection of mostly cloud-free observations Validation
Description of the retrieval HO columns have been retrieved from ERS- GOME and ENVISAT SCIAMACHY instruments by various groups using different settings (Noël et al., 999,,, 4, 5, 8; Mieruch et al., 8; Casadio et al., ; Maurellis et al., ; Lang et al., 3; 4; Lang, 3; Wagner et al., 3, 5, 6). These algorithms differ not only in the used wavelength ranges (see Fig. ), but also in the complexity of the retrievals. Most of the algorithms make use of simultaneously retrieved O or O4 absorptions to correct for the influence of clouds. A general feature is that a modified DOAS type retrieval has to be performed, because the strong and narrow-band H O (and O) absorption lines are not fully resolved by the instrument. The respective modifications are either applied in the spectral retrieval (e.g. weighting function modified DOAS) or appropriate corrections are successively applied. The SCIAMACHY water vapor retrieval described in this document analyses the H O and O absorptions in the red spectral range (64-683nm). Our water vapor product was developed with the aim to study time series and temporal trends of the atmospheric water vapor content with high accuracy. Thus no a-priori information or information from external data sets is included in our algorithm. On the other hand, this approach has also some drawbacks, especially relatively large uncertainties of individual observations mainly related to clouds. In principle, satellite observations in the visible and NIR spectral range are sensitive to the whole tropospheric HO column. However, especially due to the shielding effect of clouds, the sensitivity for the surface-near layers can be systematically reduced. In the current version of our retrieval, the height-dependent sensitivity can not be quantified (e.g. by averaging kernels), because no explicit radiative transfer simulations are applied (see below). However, the rather good agreement with independent data sets (see below) indicates that H O concentrations close to the surface are well captured by our retrievals. After the spectral retrieval the derived initial trace gas SCDs are corrected for the spectral saturation effect and finally converted into total atmospheric VCDs. Wavelength ranges for different HO analyses Lang et al., 7 Lang et al., Wagner et al., 3 Noel et al., 999 Casadio et al., Fig. Wavelength ranges for HO analyses used by different research groups
Choice of the settings for the DOAS fitting The fitting parameters used for the processing of SCIAMACHY spectra are based on those used for the analysis of GOME- observations [Wagner et al., 3; 5; 6]. In the chosen spectral range, medium strong absorption lines of HO (around 65nm) and O (around 63nm) exist. Compared to other spectral ranges (see Fig. ), these absorptions are still strong enough to ensure a high signal to noise ratio. At the same time, they are substantially weaker than e.g. at longer wavelength ranges, thus avoiding strong nonlinearities in the spectral fitting process. Another advantage of the selected fitting range is that it includes also an absorption band of the oxygen dimer O4 (at 63nm). Besides the trace gas reference spectra, also several other spectra are included in the spectral fitting process. First, a synthetic Ring spectrum is included to correct for the so called Ring effect caused by inelastic Raman scattering on air molecules. In addition, also three reference spectra for different vegetation types are included. Especially over surfaces with strong vegetation, it turned out that spectral features caused by vegetation show strong interference with the atmospheric absorbers [Wagner et al., 7]. Since the vegetation spectra describe the spectral reflectance, the natural logarithm of the reflectance spectra are used in the spectral fitting. To correct for possibly non-perfect polarisation correction of the SCIAMACHY spectra, also two polarisation sensitivity spectra from the keydata are included in the spectral analysis. Finally, an inverse solar spectrum is included to correct for possible intensity offsets, e.g. caused by instrumental straylight. The parameters for the spectral analysis are summarized in Table. An example of the spectral DOAS analysis is shown in Fig.. The absorption structures of O, HO and O4 are clearly visible. The residual contains noise, but also systematic features at 63nm and 65nm caused by imperfect match of the O absorption. However, compared to the total strength of the O and HO absorptions these features are weak. From the spectral retrieval, also the fitting errors (standard deviation) are determined and stored in the data files (see product specification document). Also an estimate for the total error is given, which takes into account the individual fitting errors and the effect of clouds (see product specification document). Fig. Example of a spectral DOAS analysis. Displayed are the fitted reference spectra (black lines) scaled to the respective spectral structures in the measured spectrum (red). 3
Table. DOAS settings used for HO spectral retrieval Property Fitting interval Sun reference Wavelength calibration Selected parameter 64-683.nm nm SCIAMACHY Sun irradiance Wavelength calibration of sun reference optimized by NLLS adjustment on convolved Kurucz solar spectrum. Remarks Absorption cross-sections HO HITRAN 6 (Rothmann et al.) 9K, convoluted with wavelength dependent instrument function O HITRAN 4 (Rothmann et al.) 9K, convoluted with wavelength dependent instrument function O4 Greenblatt et al., 99 Interpolated Additional spectra Ring spectrum Synthetic Ring spectrum calculated from sun spectrum using the MFC software (Gomer et al., 993) Conifers ASTER Spectral Library through the courtesy of the Jet Propulsion Laboratory, California Institute of Technology, http://speclib.jpl.nasa.gov/ Logarithm, interpolated Grass ASTER Spectral Library through the courtesy of the Jet Propulsion Laboratory, California Institute of Technology, http://speclib.jpl.nasa.gov/ Logarithm, interpolated Decidous ASTER Spectral Library through the courtesy of the Jet Propulsion Laboratory, California Institute of Technology, http://speclib.jpl.nasa.gov/ Logarithm, interpolated ZETA_NAD instrument keydata Interpolated OBM_s_p instrument keydata Interpolated Polynomial 3rd order (4 parameters) Intensity offset correction Inverse solar spectrum3 Kurucz, R.L., I. Furenlid, J. Brault, and L. Testerman, Solar flux atlas from 96 nm to 3 nm, National Solar Observatory Atlas No., 984. keydata version 3. 3 from SCIAMACHY solar spectrum 4
Saturation correction Before the retrieved HO SCD is converted into a vertical column density, a correction of the non-linearity caused by the limited spectral resolution of the instrument is performed. For that purpose, the non-linearity is quantitatively described by modeling of the DOAS analysis procedure. First, a high-resolved absorption spectrum is calculated from the high resolved HO cross section applying the Lambert-Beer law: I( λ ) = I ( λ ) exp( SCD H O,true σ O ( λ ) Here, a set of realistic HO SCDs (SCDHO,true) has to be assumed. Then, the respective absorption spectra (and the HO absorption cross section) are convoluted with the SCIAMACHY instrument function. Finally, the convoluted spectra are analysed using the exact DOAS-settings as used for the analysis of the measured spectra. The fit yields a H O SCD (HOfit), which is smaller than the originally assumed H Otrue. For small SCDs, HOfit depends almost linearly on HOtrue. However, for larger SCDs, H Ofit is systematically smaller than HOtrue. Fig. 3 shows the relationship of both SCDs. The determined relationship is used to correct the HOfit determined in the DOAS fit to get the true atmospheric H O SCD (HOtrue). A similar correction is also applied to the retrieved O SCD (Fig. 4). The formulae for the saturation corrections are: SCDHO,true =.596-4 * (SCDHO,fit) + 9.6779 - * SCDHO,fit + 5.7495 + SCDO,true =.6379-5 * (SCDO,fit)3 +.54986-6 * (SCDO,fit) + 9.457654 - * SCDO,fit +.6753 +3 For the conversion of O SCDs to O AMFs a O VCD of 4.5657 4 was assumed. 3E+3 Measured HO SCD [molec/cm²] y = 7E-49x3 - E-4x +.99x + E+ R = E+3 E+3 E+3 E+3 3E+3 4E+3 True HO SCD [molec/cm²] Fig. 3 Dependence of the retrieved HO SCD on the true HO SCD. 5 5E+3
AMF from spectral analysis 5 4 3 3 4 5 6 7 8 Modelled AMF Fig. 4 Dependence of the retrieved O SCD on the true O SCD. The SCD is expressed as air mass factor here, because the atmospheric O VCD is almost constant. Correction for effects of different profile shapes of O and HO In the last step, the corrected HO SCD is converted into the HO VCD. For this purpose, a measured AMF is used based on the retrieved O SCD (Wagner et al., 3, 5, 6). Since the atmospheric O VCD is (almost) constant, the ratio of the retrieved (corrected) O SCD and the O VCD yields the measured AMF for a given observation. The application of a measured AMF has the important advantage that the necessary information especially on the effects of clouds is taken from the satellite observation itself. However, this correction has also its limitations. First, since the relative profiles of H O and O are systematically different, the respective AMF show also systematic differences, with the O AMF being higher than the HO AMF. This effect is accounted for by applying a correction function (depending on the solar zenith angle, viewing geometry and the surface albedo) to the retrieved H O VCD (see Figs. 5 and 6). These correction functions is derived from radiative transfer calculations based on an average (relative) HO profile (see Fig. 7) calculated from average profiles of lapse rate and relative humidity (see Wagner et al., 6) and an O profile from the US standard atmosphere. Note that the original correction function was calculated a surface albedo of % and cloud free conditions. Similar results were obtained assuming more realistic values of 3% (over ocean) and a cloud fraction of 5% (cloud with OD of 5 between and 4km altitude). Finally, another correction function for the effect of changing surface albedo is applied to the HO VCDs (see Fig. 8). Note that this correction was not applied in the first version of our retrieval (Wagner et al., 6). The surface albedo used for the correction is taken from monthly varying albedo maps, which are composite of albedo derived from GOME- observations (Koelemeijer et al., 3) for high latitude (>5 ), and SCIAMACHY observations (Grzegorski, 9) at mid and low latitudes (<4 ). For the transition between 4 and 5, both products are interpolated linearly. Over ocean, the albedo is set to 3%. An example for January is shown in Fig. 9. It should be noted that especially for observations over low layer clouds with large cloud fractions, the effect of the different relative profile shapes can cause large errors (up to more than %). However, for high clouds and especially low cloud fractions, the relative cloud effect on the O and HO absorptions is similar and the application of the measured AMFs leads to more accurate results. 6
7 Ratio O AMF / HO AMF 6 5 4 3 3 4 5 6 7 8 9 SZA [ ] Fig. 5 Ratio of the AMF for O and HO as a function of the solar zenith angle (nadir viewing direction). The ratio is determined from radiative transfer simulations assuming a H O profile as shown in Fig. 7. 3.5 SZA: Reihe Reihe 4 Reihe3 5 Reihe4 6 Reihe5 7 Reihe6 75 Reihe7 8 Reihe8 3 Ratio.5.5-5 -3 - -9 LOS angle -7-5 -3 Fig. 6 Dependence of the correction factor on the line of sight (LOS) angle of the satellite instrument for different solar zenith angles. Here only values for a relative azimuth angle of zero are shown for simplicity. In the retrieval, however, the actual relative azimuth angle of individual measurements is taken into account. Note that for SCIAMACHY (and GOME-), only LOS between about and 6 occur. Larger deviations of the LOS angles from 9 occur for GOME-. 7
Fig. 7 HO profile used for the radiative transfer calculations (see Wagner et al., 6). For tropical regions a HO VCD of 3 molec/cm² and for polar regions a HO VCD of 3 molec/cm² was assumed..5 Correction factor.5.5..4.6.8 Surface albedo Fig. 8 Correction function taking into account the surface albedo (the H O VCDs are divided by this function). 8
Fig. 9 Surface albedo in the red spectral range for January. Since our HO data product is not based on explicit radiative transfer calculations for individual measurements, a straight forward estimation of the product uncertainty is difficult. However, in the product, a total error is given based on four potential error sources: -the fit error of HO -the fit error of O -general uncertainties of the spectral retrieval, e.g. related to uncertainties of the spectroscopic data. This error is estimated to %. -uncertainties related to radiative transfer effects, especially due to clouds. For this uncertainty, an empirical formula taking into account the measured O SCD is used: errrtm = (SCDO,meas - SCDO,max(SZA)*.6)/ 5molec/cm² This error increases with decreasing O SCD (indicating strong cloud shielding). Note that this formula can only provide a rough estimate. However, from sensitivity studies (e.g. Wagner et al., 5, 6) it was shown that for measurements with low O SCDs the retrieved HO VCDs systematically underestimate the true HO VCD (see also next section). The total (relative) error is derived from the individual error terms by the following formula: errrel,total = fit H O + fit O +. + errrtm Selection of mostly cloud-free observations Since most of the atmospheric HO column density is located close to the surface, cloud shielding has a strong influence on the retrieved HO VCD (see Wagner et al., 5). For 9
large cloud fractions, the satellite observations typically underestimate the true HO VCDs. Thus, for the SCIAMACHY HO product, only (mostly) cloud-free observations are selected. To identify strong cloud effects, we consider the simultaneously retrieved O SCD. Only satellite observations are considered, for which the retrieved O SCD is above 8% of the maximum O SCD for the respective solar zenith angle (SZA), see Figs. and. Optical depth O absorption.3.5..5..5 3 4 5 6 7 8 9 SZA Fig Measured optical depth of the O absorption at 63 nm along a satellite orbit as a function of the SZA. The drawn line shows the maximum O absorption for (mostly) nadir observations. (here observations of the GOME- instrument are shown (Wagner et al., 5). The LOS dependence is considered by multiplication of an additional function (see Fig. ).4 Correction factor.3 6 7 - -9 LOS [ ] 4 75 5 8...9-5 -3-7 -5-3 Fig. LOS dependence of the threshold for mainly cloud-free observations. In Figure an example for the monthly mean HO VCDs using only mostly cloud-free observations is shown (for March 3).
Fig. : Monthly mean HO VCD derived from SCIAMACHY for mostly cloud-free conditions (March 3). Validation We compared HO VCDs retrieved from SCIAMACHY with results from GOME- and GOME-. Results from GOME- and GOME- have been validated using observations from SSM/I observations and radiosondes (Wagner et al., 5; Slijkhuis et al. 8, EUMETSAT, ). For mainly cloud-free conditions, (effective cloud fraction <%), the mean relative error (compared to radio sonde observations) is typically < %. Over regions with high albedo (low albedo) our retrieval tends to underestimate (overestimate) the true water vapor column. In Fig. 3 the HO VCD from GOME- for March 3 is shown. It was retrieved with the same settings as described in Table (but for the GOME- retrieval no polarisation spectra were included). In Figs. 4 and 5 the difference between SCIAMACHY and GOME results for March 3 are presented. Fig. 6 shows time series of HO VCDs for the different sensors for different latitude bands. During the overlap periods good agreement between the results of the different instruments is found. Good agreement is also obtained from correlations studies of between the different sensors (Fig. 7).
Fig. 3 Monthly mean HO VCD derived from GOME- for mostly cloud-free conditions (March 3). Fig. 4 Difference between the monthly mean HO VCDs from SCIAMACHY (Fig. 7) and GOME- (Fig. 8) for March 3. Most of the noise-like patterns are related to the different spatial sampling of both instruments.
Fig. 5 Like Fig. 4, but only for collocated observations Fig. 6 Time series of monthly mean HO VCDs for different latitude bands from 996-9 from GOME-, SCIAMACHY and GOME-. Good agreement between the overlap periods is found. 3
Fig. 7 Correlation analyses between SCIAMACHY results (x-axis) and GOME- and GOME-. Recommendations for product validation Our HO product was extensively validated for different measurement conditions (e.g. EUMETSAT, ). Additional validation exercises should focus on measurements over regions with high surface albedo (e.g. deserts) or elevated terrain. Known issues Our water vapor retrieval is optimised to yield stable and unbiased time series of the atmospheric HO VCD. This was the reason to chose the rather simple retrieval scheme without using additional data sets (e.g. on cloud properties). As a consequence, however, in specific cases, also large deviations of the retrieved H O VCD from the true atmospheric value can occur. In spite of the application of measured AMFs, the largest errors are expected to occur over low lying clouds (e.g. at the west coasts of South America or Africa), because of the different profile shapes of H O and O. For observations above low clouds most of the atmospheric HO VCD is shielded, while most of the O VCD is still seen by the satellite instrument. Thus in such cases the use of a measured AMF leads to a systematic underestimation of the retrieved HO VCD. Over regions with high albedo (low albedo) our retrieval tends to underestimate (overestimate) the true water vapor column [EUMETSAT, ]. While high fitting errors (given in the data product) indicate measurements with high errors of the HO VCD product, additional errors might be introduced in cases with strong differences of the HO profile from the assumed profile. An errors estimation including different error 4
sources are given in the result files (see product specification file). These errors are difficult to quantify. In particular, they can occur although the fitting errors are small. These errors Currently we are working on an improved version of the H O retrieval including a better cloud correction. Up to now, no explicit treatment of measurements over elevated surface terrain is included in our retrieval. Thus over high mountains areas (>m) the HO VCD is systematically underestimated. References Casadio, S., Zehner, C., Pisacane, G., and Putz, E., Empirical retrieval of the atmospheric air mass factor (ERA) for the measurement of water vapour vertical content using GOME data, Geophys. Res. Lett., 7, 483 486,. EUMETSAT, Algorithm Theoretical Basis Document for GOME- Total Column Products of Ozone, NO, tropospheric NO, BrO, SO, HO, HCHO, OClO and Cloud Properties (GDP 4.4 for O3M-SAF OTO and NTO, http://atmos.caf.dlr.de/gome/docs/dlr_gome_atbd.pdf,. Greenblatt, G. D., J. J. Orlando, J. B. Burkholder, and A. R. Ravishankara, Absorption measurements of oxygen between 33 and 4 nm. J. Geophys. Res., 95, 8 577 8 58, 99. Grzegorski, M., Cloud retrieval from UV/VIS satellite instruments (SCIAMACHY and GOME), PhD thesis, University of Heidelberg, 9. Gomer, T., T. Brauers, F. Heintz, J. Stutz, and U. Platt, MFC user manual, version.98, Institut für Umweltphysik, University of Heidelberg, Germany, (http://ich33.ich.kfajuelich.de/mfc/mfc.htm), 993. Koelemeijer, R. B. A., J. F. de Haan, and P. Stammes, A database of spectral surface reflectivity in the range 335--77 nm derived from 5.5 years of GOME observations, J. Geophys. Res., 3, Vol. 8 (D), doi.9/jd49 Lang, R., Williams, J. E., van der Zande, W. J., and Maurellis, A.N., Application of the Spectral Structure Parameterization technique: retrieval of total water vapour columns from GOME, Atmos. Chem. Phys., 3, 45 6, 3. Lang, R., Water vapour measurements from space, Status report, MPI for Chemistry, Mainz, Germany, 3. Lang, R., and M. Lawrence, Evaluation of the hydrological cycle of MATCH driven by NCEP reanalysis data: comparison with GOME water vapor field measurements, Atmos. Chem. Phys. Discuss., 4, 797-7984, 4. Liu, W., W. Tang, F.J. Wentz, Precipitable water and surface humidity over global oceans from Special Sensor Microwave Imager and European Center for Medium Range Weather Forecasts, J. Geophys. Res., 97, 5-64, 99. Maurellis, A.N., R. Lang, W.J. van der Zande, I. Aben, W. Ubachs, Precipitable Water Column Retrieval from GOME data, Geophys. Res. Lett., 7, 93-96,. 5
Mieruch, S., Noël, S., Bovensmann, H., and Burrows, J. P.: Analysis of global water vapour trends from satellite measurements in the visible spectral range, Atmos. Chem. Phys., 8, 4954, 8. Noël, S., M. Buchwitz, H. Bovensmann, R. Hoogen, J.P. Burrows, Atmospheric Water Vapour Amounts Retrievd from GOME Satellite data, Geophys. Res. Lett., 6, 84-844, 999. Noël, S., Bovensmann, H., and Burrows, J. P., Water vapour retrieval from GOME data including cloudy scenes, Proc. ENVISAT/ERS Symposium, Gothenburg,. Noël, S., Buchwitz, M., Bovensmann, H., and Burrows, J. P., Retrieval of Total Water Vapour Column Amounts from GOME/ERS- Data, Adv. Space Res., 9, 697 7,. Noël, S., Buchwitz, M., and Burrows, J. P.: First retrieval of global water vapour column amounts from SCIAMACHY measurements, Atmos. Chem. Phys., 4, -5, 4. Noël, S., Buchwitz, M., Bovensmann, H., and Burrows, J. P.: Validation of SCIAMACHY AMC-DOAS water vapour columns, Atmos. Chem. Phys., 5, 835-84, 5. Noël, S., Mieruch, S., Bovensmann, H., and Burrows, J. P.: Preliminary results of GOME- water vapour retrievals and first applications in polar regions, Atmos. Chem. Phys., 8, 5959, 8. Rothman, L., A. Barbe, D. Chris Benner, L.R. Brown, C. Camy-Peyret, M.R. Carleer, K. Chance, C. Clerbaux, V. Dana, V.M. Devi, A. Fayt, J.-M. Flaud, R.R. Gamache, A. Goldman, D. Jacquemart, K.W. Jucks, W.J. Lafferty, J.-Y. Mandin, S.T. Massie, V. Nemtchinov, D.A. Newnham, A. Perrin, C.P. Rinsland, J. Schroeder, K.M. Smith, M.A.H. Smith, K. Tang, R.A. Toth, J. Vander Auwera, P. Varanasi, K. Yoshino, The HITRAN molecular spectroscopic database: edition of including updates through. J. Quant. Spectrosc. Radiat. Transfer 8, 5-44 (3). Slijkhuis, S., Comparison of HO DLR/GOME/HO/ Iss., 6..8 retrievals from GOME and GOME-, Wagner, T., S. Beirle, M. Grzegorski, S. Sanghavi, U. Platt, El-Niño induced anomalies in global data sets of water vapour and cloud cover derived from GOME on ERS-, J. Geophys. Res,, D54, doi:.9/5jd597, 5. Wagner T., S. Beirle, M. Grzegorski, U. Platt, Global trends (996 3) of total column precipitable water observed by Global Ozone Monitoring Experiment (GOME) on ERS- and their relation to near-surface temperature, J. Geophys. Res.,, D, doi:.9/5jd653, 6. Wagner, T., Beirle, S., Deutschmann, T., Grzegorski, M., and Platt, U.: Satellite monitoring of different vegetation types by differential optical absorption spectroscopy (DOAS) in the red spectral range, Atmos. Chem. Phys., 7, 69-79, 7. 6