Seasonal variations of greenhouse gas column-averaged dry air mole fraction retrieved from SWIR spectra of  GOSAT TANSO-FTS  Nawo Eguchi* 1 , Yukio Yoshida 2 , Isamu Morino 2 , Nobuyuki Kikuchi 2 , Tazu Saeki 2 , Makoto Inoue 2 , Osamu Uchino 2 , Shamil Maksyutov 2 , Hiroshi Watanabe 2  and Tatsuya Yokota 2 1: Tohoku University (Now at Kyushu University) 2: National Institute for Environmental Studies * nawo@riam.kyushu-u.ac.jp
Contents Status of SWIR Level 2 current version (Ver01.xx) Seasonal variations of XCO 2  and XCH 4 Comparison with SCIAMACHY(2003-05) Summary Possibility to scientific research use
Greenhouse gases Observing SATellite Top-Down approach Synoptic scale – Global Intra-seasonal, Seasonal, Inter-Annual scales CO 2 , CH 4 , H 2 O, Clouds, Aerosol JAXA/NIES/MOE ( 宇宙航空研究開発機構・環境研・環境省 ) MAP method [e.g., Rodgers, 2000] Column & Profile : CO 2 , CH 4 , H 2 O Interferogram   ( 干渉光 ) フーリエ変換分光器 TANSO-FTS ( T hermal  A nd  N ear-infrared  S ensor for carbon  O bservation -  F ourier  T ransform  S pectrometer  23 January 2009 種子島宇宙センター SWIR Band2 Complex Fourier Inverse  Transform Solar CO 2 CH 4 H 2 O Cirrus Aerosol IFOV ( 観測視野 ) FTS 10.5km Sun Alt. 666km Measurement of reflection  from Surface, clouds and so on
Optimal Estimation Method (Rodgers [2000]) eq.  (1) eq.  (2) The optimal  x  is found when an iterated solution  Cost function J (x)   is a minimum value.  The columns and profiles of CO 2  and CH 4  are retrieved by the optimal estimation method based on Rodgers [2002] from the GOSAT TANSO-FTS SWIR (Shortwave InfraRed;  0.76, 1.6, and 2 micron ) and TIR (Shortwave InfraRed;  0.76, 1.6, and 2 micron ) spectrum data. Optimal solution from eq.(1) eventually required the accurate  Sa  (a priori error covariance matrix) and its assessment. In the GOSAT retrieval, a priori ( X a ) and its covariance matrices ( S a ) of CO 2  and CH 4  are obtained from the simulated data of NIES Transport Model [Maksyutov et al., 2008]. Prior covariance matrix is consisted of variances on the three temporal scales: Synoptic scale variability ( S Synoptic ) in 2-week using NIES TM to obtain concentrations on global (every grids), Interannual variability ( V Interanuual ) using observed concentration to obtain variability for a long term (several decade), (3) Seasonal cycle bias ( B Season ): to estimate the effects of the errors in the simulated seasonal cycles. SWIR L2 ATBD [2010] 先験値情報を設定する意味 もっともらしい値から計算を開始するため。 また、 Sa による 先験値への拘束は ・観測ノイズに起因する解の発散の抑制 ・非線形問題における  local minimum  への収束の回避 Measurement residual Difference from a prior Factor : Error covariance of observation  : Optimal concentration : Observed radiance spectra :  ε S x y a a Error covariance of prior : Prior concentration  : Unit matrix for scaling : S x D Simulated spectra : ) ( x F
Status of SWIR Level2 Ver.01.xx Improved point from previous version (Ver.00.xx) Cirrus detection method Surface Pressure retrievals by using TANSO-FTS SWIR Band 1 (O 2 A band) Explicitly-retrieval of equivalent path length which is closely related with aerosol and surface pressure in retrieval field Spectroscopic parameter of CH 4 ,  line-mixing etc… Period of data available to General User (GU) 6 April 2009 to 19 April 2011 (except May 2009)
Comparison Ver. 01.xx with Ver. 00.xx Yoshida et al. (MSJ meeting 2010) High and low retrieved values are removed because of improved method treating cirrus and surface pressure (aerosol)
Screening strategy of TANSO-FTS SWIR Level 2 data To keep a certain quality of retrieved parameter, the filtering and screening of data are conducted before and after the retrieval process, respectively. Table 2 :  Data number of data passed by L1B quality flag and CAI cloud flag Clear sky ratio (from MODIS) 16 ~ 17 %  Eguchi and Yokota [GRL, 2008] Before the retrieval process, the level 1B data are filtered out by Level 1B quality flag (spike noise, saturation and so on) approximately  60%  NG   (Ver01.10),  approximately  20 % NG   (Ver01.20, 30) CAI cloud flag (remove scan which having cloud pixels) approximately  80 % NG   Totally,  93%  ( 82% ) NG before the retrieval process   Finally  2 ~ 3% * With respect to CAI available data number Ver 01.30  (Ver.01.20 is also same feature) The L1B quality flag check is weak. Most of added data are low SNR data . Period Total L1BFlag CAIcld L1BFlag + CAIcld L2 for GU XCO 2 XCH 4 2009/04/23-25 27543 9522 (34.6) 4347 (19.6*) 1688 (6.3*) 703 681 2009/07/24-26 27973 11484 (41.1) 4253 (19.8*) 1728 (6.6*) 640 590 2010/01/16-18 26185 9518 (36.4) 4259 (20.3*) 1832 (7.2*) 959 951 2010/03/20-22 27096 21455 (79.2) 4550 (20.3*) 4116 (18.4*) 785 739
Table3 : Surviving ratio of retrieved data by screening items ( function of land/ocean 、 clear-sky ratio )  2009 7/24-26  (Shade indicates less than 50%) Effective screening item is  AOD (variety of path length)  for land and  CAI coherent test  for ocean.  χ 2   and   2μ scatering material (cirrus) determinations  are closely correlated with clear-sky ratio within FOV. Spectrum fitting Sufficiency information of spectrum Check cloud remain Evaluation of simultaneous retrieved parameter Convergence of retrieval process Screening (After the retrieval process)
Seasonal characteristic of XCO 2 Apr 2009  ~ Jun 2011 (GU : Apr 2009 – Apr 2011) White color indicates that the data are removed by screening. The sunglint region is primary measurement area over ocean. Ave. XCO 2  (whole period) The retrieved values at high latitudes are low because the GOSAT measured summer time over there. The CO 2  value at summer time is lowest through the year.  The Level 1B quality is low at the tropics and Asian monsoon regions where the clouds cover frequently.
Northern Hemisphere Southern Hemisphere Max. May  /  Min. September Monthly mean STD  3 ppmv Amplitude  5 ~ 10 ppmv Diff from prior 8 ppmv ( ~ 2% low bias) Only the grids with more than six months of data were taken into consideration. Seasonal variation of XCO 2  (Monthly mean) Amplitude (peak-to-peak) Month with the maximum prior (NIES Transport  Model Ver05) 2009 Apr 2010 Apr 2011 Apr SCIAMACHY Interhemispheric Difference (NH-SH)
Regional Characteristic of XCO 2 +0.98 [ppmv/year] +0.88 [ppmv/year] +2.4 [ppmv/year] +0.78 [ppmv/year] +0.67 [ppmv/year]
Seasonal characteristic of XCH 4 Apr 2009  ~ 2011 Jun  (GU : Apr 2009 – Apr 2011) White color indicates that the data are removed by screenings. The sunglint region is primary measurement area over ocean. Ave. XCH 4  (whole period) The seasonal variation in L2 current version is consistent with the previous knowledge.  The contrasts of inter-hemispheric and between east and west North America are seen, also the high XCH 4  is seen over Asia.
Land Ocean Land   Max : Sep-Nov  Min : Apr- Jul Amplitude : 20 ppbv Ocean  ??? Seasonal variation of XCH 4  (Monthly mean) Amplitude (peak-to-peak) Month of the maximum value The dip is caused by the seasonal march of observation latitudinal band. delay 2009 Apr 2010 Apr 2011 Apr Only the grids with more than six months of data were taken into consideration. Non-correction by factor Higher than a prior ( 〜 1%)
Regional characteristic of XCH 4 Non-correction by factor +9.6 [ppbv/year] +6.5 [ppbv/year] -5.8 [ppbv/year]
Summary Quality check of Level 2 current version (Ver.01.xx) Most of level 1B data (93%) are removed by L1B quality check and CAI cloud flag. There is room for improvement of the screening method of cirrus and aerosol (effective path length), esp. thin cirrus rejection and its effect on retrieved value. Seasonal Variations of XCO 2 , XCH 4 It is found that the seasonal variation on the continental scale is similar to the variation by a prior (NIESTM-05) (phase and amplitude), but the XCH 4  seasonal variation (at several regions) is more complex than that of XCO 2 . XCO 2 Large differences from a prior are found in the areas of NH where plant activity is high. XCH 4 Large variances are found over Asia and North America.
Potential to scientific research use Remain  negative bias  of  〜 2% ( 〜 9ppmv) for XCO 2 ,  〜 1% ( 〜 20ppbv) for XCH 4 [Morino et al., AMT, 2011] Improvement of retrieval process Further validation is needed (discussion for seasonal and regional biases). Impacting on flux estimation (Level 4 product) research Seasonal cycle  (phase and amplitude) and  annual mean  (low and middle latitudes) are consistent with the previous knowledge. XCO 2 : Large differences from prior are located over high activity regions of plant. XCH 4 : Large variances are located over East Asia.  Research of  Inter-annual variation  requires data accumulation.  Rejection of abnormal values near sources and sinks Analysis considering  synoptic scales  can be done, if the data quality and number meet the level of quality for science.
Thank you for your attention

IGARSS2011_eguchi.ppt

  • 1.
    Seasonal variations ofgreenhouse gas column-averaged dry air mole fraction retrieved from SWIR spectra of GOSAT TANSO-FTS Nawo Eguchi* 1 , Yukio Yoshida 2 , Isamu Morino 2 , Nobuyuki Kikuchi 2 , Tazu Saeki 2 , Makoto Inoue 2 , Osamu Uchino 2 , Shamil Maksyutov 2 , Hiroshi Watanabe 2 and Tatsuya Yokota 2 1: Tohoku University (Now at Kyushu University) 2: National Institute for Environmental Studies * nawo@riam.kyushu-u.ac.jp
  • 2.
    Contents Status ofSWIR Level 2 current version (Ver01.xx) Seasonal variations of XCO 2 and XCH 4 Comparison with SCIAMACHY(2003-05) Summary Possibility to scientific research use
  • 3.
    Greenhouse gases ObservingSATellite Top-Down approach Synoptic scale – Global Intra-seasonal, Seasonal, Inter-Annual scales CO 2 , CH 4 , H 2 O, Clouds, Aerosol JAXA/NIES/MOE ( 宇宙航空研究開発機構・環境研・環境省 ) MAP method [e.g., Rodgers, 2000] Column & Profile : CO 2 , CH 4 , H 2 O Interferogram   ( 干渉光 ) フーリエ変換分光器 TANSO-FTS ( T hermal A nd N ear-infrared S ensor for carbon O bservation - F ourier T ransform S pectrometer 23 January 2009 種子島宇宙センター SWIR Band2 Complex Fourier Inverse Transform Solar CO 2 CH 4 H 2 O Cirrus Aerosol IFOV ( 観測視野 ) FTS 10.5km Sun Alt. 666km Measurement of reflection from Surface, clouds and so on
  • 4.
    Optimal Estimation Method(Rodgers [2000]) eq. (1) eq. (2) The optimal x is found when an iterated solution Cost function J (x) is a minimum value. The columns and profiles of CO 2 and CH 4 are retrieved by the optimal estimation method based on Rodgers [2002] from the GOSAT TANSO-FTS SWIR (Shortwave InfraRed; 0.76, 1.6, and 2 micron ) and TIR (Shortwave InfraRed; 0.76, 1.6, and 2 micron ) spectrum data. Optimal solution from eq.(1) eventually required the accurate Sa (a priori error covariance matrix) and its assessment. In the GOSAT retrieval, a priori ( X a ) and its covariance matrices ( S a ) of CO 2 and CH 4 are obtained from the simulated data of NIES Transport Model [Maksyutov et al., 2008]. Prior covariance matrix is consisted of variances on the three temporal scales: Synoptic scale variability ( S Synoptic ) in 2-week using NIES TM to obtain concentrations on global (every grids), Interannual variability ( V Interanuual ) using observed concentration to obtain variability for a long term (several decade), (3) Seasonal cycle bias ( B Season ): to estimate the effects of the errors in the simulated seasonal cycles. SWIR L2 ATBD [2010] 先験値情報を設定する意味 もっともらしい値から計算を開始するため。 また、 Sa による 先験値への拘束は ・観測ノイズに起因する解の発散の抑制 ・非線形問題における local minimum への収束の回避 Measurement residual Difference from a prior Factor : Error covariance of observation : Optimal concentration : Observed radiance spectra :  ε S x y a a Error covariance of prior : Prior concentration : Unit matrix for scaling : S x D Simulated spectra : ) ( x F
  • 5.
    Status of SWIRLevel2 Ver.01.xx Improved point from previous version (Ver.00.xx) Cirrus detection method Surface Pressure retrievals by using TANSO-FTS SWIR Band 1 (O 2 A band) Explicitly-retrieval of equivalent path length which is closely related with aerosol and surface pressure in retrieval field Spectroscopic parameter of CH 4 , line-mixing etc… Period of data available to General User (GU) 6 April 2009 to 19 April 2011 (except May 2009)
  • 6.
    Comparison Ver. 01.xxwith Ver. 00.xx Yoshida et al. (MSJ meeting 2010) High and low retrieved values are removed because of improved method treating cirrus and surface pressure (aerosol)
  • 7.
    Screening strategy ofTANSO-FTS SWIR Level 2 data To keep a certain quality of retrieved parameter, the filtering and screening of data are conducted before and after the retrieval process, respectively. Table 2 : Data number of data passed by L1B quality flag and CAI cloud flag Clear sky ratio (from MODIS) 16 ~ 17 % Eguchi and Yokota [GRL, 2008] Before the retrieval process, the level 1B data are filtered out by Level 1B quality flag (spike noise, saturation and so on) approximately 60% NG (Ver01.10), approximately 20 % NG (Ver01.20, 30) CAI cloud flag (remove scan which having cloud pixels) approximately 80 % NG Totally, 93% ( 82% ) NG before the retrieval process Finally 2 ~ 3% * With respect to CAI available data number Ver 01.30 (Ver.01.20 is also same feature) The L1B quality flag check is weak. Most of added data are low SNR data . Period Total L1BFlag CAIcld L1BFlag + CAIcld L2 for GU XCO 2 XCH 4 2009/04/23-25 27543 9522 (34.6) 4347 (19.6*) 1688 (6.3*) 703 681 2009/07/24-26 27973 11484 (41.1) 4253 (19.8*) 1728 (6.6*) 640 590 2010/01/16-18 26185 9518 (36.4) 4259 (20.3*) 1832 (7.2*) 959 951 2010/03/20-22 27096 21455 (79.2) 4550 (20.3*) 4116 (18.4*) 785 739
  • 8.
    Table3 : Survivingratio of retrieved data by screening items ( function of land/ocean 、 clear-sky ratio ) 2009 7/24-26 (Shade indicates less than 50%) Effective screening item is AOD (variety of path length) for land and CAI coherent test for ocean. χ 2 and   2μ scatering material (cirrus) determinations are closely correlated with clear-sky ratio within FOV. Spectrum fitting Sufficiency information of spectrum Check cloud remain Evaluation of simultaneous retrieved parameter Convergence of retrieval process Screening (After the retrieval process)
  • 9.
    Seasonal characteristic ofXCO 2 Apr 2009 ~ Jun 2011 (GU : Apr 2009 – Apr 2011) White color indicates that the data are removed by screening. The sunglint region is primary measurement area over ocean. Ave. XCO 2 (whole period) The retrieved values at high latitudes are low because the GOSAT measured summer time over there. The CO 2 value at summer time is lowest through the year. The Level 1B quality is low at the tropics and Asian monsoon regions where the clouds cover frequently.
  • 10.
    Northern Hemisphere SouthernHemisphere Max. May / Min. September Monthly mean STD 3 ppmv Amplitude 5 ~ 10 ppmv Diff from prior 8 ppmv ( ~ 2% low bias) Only the grids with more than six months of data were taken into consideration. Seasonal variation of XCO 2 (Monthly mean) Amplitude (peak-to-peak) Month with the maximum prior (NIES Transport Model Ver05) 2009 Apr 2010 Apr 2011 Apr SCIAMACHY Interhemispheric Difference (NH-SH)
  • 11.
    Regional Characteristic ofXCO 2 +0.98 [ppmv/year] +0.88 [ppmv/year] +2.4 [ppmv/year] +0.78 [ppmv/year] +0.67 [ppmv/year]
  • 12.
    Seasonal characteristic ofXCH 4 Apr 2009 ~ 2011 Jun (GU : Apr 2009 – Apr 2011) White color indicates that the data are removed by screenings. The sunglint region is primary measurement area over ocean. Ave. XCH 4 (whole period) The seasonal variation in L2 current version is consistent with the previous knowledge. The contrasts of inter-hemispheric and between east and west North America are seen, also the high XCH 4 is seen over Asia.
  • 13.
    Land Ocean Land Max : Sep-Nov Min : Apr- Jul Amplitude : 20 ppbv Ocean ??? Seasonal variation of XCH 4 (Monthly mean) Amplitude (peak-to-peak) Month of the maximum value The dip is caused by the seasonal march of observation latitudinal band. delay 2009 Apr 2010 Apr 2011 Apr Only the grids with more than six months of data were taken into consideration. Non-correction by factor Higher than a prior ( 〜 1%)
  • 14.
    Regional characteristic ofXCH 4 Non-correction by factor +9.6 [ppbv/year] +6.5 [ppbv/year] -5.8 [ppbv/year]
  • 15.
    Summary Quality checkof Level 2 current version (Ver.01.xx) Most of level 1B data (93%) are removed by L1B quality check and CAI cloud flag. There is room for improvement of the screening method of cirrus and aerosol (effective path length), esp. thin cirrus rejection and its effect on retrieved value. Seasonal Variations of XCO 2 , XCH 4 It is found that the seasonal variation on the continental scale is similar to the variation by a prior (NIESTM-05) (phase and amplitude), but the XCH 4 seasonal variation (at several regions) is more complex than that of XCO 2 . XCO 2 Large differences from a prior are found in the areas of NH where plant activity is high. XCH 4 Large variances are found over Asia and North America.
  • 16.
    Potential to scientificresearch use Remain negative bias of 〜 2% ( 〜 9ppmv) for XCO 2 , 〜 1% ( 〜 20ppbv) for XCH 4 [Morino et al., AMT, 2011] Improvement of retrieval process Further validation is needed (discussion for seasonal and regional biases). Impacting on flux estimation (Level 4 product) research Seasonal cycle (phase and amplitude) and annual mean (low and middle latitudes) are consistent with the previous knowledge. XCO 2 : Large differences from prior are located over high activity regions of plant. XCH 4 : Large variances are located over East Asia. Research of Inter-annual variation requires data accumulation. Rejection of abnormal values near sources and sinks Analysis considering synoptic scales can be done, if the data quality and number meet the level of quality for science.
  • 17.
    Thank you foryour attention

Editor's Notes

  • #2 IGARSS oral presentation 20min 17-18min 2-3min for QA Font size 24 or more Plan on covering at most 6 points per slide, covered by 6 to 12 spoken sentences and no more than about two spoken minutes. Paper 2864: SEASONAL VARIATIONS OF GREENHOUSE GAS COLUMN-AVERAGED DRY AIR MOLE FRACTIONS RETRIEVED FROM SWIR SPECTRA OF GOSAT TANSO-FTS Session: TH1.T07.3 - Satellite-Based Atmospheric Sounding and Imaging Nawo Eguchi* , Yukio Yoshida, Isamu Morino, Nobuyuki Kikuchi, Tazu Saeki, Makoto Inoue, Osamu Uchino, Shamil Maksyutov, Hiroshi Watanabe and Tatsuya Yokota   Center for Global Environmental Research (CGER), National Institute for Environmental Studies (NIES), Onogawa 16-2, Tsukuba, 305-8506, Japan *now at Center for Atmospheric and Oceanic Studies (CAOS), Graduate School of Science, Tohoku University, Aramaki-aza Aoba 6-3, Aobaku, Sendai, 980-8578, Japan nawo@m.tohoku.ac.jp, +81-22-795-6743   Presentation Type: Oral Session Location: Room 13 Session Time: Thursday, July 28, 08:20 - 10:00 Paper Presentation Time: Thursday, July 28, 09:00 - 09:20
  • #5 Optimal estimation method is adapted by the GOSAT retrieval. The evaluation function, that is cost function consists two terms, First one is measurement residual, that includes the measurement information. Second one is a prior covariance matrix plays quite important role to avoid the unphysical solutions and to obtain solution stably. The optimal x is found when an iterated solution is a minimum value. The 2nd term avoids unphysical solutions resulted from the 1st term.
  • #8 コメント: L1B 品質フラグは『安全性を重視した厳しい設定』から『最適な設定』へ変更した、というのが表向きの表現であるが、逆に見落とされてフラグが付加されなくなった低品質データも存在する。 passed data よりは added data の方が良いかと。 前者だと一般的に L1B 品質をパスしたデータは SNR が低い、と思われる可能性有り。 なお、スパイクノイズフラグの変更は V01.20 時点で実施されており、この段階ですでに数が増えています。 V01.30 での変更は飽和フラグで、むしろ SNR が高いデータが ( 数としてはあまり多くはありませんが ) 増えています。