Global intercomparison of SCIAMACHY XCH4 with NDACC FTS what can we learn for GOSAT validation by TCCON FTS? R. Sussmann, M. Rettinger, F. Forster, T. Borsdorff took some notes from observations beside the road : SCIA GOSAT mir nir NDACC TCCON dry air column O2 column SFIT GFIT microwindows full band profiling VMR-profile scaling smile frown annual cycle true airmass dep. artifact bias? diurnal variation true spectroscopic artifact diurnal cylce precision?
Global intercomparison of SCIAMACHY XCH4 with NDACC FTS what can we learn for GOSAT validation by TCCON FTS? R. Sussmann, M. Rettinger, F. Forster, T. Borsdorff Outline: global intercomp. of SCIA XCH4 versus 12 NDACC MIR-FTIR stations first GFIT results from nearly 2 years TCCON operations at Garmisch annual cycles: SCIA versus g.-b. MIR versus g.-b. NIR diurnal variations: g.-b. MIR versus g.-b. NIR summary
XCH4 anomaly Conclusions from ACP (2005)*: SCIA WFMD v0.41 versus Zugspitze MIR-FTIR Zugspitze FTIR: precision of daily means is 0.4 % can retrieve true annual-cycle amplitude: 1 % can retrieve true day-to-day variability: 1 % 1.2 1.1 SCIA 2000 km daily means SCIA 1000 km daily means Zugspitze FTIR daily means SCIAMACHY WFMD v0.41: time-dependent biases 3 %/month due to icing problems in channel 8 could not retrieve true day-to-day variability precision of the daily means 0.6 % (calculated from the individual pixels, 1000-km selection radius) would be sufficient to capture true day-to-day variability and the annual cycle - if the icing issue could be eliminated. 1.0 0.9 2003 0.8 Dec-02 Mar-03 May-03 Aug-03 Nov-03 Date * Sussmann, Stremme, Buchwitz, de Beek, Atmos. Chem. Phys, 5., 2419 2429, 2005
SCIAMACHY nadir spectrum & fitting windows CH4: 1.66 m O2 CO2: 1.58 m CH4, N2O CO
New questions for validating SCIA IMAP-DOAS v49 & WFM-DOAS v1.0 After validation in 2005, the SCIAMACHY icing issue was overcome by reimplementing SCIA retrievals from channel 8 to channel 6. Two years of SCIA data are now available: 2003 & 2004. Possible to revisit the questions 1: Can the new SCIA channel 6 retrievals capture true day-to-day variability? 2: Can the new SCIA retrievals reflect annual cycles as seen by FTIR? 3: Is there a latitudinal dependency of biases for IMAP-DOAS v49 and WFM-DOAS v1.0? Make sure that g.-b. CH4 columns are retrieved in a perfectly consistent manner at all globally distributed ground-based MIR-FTIR stations.
The 12 NDACC-MIR-FTIR stations of our SCIA MIR-FTIR intercomp. study station latitude longitude station altitude Spitzbergen Thule Kiruna Harestua Bremen Zugspitze Garmisch Jungfraujoch Toronto Izaña St-Denis Wollongong 78.92 N 76.53 N 67.84 N 60.22 N 53.11 N 47.42 N 47.48 N 46.55 N 43.66 N 28.30 N 20.90 S 34.41 S 11.92 E 68.74 W 20.41 E 10.75 E 8.85 E 10.98 E 11.06 E 7.99 E 79.40 W 16.48 W 55.48 E 150.88 E 20 m 225 m 419 m 596 m 29 m 2964 m 745 m 3580 m 174 m 2367 m 50 m 40 m Number of columns in 2003/2004 113 177 338 1234 179 999 498 702 185 207 141 633 tropopause height 8.95 km 8.51 km 9.62 km 10.20 km 10.74 km 11.25 km 11.25 km 11.38 km 13.25 km 14.44 km 15.66 km 12.53 km
MIR-FTIR CH4 retrieval homogenization: one common micro-window set: 5 windows 2613-2921 cm -1 1.0 0.9 0.8 Total CH4 CO2 Solar HDO H2O 1.04 1.02 1.00 0.98 0.96 0.94 0.92 0.90 0.88 0.86 0.84 CH4 Total HDO CO2 CO22 Solar 1.08 1.06 1.04 1.02 1.00 0.98 0.96 0.94 0.92 0.90 0.88 0.86 CH4 Total Solar N2O O3 1.1 1.0 0.9 0.8 0.7 1.2 1.1 1.0 Total CH4 0.9 NO2 OCS 0.8 Solar 0.7 0.6 Total CH4 HDO NO2 H2O Solar 2613.8 2614.5 2615.2 0.82 0.80 2650.5 2650.6 2650.7 2650.8 2650.9 2651.0 2651.1 2651.2 2651.3 2651.4 0.84 0.82 0.80 2835.50 2835.55 2835.60 2835.65 2835.70 2835.75 2835.80 0.6 2903.6 2903.7 2903.8 2903.9 2904.0 2904.1 0.5 0.4 2921.0 2921.1 2921.2 2921.3 2921.4 2921.5 2921.6 harmonized treatment of interfering species: same binput-file identical spectroscopy for all: same cfgls (HIT04 & Hase update) common source of pt-input profiles: NCEP one consistent set of a priori profiles: Toon with Meier correction one consistent set of regularization matrices and altitude grids: Tikhonov-L 1 on the %-VMR scale
Harmonizing the g.-b. MIR XCH4 data set: how to calculate the air column in situ p measurements not continuously available at all sites for 2003/2004 radio sondes not available for all sites NEP p-profiles and ECMWF in situ p available NCEP favored because: The forward model of the g.-b. FTIR retrievals uses NCEP pressure profiles to calculate the airmass-profile ( fas.mas file) which is internally used to transfer VMR profiles to partial column (CH4) profiles I.e., the air columns calculated from NCEP are i) readily available, and ii) perfectly consistent to the CH4 total columns retrieved.
Harmonizing the g.-b. MIR XCH4 validation data set: dry versus wet XCH4 previous SCIA validation studies derived XCH4 from FTIR via rationing by wet air columns (derived from pressure data). inconsistency to SCIAMACHY, which retrieves dry XCH4. FTIRs should substract the water vapor column from the air column: where do we get water columns from? not all FTIR groups have a maturated FTIR water retrieval; are NCEP water columns reliable?
Water vapor column (cm-2) Harmonizing the g.-b. MIR XCH4 validation data set: how to calculate dry XCH4 validate NCEP water columns against Optimized Zugspitze FTIR water vapor column retrieval (ACPD 2009, IRWG science talk) 4.5E+22 4.0E+22 Zugspitze FTIR NECP 3.5E+22 3.0E+22 2.5E+22 2.0E+22 1.5E+22 1.0E+22 NCEP water ok, use for all stations 5.0E+21 0.0E+00 20.12.02 30.03.03 08.07.03 16.10.03 24.01.04 03.05.04 11.08.04 19.11.04 Date
XCH4 Harmonizing the g.-b. MIR XCH4 validation data set: dry versus wet XCH4 impact on bias, annual cycle Wollongong 2004 1.75E-06 1.73E-06 FTIR dry XCH4 (minus NCEP water) FTIR wet XCH4 1.71E-06 1.69E-06 1.67E-06 1.65E-06 1.63E-06 1.61E-06 1.59E-06 1.57E-06 1.55E-06 01.01.2004 10.04.2004 19.07.2004 27.10.2004 Date
MIR-FTS, homogenization: data selection / quality control FTIR spectra quality (rms of the spectral fitting residuals) harmonized selection thresholds for all FTIR data-sets
MIR-FTS, homogenization: averaging kernels In order to reduce intercomparison errors, the retrieved FTIR CH4 vertical profiles (dofs 2-3) were folded by the total column averaging kernels from the SCIAMACHY retrievals (dofs = 1) according to Rodgers and Connor (JGR, 2003)
Intercomp. results: global XCH4 biases SCIA - MIR-FTIR (same day, 500 km radius) 2004 open symbols: without averaging kernel correction
Intercomp. results: validation plots (daily means FTIR, SCIA-200, -500, -1000 km)
Intercomp. results: validation plots (daily means FTIR, SCIA-200, -500, -1000 km)
Intercomp. results: validation plots (daily means FTIR, SCIA-200, -500, -1000 km)
Intercomp. results: validation plots (daily means FTIR, SCIA-200, -500, -1000 km)
Intercomp. results: validation plots (daily means FTIR, SCIA-200, -500, -1000 km)
XCH4 Does SCIA just reflect dynamical prior, i.e., is not able to retrieve the 1 % CH4 changes above clean sites? 1.90E-06 1.85E-06 1.80E-06 Red: XCH4 calculated from fixed a priori profile taking daily pt-profiles and tropopause altitudes into account 1.75E-06 1.70E-06 1.65E-06 1.60E-06 1.55E-06 1/1/03 20/2/03 11/4/03 31/5/03 20/7/03 8/9/03 28/10/03 17/12/03 probably not: WFM-DOAS uses 1 fixed a priori profile
XCH4 XCH4 annual cycles: Different MIR instruments and differing ray tracing algo s. impact on annual cycles? No. 1.80E-06 Jungfraujoch FTIR Zugspitze FTIR daily means 1.75E-06 1.70E-06 1.65E-06 1.60E-06 1.55E-06 1/1/03 20/2/03 11/4/03 31/5/03 20/7/03 8/9/03 28/10/03 17/12/03
XCH 4 (ppbv) Annual cycles: g.-b. XCH4 MIR versus NIR 1860 1840 1820 1800 1780 XCH4_GFIT_ver_Dec 06 XCH4_GFIT_ver_Feb_09 XCH4_MIR same spectra (!) same instrument (!) 1760 1740 1720 1700 1680 1660 1640 1620 01.06.2007 01.12.2007 01.06.2008 01.12.2008 01.06.2009
Dry Airmass (*1E25) Annual cycles: trust in dry airmass derived from O2, NCEP and p measurements 2,25 2,20 2,15 NIR: GFIT ver Dec 06 NIR: GFIT ver Feb 09 MIR: calculated from NCEP profiles MIR: calculatd from surface p 2,10 2,05 2,00 1,95 1,90 1,85 8-1-2007 9-1-2007 10-1-2007 11-1-2007
Annual cycles: airmass-dependent artifacts taken from Geoff s 2006 talk taken from Geoff s 2006 talk
Annual cycles: airmass-dependent artifacts taken from Geoff s 2006 talk taken from Geoff s 2006 talk
Annual cycles: airmass-dependent artifacts taken from Geoff s 2006 talk Unfortunately, virtually all modern retrieval algorithms use NLLS methods, which do not match equivalent widths. Since NLLS methods minimize the square of the residuals, and since the largest residuals typically arise at line center, NLLS methods try harder to fit the line center than elsewhere on the line profile. taken from Geoff s 2006 talk
XCH4 (ppbv) XCH4 diurnal variation: g.-b. MIR versus NIR 1820 1800 XCH4_NIR_GFIT_ver_Dec_06 XCH4_NIR_GFIT_ver_Feb_09 XCH4_MIR 1780 1760 1740 1720 1700 1680 1660 4 6 8 10 12 14 16 18
CH4 Total Coumn (*1E19) O2 Total Column (*1E24) Diurnal variation O2 and CH4: old GFIT versus newest version GFIT 4,5 O2_GFIT_ver_Dec_06 O2_GFIT_ver_Feb_09 4,4 3,80 4,3 4,2 3,75 3,70 CH4_GFIT_ver_Dec_06 CH4_GFIT_ver_Feb_09 4,1 3,65 4 6 8 10 12 14 16 18 3,60 3,55 3,50 3,45 3,40 3,35 3,30 4 6 8 10 12 14 16 18
XCH4 (ppbv) XCH4 (ppbv) XCH4 (ppbv) XCH4 (ppbv) XCH4 (ppbv) XCH4 (ppbv) Diurnal cycles: measured in alternating NIR-MIR mode with Garmisch FTS NIR MIR 1780 1770 1760 1750 1740 1730 1720 1710 1700 1690 NIR MIR 6 8 10 12 14 16 18 1 % 1780 1770 1760 1750 1740 1730 1720 1710 1700 1690 NIR MIR 6 8 10 12 14 16 18 NIR averaged to MIR integration time 1 % 1780 1780 1770 1760 1750 1740 NIR MIR 1770 1760 1750 1740 NIR MIR 1730 1730 1720 1710 1700 1690 6 8 10 12 14 16 18 1 % 1720 1710 1700 1690 6 8 10 12 14 16 18 1 % 1780 1780 1770 1760 1750 NIR MIR 1770 1760 1750 NIR MIR 1740 1740 1730 1730 1720 1710 1700 1 % 1720 1710 1700 1 % 1690 6 8 10 12 14 16 18 6 8 10 12 14 16 18 1690
MIR-FTIR XCH4 precision/diurnal variation: impact of regularization? 1.05 1.04 1.03 1.02 1.01 1.00 0.99 0.98 0.97 0.96 XCH4 18 Sep 2003 i of day i (18 Sep) = 0.13 % 10 min integration per column Zugspitze 2003: AV i ( i ) 0.5 % variation mainly due to clouds impact of regularization? 0.95 7:00:00 AM 9:24:00 AM 11:48:00 AM 2:12:00 PM 4:36:00 PM Time of Day
AV i ( i ) (%) L 1 as a function of (dofs): minimize diurnal variation averaged over all days 0,40 0,35 0,30 0,25 0,20 0,15 L 1 is robust even for low dofs with diagonal S a (L 0 ) you can understimate true variability with s down to zero 0,10 Tikhonov L 1 Diagonal S a (=Tikhonov L 0 ) 0,05 0,00 0 1 2 3 4 ( ) degrees of freedom of signal
Sigma (%) Zugspitze result reproduced at ISSJ: Tikhonov L 1 retrievals optimization: intra-day CH 4 column variability vs mean DOFS 0.36 0.34 0.32 0.3 Duchatelet, Oct 2009 0.28 0.26 Jungfraujoch : minimum reached for S/N ~1500 0.24 1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9 3.1 3.3 3.5 DOFS
MIR-FTIR CH4 precision: why does Tikhonov L 1 reduce cloud-induced scatter? Unfortunately, virtually all modern retrieval algorithms use NLLS methods, which do not match equivalent widths. Since NLLS methods minimize the square of the residuals, and since the largest residuals typically arise at line center, NLLS methods try harder to fit the line center than elsewhere on the line profile. taken from Geoff s 2006 talk
MIR-FTS CH4 precision: Tikhonov L 1 better integrates line area than VMR-scaling ^ certainly at the cost of somewhat unphysical profile shapes at the presence of clouds, but never mind if interested in columns different effective apodization parameters set fix in forward model time of day time of day
Global validation of SCIAMACHY XCH4 by NDACC FTS what can we learn for GOSAT validation by TCCON FTS? Summary based on HYMN CH4 profile retrievals our global SCIA validation study with 12 MIR-FTS stations has lead to a maturated XCH4 retrieval from MIR-FTS which can be used to validate both TCCON-FTS and GOSAT XCH4 annual cycles above clean sites with 1 % amplitude differ on the 1 % level between SCIA MACHY & g.-b. MIR-FTS & g.-b. NIR-FTS ( GOSAT?) GFIT-O2 compares very well to dry airmass derived from NCEP or in situ p CH4 smiles are reduced with GFIT ver Feb 09 compared to GFIT ver Dec 06 still both CH4 MIR and newest version GFIT shows > 1 % smiles on some days although precision of g.-b. NIR FTS is better, g.-b. MIR is astonishingly good CH4 profile retrieval with SFIT 2 (Tikhonov L 1 on %-VMR scale) helps to reduce cloud-induced scatter by up to a factor of 3 %
Research Center Karlsruhe, IMK-IFU, Garmisch-Partenkirchen, Germany Ralf Sussmann: 10 Years of Solar FTIR Spectrometry at the Zugspitze 36 MIR-FTIR regularization: Tikhonov first derivative regularization L 1 n n T a 1 1 0 0 1 2 0 0 2 1 0 0 1 1 1 1 1 L L S R with regularization strength.
MIR-FTIR regularization: Suggested Tikhonov first-derivative (L 1 ) regularization applied to profiles given in units of %-VMR changes rel. to a priori profile dofs 1 (VMR profile scaling) dofs 2 dofs 2.5 dofs 3 dofs 4 Research Center Karlsruhe, IMK-IFU, Garmisch-Partenkirchen, Germany Ralf Sussmann: 10 Years of Solar FTIR Spectrometry at the Zugspitze