SENSITIVITY OF OBSERVATIONAL DATA TO TIME DEPENDANT INVERSION Takashi Maki Senior Coordinator for Chemical Transport Modeling Atmospheric Environment Division Japan Meteorological Agency CONTENTS 1. Purposes 2. Experimental Methods 3. Results 4. Summary and Future Plans
1-1. Background We usually adopt CO 2 observation dataset provided by CMDL ( GLOBALVIEW ) in inversion In GLOBALVIEW, CO 2 data are smoothed, interpolated and extrapolated We tried to use CO 2 raw data to estimate CO 2 flux more realistically in inversion There is a possibility that we tend to underestimate the CO 2 flux variability
1-2. Outline of our experiments TransCom 3 Level 3 control run Inversion 1 Inversion 2 Inversion 3 Inversion 2.5 As similar as possible Preparation for raw data Use raw data (WDCGG) Use as many raw data as possible
2-1. Outline of Inversion 1 Model outline JMACDTM (2.5deg. 32levels) Offline, 6 hourly JMA-winds (1997) Diffusion (Cumulus, Turbulent and shallow convection) Inversion method Bayesian (Greens functions) approach Internal shape, landCASA Internal shape, oceanFlat Prior flux and errorsmall error Prior data and errorGLOBALVIEW Offset of CO ppm
Difference between Inversion 1 and 2 Aim: Reduce Constraints when there are no available observational data Inversion 2 provide a base data in estimating inversion Outline of Inversion 2 Inversion 1Inversion 2 Data Uncertainty As in T3L2Multiple by 10.0 when there are no raw data
Difference between Inversion 2 and 2.5 Aim: Replace GLOBALVIEW data with WDCGG raw data if available Outline of Inversion 2.5 Inversion 2Inversion 2.5 Prior dataAs in T3L2Use WDCGG monthly data if available Data error As in Inversion 2 Raw data GV x 1 Others GV x 10
Difference between Inversion 2.5 and 3 Aim: Use as many sites as possible We did not use site where the annual mean concentration is too high as zonal mean concentration (as in WDCGG No.27) We select 106 sites 2-4. Outline of Inversion 3 Inversion 2.5Inversion 3 Sites selection As in T3L2 Select site where there are enough raw data (60%)
2-5. GLOBALVIEW and WDCGG GLOBALVIEWWDCGG OrganizationCMDL/NOAA JMA (WMO/GAW) Data intervalWeekly etc.Monthly etc. Data management Smoothed, interpolated and extrapolated Reported data (if not available, calculated by WDCGG) MediaCD-ROM, Internet
2-6. Data management by WDCGG Format check Threshold value check (300ppm-500ppm) Unnatural value check (same value etc.) WDCGG contact each laboratory and correct data if possible. The data are updated every month ( Suspicious data In principle, data are edited and selected by the data submitter.
2-7. Outline of 106 Sites (Inv. 3) Data from WDCGG 98 there is no raw data, use data from GLOBALVIEW Data from GLOBALVIEW 8 Sites (car030, car040, car050, car060, cri02, lef030, ljo04, opw00) Rejected sites in WDCGG Lack of data amount 9 sites Scale is unknown 4 sites Conc. are too high (low)15 sites
3-1. Data amount and estimated error Southern Hemisphere Northern Hemisphere Tropical Region Total estimated error
3-2. Annual estimated flux Estimated flux in global scale Inversion 3 shows larger fluctuation
3-3. Annual estimated flux Northern Hemisphere Tropical Region Southern Hemisphere N.H. Control and Inv.3 is similar Inv.2 and Inv. 2.5 is same T.R. Control and Inv.1 is same Inv.2 and Inv. 2.5 is same S.H. Inv.1 – Inv.3 are similar
3-4. Inversion 1 and control run Correlation coefficient for every month In Tropical region, the coefficients tend to small. Land Land Ocean Ocean Land Land Ocean Ocean Land Land Ocean Ocean Land Land Ocean Ocean Land Land Ocean Ocean Land Ocean
3-5. Inversion 1 and Inversion 2 Standard deviation of monthly flux gap RegionFlux gap Temp. Asia0.080 Europe0.071 Temp. N. America0.061 Bor. Eurasia0.058 N. Africa0.057 Trop. Asia0.047 Data amount reduced regions RegionRate Australia31.3% Temp. N. America26.2% Europe24.3% W. Trop. Pacific23.4% Temp. Asia22.5% N. Ocean22.4% Flux gap appears in the region where there are less raw There are no data in B. Eurasia, N. Africa and Tr. Asia
3-6. Inversion 1 and Inversion 2 Increase in mean monthly error gap Increase in Error is small (in land region) E. Trop. Pacific North Pacific Southern Ocean South Pacific Bor. Eurasia Europe Temp. N. America Trop. Asia Temp. Asia Err gapRegion
3-7. Inversion 2 and Inversion 2.5 Standard deviation of monthly error gap Australia Europe B. N. America B. Eurasia S. Indian Ocean Temp. N. America Difference appeared (small) in data amount reduced region! Estimated errors tend to increase (small) Data amount reduced regions S. Indian Ocean15.5% B. N. America8.0% Europe5.2% Australia3.1% E. Trop. Pacific1.2% S. Pacific0.8%
3-8. Result of Inversion 3 Estimated flux in From inversion 3, CO2 flux increased in Tropical land regions at 1997 – 1998.
3-9. Result of inversion 1-3 Average of estimated flux and data error In inversion 2 and 2.5, errors increased. In inversion 3 flux error reduced. ExperimentFlux (GtC/y)Data (unit less) Inversion Inversion Inversion Inversion
4-1. Result from inversion 1 In global scale, inversion 1 is similar to control run. Inversion 1 underestimate N.H flux and overestimate S.H flux. In tropical region, total flux is similar to control run. But in each region, there is a difference between control run and inversion 1 There is a room to modify greens function
4-2. Result from inversion 2 The effect of increasing prior data errors are not so large. Flux gap and error increase appears in data amount reduced region. Estimated data error tend to increase from inversion 1. Estimated flux error have a good correlation with data coverage.
4-3. Result from inversion 2.5 The result shows that we can combine GLOBALVIEW and WDCGG dataset without reduce accuracy. The change occurs data amount reduced regions. Estimated flux and data error tend to increase but not so large from inversion 2. This shows a option to use raw data.
4-4. Result from inversion 3 Data extension made possible to reduce total estimated flux error. Estimated data error are larger than inversion 1, 2 and 2.5. Data coverage and quality of data could affect this result.
4-5. Conclusions Sensitivity of dataset to time dependant inversion is not so large when we select measurement appropriately. There is one option to use raw data to time dependant inversion. We need more data which are well selected to run time dependant inversion. Please submit data to WDCGG!
4-6. Future Plan Enhance data quality control Statistical analysis for observations is needed. Use internally varying NCEP reanalysis or JRA25 (planned) Use On-line Based upon Kosa prediction model Estimate carbon flux more precisely