Docstoc

AWAJI_CLIVAR_GODAE2006

Document Sample
AWAJI_CLIVAR_GODAE2006 Powered By Docstoc
					    Four-Dimensional Variational
  Coupled Data Assimilation by K-7
                                   Toshiyuki Awaji
            Japan Agency for Marine-Earth Science and Technology (JAMSTEC)
              Department of Geophysics, School of Science, Kyoto University

Members:
N. Sugiura, S. Masuda, T. Mochizuki, T. Miyama, H. Igarashi, N. Ishida, T. Toyoda (JAMSTEC)
K. Horiuchi, H. Hiyoshi (AESTO), N. Komori (JAMSTEC/Earth Simulator Center)
Yoichi Ishikawa (Kyoto Univ.),Masa Kamachi(JMA-MRI)

                                                         “K7” is the symbolic name of
Outline:                                                 Category 7 of MEXT’s
                                                         Research Revolution 2002
    1) Introduction
                                                         (RR2002) Project Using the
    2) CDA System & Strategy
                                                         Earth Simulator (called
    3)Results : CDA for 97/98 El Nino                    “KYOUSEI 7” in Japanese)
                (and long-term ODA)
    4)Summary
                      INTRODUCTION
Main Objectives;
 Construction of 4D-VAR CDA (Coupled Data Assimilation)
System capable of offering greater information content
and forecast potential for S-I processes by creating a
higher-quality dataset than do models or data alone
 Application to State Estimation of S-I
                                       Variabilities in
1990s Focusing on 1997/98 biggest EL Nino
Methodology:
4DVAR approach is one of the most attractive assimilation methods because
it can provide a dynamically self-consistent output.
 CDA has the ability to provide a best guess time trajectory of S-I phenomena
(inherently coupled process) by optimizing both initial conditions and important
parameter values compared with AGCM or OGCM alone.
               Experimental Settings
   Coupled Model on ES (CFES):
      T42L24 AFES for AGCM (originally CCSR/NIES AGCM and then improved:
       e.g.: new radiation code MstrnX [Nakajima et al. 2000] and diagnostic code of
       marine stratocumulus [Mochizuki et al. 2006] )
      1x1deg L36 MOM3 for OGCM
      Canopy type Model for Land (MATSIRO)
      IARC Sea-Ice Model
   Assimilation Method: 4DVAR
      Adjoint OGCM and adjoint AGCM are coupled
   Assimilation Window
      Climatological seasonal exp.: 1 year window (using 1-month OBS data)
      1996,97,98 (3-year-long specific) reanalysis exp.: a sequence of 9-month
       window (using series of 10-day means of OBS data) and 11ensemble
       experiments were conducted to cope with weather modes
   Major Assimilated Elements
      Atmosphere: NCEP’s BUFR data and SSM/I sea wind
      Ocean: WOA data, T/P altimeter data, FNMOC dataset, OISST values, and
       ARGO float data from the Coriolis Data Center
   Diagnostic Run: First guess field
      20-yr free integration of CGCM (or OGCM)
               What do we control?
1.   Ocean initial condition
2.   Bulk parameters controlling Air-sea fluxes of
        Momentum           Fv    M CM v v
        Sensible heat      F  c p H C H v  g   
        Latent heat        Fq   E C E v q g  q 

 hFor Smaller-scale Parameterization:
 we made pre-optimization using the Green function approach.
                                                                      Cost Function
                                                                                                            X                                      
          50
                                                                                                                                  T
          45
                                                                                      J  αT B 1α                                1
                                                                                                                     atm  X atm R atm X atm  X atm
                                                                                                                             obs                 obs

                                                                                                          10 daily


                                                                                                 X                                      
          40                                                                                                            T
                                                                                                                       1
                                                                           alpha_m
                                                                           alpha_h
                                                                                                         ocn  Xobs R ocn X ocn  Xobs ,
                                                                                                                 ocn                ocn
          35                                                                                   10 daily
                                                                           alpha_e

                                                                                                        xi  x j 2 yi  y j 2 
                                                                           u10
          30
                                                                           u
                                                                           v          B ij   i j exp                         
   cost




          25
                                                                           t                            2i  j
                                                                                                              x x          y y 
                                                                                                                        2i  j 
          20
                                                                           q                           
                                                                           SST
                                                                           T_ocn
          15                                                               S_ocn
                                                                           SSHa
                                                                                     Overall,
          10
                                                                                           Each cost term shows the
           5                                                                                decrease.
           0                                                                               Atmospheric costs terms
               1
                   3
                       5
                           7
                               9
                                   11
                                        13
                                             15
                                                  17
                                                       19
                                                            21
                                                                 23
                                                                      25




                                    iterations                                              decrease (not monotonically
Cost reduction process for specific experiment                                              with some fluctuation).
               Window: Jul.1996-Mar.1997
 Time scales of Atmosphere and Ocean differ too much!
 Our    assimilation Strategy for S-I processes:
    Forward model: forecasts both weather + seasonal modes
    Backward adjoint: assimilates monthly mean OBS data for
     climatological exp. and 10-day mean OBS data for specific exp.
     to improve background State by 4DVAR approach (energy
     source of weather modes)
    Leave weather mode to more ability of CGCM with improved
     background states                          Aiming at optimized
                                                             tuning of initial
                                                             conditions and
      forward                                               parameters to
     backward                                     Obs.
                                                             practically obtain
                                                             approximate solution
                                                             by 4dvar iterative
                                                        Obs.
                                Best guess trajectory        procedure of forward
       First guess field
                           Obs.    not snapshot              and backward
                                     Assimilation            calculations
                                        period
  First description
                        Vertical VS latitudinal cross section
Climatological                                                     Global
 Seasonality            Left side:K7          Right-side:ERA40 atmospheric
 Examples:                           comparison                structure: well
                                                              defined in CDA




              SP        EQ            NP          SP            EQ      NP
                             Zonal mean Temp. (January)
                   SP     EQ               NP




              SP        EQ            NP         SP             EQ     NP
                                   Zonal mean wind
Estimated Latent Heat flux      Confirm goodness

                                     Good quality
                                     of mapping




   Observation (COADS)       Our Assimilation




         NCEP2                   ERA40
        For regional processes
Annual march of monsoon precipitation
averaged over the Indo-china Peninsula



    CMAP OBS


    Assimilation
                               simulation




               monsoon onset : better define
                    Example of 4DVAR advantage
                  in case of heavy rainfall over Russia

                                                        Sensitivity exp. by adjoint
                                                       calculation identifies PDF of
                                                       moisture sources of rainfall
time series of daily rainfall in outlined area
                                                                            PDF of source




                  Similar result to that
                  by Gong & Eltahire
                                                      Adjoint variable of surface
                   (1996) approach               evapotranspiration in the case of July9
   rainfall & water vapor flux field
Comparison between Adjusted bulk parameter
         同化(ens4_test4s2)による中立時抵抗係数と
     values and experimental results
                                                 Charnockの比較
                   2.3             Comparison with Charnock’s experiment
                   2.1
                                                                                       deviation
                   1.9            Adjusted
                                 values lie in
   CD*0.001(10m)




                   1.7                                                                           ±1σ
                                 reasonable
                   1.5              range
                                                                                             平均
                   1.3                                                                      adjusted
                                                                                             mean
                                                                                      Charnock
                   1.1
                                                                              Charnock’s
                   0.9                                                      experimental law
                   0.7
                     0
                         1
                             2
                                 3
                                     4
                                         5
                                             6
                                                 7
                                                     8

                                                         9
                                                             10
                                                                  11
                                                                       12

                                                                            13
                                                                                 14
                                                                                       15
                                                                                            16

                                                                                                 17
                                                                                                      18
                                                                                                           19
                                                     海上10mの中立風速(m/s)

                                                         U 10 m/s
Move on
     1996/97/98 reanalysis results by CDA
            Best Time Trajectory to observation
                                                      Obs.

                                                 9 month
                                                             Obs.
          First guess
                               Obs.
          field                        Assimilation
                                         period
 For example; NINO3.4 SST reanalysis from 1997January to 1997 September
                    First Guess (by IAU)                             Reanalysis by single CDA




                                                                    Black: Observation
            Black: Observation                                      Red:CDA single run gives
            Red:coupled model                                       better trajectory capable
            simulation run                                          of capturing obs features
                                                 Using (1) adjusted ocean initial condition
                                                           and
                                                        (2) Adjusted bulk coefficient to S-I
    A sequence of assimilation windows for the year 1996-1998
   reanalysis with 10-day mean OBS
96/1       96/9     97/1        97/9     98/1       98/9

         96/7        97/3   97/7        98/3
                                                98/7           99/3


                                           * 97/1 = January of 1997


Coupled Data Assimilation for the year 1996-1998 reanalysis
  with 9-month assimilation window, of which 2-month
        overlapping at both ends of each window
Taking into account the influence of uncontrolled short-term fluctuations, we made
  11 ensemble runs with different atmospheric initial conditions
  (shifting atmospheric data from 27 Dec 1996 to July 6 1997 by 1 day interval)

            Ensemble simulation using AIU                 Ensemble of Assimilation

 Weather component
 included statistically




                                                                  Black: Observation
                                                                  green:each ensemble
                                                                  (Blue:Jan 1st or Jul 1st)
                                                                  Red:ensemble mean




     Ensemble 4DVAR including weather mode-like fluctuations shows the robustness of
 the previous assimilation
                                                                     is made
   and suggest that ENSO could be predictable if good initializationCome back
 somehow.           Implication to enhanced predictability again
       Indian Dipole Mode Index
Example: From July 1997 to March 1998

        Ensemble simulation                     Ensemble of Assimilation
                                                      Positive dipole event




                                                          Black: Observation
                                                          green:each ensemble
                                                          (Blue: Jul 1st)
                                                          Red:ensemble mean



                                    Best trajectory to observation with
                                    (1) adjusted ocean initial condition
                                              and
                                    (2) adjusted bulk coefficient
ERA40        Features of atmospheric
             anomalies relating to the 97/98
             ENSO event : well captured

                    In case of PNA Pattern
                         (FM1998-FM1997)
  Basic agreement


A good ensemble                        Ensemble mean
member in pattern                      - good in pattern
and intensity                          - weak in amplitude
for PNA pattern
Time series of NINO3 SST variation in CDA




                                  Red: Simulation
                                  Blue: Reynolds SST
                                  Green: Assimilation




   Nino3 SST time series of CDA product is close to
    that of Reynolds SST.
SST variation in the equatorial region in CDA
      Simulation         Assimilation       Reynolds SST




97-98 ENSO event is well reproduced in the assimilation field.
Correction of Westerly Wind Burst by CDA
        Simulation      Assimilation         NCEP




WWB events become to occur realistically by CDA procedure.
Ensemble 4DVAR runs show the
Implication to enhanced predictability
                                                  Thus the subsequent
                                                  15-month “pure”
                                                  forecast with the
                                                  adjusted ocean initial
                                                  condition : conducted


            9-month assimilation                  15-month “pure” prediction




                     Half year-period 4DVAR initialization
                      is underway towards 1-year-longer
                          lead prediction of EL NINO



          Predictability continues beyond the assimilation period
Comparable or more




        predictability
Long-term Ocean Data Assimilation
       Product in K7 group
        just one example
     Subarctic North Pacific Water         Sensitivity experiment
                                     o                      2002/07
characterized by two water masses (47 N)

                                                           2002/06


                       Dichothermal
                       structure                           2002/05
                   A
                                           Outcropping     2002/04
                   B
                       Mesothermal
                       structure
                                                         2002/07                   2001
                                                                                   /01



  An “artificial cost” input at
                                                         2002                     2000/07
  A (Dichothermal water),                                /01

  B (Mesothermal water)
  for 3-yr backward calculation.
                                                          2001/07
=> Origin of water mass
                                                                    Subtropical:Subarctic
=> Observe                                                               = 6.9 : 1
             Concluding remarks
   Dynamically consistent data set capable of offering greater
    information content and forecast potential for S-I (e.g.
    ENSO variability, subsurface water masses) is obtained. by
    our CDA experiment
   Coupled data assimilation by simply controlling bulk
    coefficients brings some improvements to coupled model
    climatology.
   We are constructing 1990’s coupled model assimilated data
    and a forecast system ( will be open in this year).
   Assimilation datasets and observational
    datasets are available at
     http://www.jamstec.go.jp/frcgc/k7-base2/
      Time series of Indonesia through flow
            transport & NINO3 SST




  Observed coherency is clearly detected between these time series.
Gordon &Fine (1996, Nature):      Mean value for 1987-2004: 11.9Sv
 7~18.6 Sv                        6-month running mean (figure)

				
DOCUMENT INFO
Shared By:
Categories:
Stats:
views:6
posted:2/28/2010
language:English
pages:25