Download presentation
Presentation is loading. Please wait.
Published byOswald Bridges Modified over 9 years ago
1
増田周平, 淡路敏之, 杉浦望実, 石川洋一, 五十嵐弘道, 日吉善久, 佐々木祐二, 土居知将 JAMSTEC Kyoto University 四次元変分法データ同化手法を用いた 全球海洋環境の再現
2
はじめに 海洋環境再現実験 近年、海洋・気候学の分野でデータ同化手法を用いた海洋観測データと数値モデル計算結 果の統合が活発に研究されるようになってきた。その背景には定常観測の難しい海洋亜表 層の気候変動現象に対する重要性の認識や計算機科学の発達などがある。(独)海洋研究 開発機構では京都大学と共同して四次元変分法アジョイント手法を応用した海洋データ同 化システムを構築し、過去 50 年間にわたる海洋環境の再現を試みた。 気候変動メカニズムの解明 得られた海洋環境統合データセットを用いて気候変動現象、とくに季節 ― 経年スケール の現象に着目し、力学解析を行うことでそのメカニズムの解明を目指す。 最適観測システム構築への応用 データ同化システムを用いた応用研究として海洋環境再現のためにどの海域を重点的に観 測すれば効率が良いかを示唆する観測システム研究を行った。主に赤道太平洋域に焦点を 当て、気候変動イベントに対する感度の高い海域を同定し、そこでの観測のインパクトを 調べる。
3
Data Synthesis Efforts in Oceanography
4
Data Assimilaion An optimal synthesis of observational data and model results. Data assimilation can provide analysis fields in superb quality through 4-dimensional dynamical interpolation of in-situ observations. Observations Numerical model Merit Truth ( real ocean ) Equal quality in 4-d continuum Demerit Spatially and temporally sporadic sometimes Unrealistic; Model bias, parameterization
5
No Model z-Level Model EN3 DePreSys MIT MOM POPOPA/NEMO ERA40 NCEP Mercator URDG SODA GFDL GODAS K-7 HOPE ECMWF INGV 3D-Var/OI 4D-Var.25 o x.25 o 1 o x1 o 2 o x2 o CORE GECCO Bias corr. E-P. Relax. QSCAT GPCP DATA World’s Ocean Data Synthesis efforts (CLIVAR/GSOP)
6
4D-VAR approach A 4-dimensional variational (4D-VAR) adjoint data assimilation system can provide a dynamically self- consistent dataset. The obtained products are applicable to dynamical analysis, adjoint sensitivity experiment, Observation System Experiment, ecosystem modeling, forecast study. High computational cost is required.
7
Adjoint sensitivity analysis by using a 4D-VAR ocean DA system surface Heat flux The adjoint sensitivity analysis gives the temporal rate of change of a physical variable in a fixed time and space when model variables (e.g., water temperature, salinity, velocity, or surface air-sea fluxes) are arbitrarily changed in the 4-dimensional continuum of one temporal and three spatial coordinates. This is equivalent to specifying the “sensitivity” of a variable to small perturbations in the parameters governing the oceanic state. An adjoint sensitivity analysis moves the ocean representation backward in time! Wind stress ΔT = T model - T obs
8
Obs. Assimilation window Obs. First guess field Search of best time trajectory 4D-VAR data assimilation approach seeks for optimized 4-dimensional model states by minimizing a cost function (differences between observed and model analysis fields). In that process, Forward & Backward model are executed iteratively within assimilation window. Assimilation window should be well chosen taking the “ memory ” of oceanic phenomena of interest into consideration. In general, the longer the memory of the phenomenon is, the longer the assimilation window must be. forward backward Best time trajectory
9
K7 Ocean Data Assimilation System
10
Recent high quality observational surveys conducted during the WOCE and the WOCE revisit have revealed the sobering fact that the deepest waters of the major oceans have warmed significantly during recent decades “Bottom-water warming” Blue Earth (2004) Such temporal changes are important to understand the variability of abyssal circulation which have implications for large-scale thermohaline transport and thus for the global 3-dimensional heat budget that is presently of vital concern. MIRAI RV participate in WOCE revisit in the subarctic North Pacific in 1999 Fukasawa et al. (2004), Kawano et al. ( 2006) This bottom-water warming ranges in magnitude from 0.003 to 0.01 o C in the Pacific Ocean over the period 1985-1999.
11
OGCM : GFDL MOM3, quasi-global 75 o S-80 o N horizontal res : 1 o x1 o, vertical res:45 levels Spinup : 1. 3000-year with a climatological forcing (accelerated method) 2. 120-year as climatological seasonal march. 3. 10-year with interannual forcings from NCEP/DOE. Use of optimal parameters : Green’s function method is applied to some physical parameters (Toyoda et al.,20XX). Data sysnthesis : method : strong constrain 4D-VAR adjoint. adjoint coding: by TAMC with some modifications. assimilation window : 50 years (1957-2006) control variables : initial conditions, 10-daily surface fluxes first guess : results from Spinup 3 assimilated elements : OISST,T,S ( Ensembles ver.3 + Mirai RV independent dataset ),AVISO SSH anomaly. System
12
Green’s function : Optimization of physical parameters Modeling: 1. Short-wave radiation scheme is modified (collaboration with Ocean Circulation Team) 2. BBL scheme, GM scheme 3. Anomalous mending for polar region dynamics, Bland-new QCed Data : Download EN3_v2a_NoCWT_WijffelsTable1XBTCorr with independent RV MIRAI data Deep ocean data synthesis : State-of-the-art adjoint coding. Low resolution compiling for deep ocean obseravations. Assimilation with long-term window Control of model trend (collaboration with Univs.). Our efforts to reproduce “bottom-water warming” in reanalysis dataset (legacy of 4 th 研究開発促進アウォード )
13
Tsujino and Suginohara (2000) は全球で水平一様な鉛直拡散係数の鉛直プロファイ ルを経験的に推定した。 Hasumi and Suginohara (1999) は地形性内部波による混合に注目し、海底地形の粗 度によるパラメタリゼーションを行った。 Gargett (1986) は成層状態に依存したパラメタリゼーションを行い (κ~N -α ) 、 OGCM でも使用されている (Cummins et al., 1990) 。 3 つの視点の異なる鉛直拡散パラメタリゼーションの線形結合を考え、観測デー タをもとに客観的に結合係数を決定する。 Gargett (1986) の拡散については、 K GGT =10 -3 N -1 (in cgs) とし、大西洋の恒久躍層に 対する議論 (e.g., Marzeion et al., 2007) から 2000m 以深で有効とした。 更に、鉛直拡散に寄与する二重拡散 (Schmitt, 1988; Marmorino and Caldwell, 1976) 、等密度面・層厚拡散 (Redi, 1982; Gent and McWilliams, 1990) のパラメタリ ゼーション、海底境界層とモデル最深層との拡散係数も推定に加えた。 これらチューニングパラメータ群の初期推定値は η 0 =(f TJN,f GGT,f HSM,H HSM,f sf,f dc,A isp,A thk,K BBL ) =(1/3,1/3,1/3, 700, 1, 1,10 3,10 3,2*10 -4 ) 推定するパラメータ
14
コントロールラン (CTL) として、静止状態から月平均気候値フォーシングを与えて、 加速法 (Bryan, 1969) を用いて 4000 年積分後、加速しないで 320 年積分した。 ( 4000m 以深平均の水温トレンドは最後の 100 年で 0.0002 ℃。) 320 年の(加速なし)部分を 9 つのパラメータそれぞれについての擾乱実験を行い、 以下のグリーン関数法 (Menemenlis et al., 2005) により、最適なパラメータ群 η を推 定した。 線形を仮定( )すると、 コスト関数: を最小にするインクリメントは である。 ここで、 x は状態ベクトル(最後の 20 年の年平均気候値)、 y は観測値である。また、 R 、 B はそれぞれ観測値、制御変数に対する誤差共分散行列(対角)で、 R -1 の成分 (観測の重み)は各モデルグリッドにおける、(グリッドの占める体積) / (観測 値の分散)から決定した。 観測値は、既存の水温・塩分データ (EN3 by Met Office) に、海洋地球観測船みらい による CTD/XCTD データを加えて作成した。このとき、南シナ海・スル海、日本海、 メキシコ湾はコストが集中するが、今回のパラメータ調整では修正しきれないモデ ルバイアスがあるとして観測から除いた。 得られた最適値は以下(コスト関数は約 5% 減) η=(f TJN, f GGT, f HSM, H HSM, f sf, f dc, A isp, A thk, K BBL ) =(0.43,0.08,0.72, 580,0.91,0.91,1.0*10 3,1.3*10 3,2.1*10 -4 ) これを用いて同様の 320 年の積分を行った (ADJ) 。また比較として Tsujino, Gargett, Hasumi スキームのみを用いた実験(それぞれ TJN,GGT,HSM )を行った。 実験
15
Mean profile of vertical diffusive coefficient. Distributions of VDC at 2000m-depth CTL AD J TJN GG T HS M TJNGGT HSM ADJ K ADJ =0.43K TJN +0.08K GGT +0.72K HSM +…
16
Water temperature at 4000-5500m-depth ADJ(Green’s functions)WOA Observation ( left ) Tsujino ( mid ) Gargett ( right ) Hasumi ℃ (Toyoda et al. ‘09JOS annual mtg.)
17
Improved Ocean State Estimate
18
Cost function Assimilated elements: Temperature, Salinity (ENSEBLES v.3+JAMSTEC observations), SST (reconstructed Reynolds+OISST ver.2), SSH anomaly data (AVISO). First guess is generated from momentum, net heat, shortwave, latent heat flux of NCEP/DOE.
19
Time change of the each component of the cost function, i.e. the difference between simulation and observation. Reduction by iteration processes means progress in synthesis. SST SSHa 1991 2006 Subsurface T (←Argo) Optimal Synthesis (dynamical interporation) by 4D-VAR adjoint method
20
Our results provides a consistent view NCEP atmospheric reanalysis Reanarysis 50yr Assim.J-OFURO Comparing with other products RE50NCEP2 Estimated net heat flux (a control variable)
21
Our reanalysis data x ERA40 Estimated wind stress field (a control variable) Stress Jan Stress Jul Curl Jan Curl Jul
22
Climate indices during 1957-2006 * * * * Nino3 SST DMI ITF mass transport (8-14Sv) ACC mass transport (130-140Sv) Atlantic MOC (14-20Sv) * Bryden et al. (2005) Our result provides realistic time series of important climate indices.
23
Temporal change of global heat content Comparison with observed heat content trends. Comparison of year to year changes in heat content from 1960 implies => Our trend would be robust
24
Comparing with TOGA-TAO ADCP Also, obtained 4-D velocity field is by and large consistent with independent observations by TAO array. Validation for un-easily-observable variables
25
Atlantic 48N Indian Ocean 10S Equatorial Pacific Atlantic 25N CLIVAR/GSOP Intercomparison: Heat Transport Anomaly (PW)
26
Heat Transport Correlation (5 ys low pass) Global ECMWF GECCO INGV SODA GODAS K7
27
The observation number of salinity had been small before ARGO era ( ~ ‘00). As a result, the interannual variance of an OI dataset is relatively small for these periods. 4D-VAR dataset can resolve this issue thanks to both numerical model & adjoint method. interannual variance (psu 2 ) Number of obs. for subsurface salinity 0-100m 100-400m 400-700m 700-2000m 2000m- [ ] 1957-1966 [ ] 1967-1976 [ ] 1977-1986 [ ] 1987-1996 [ ] 1997-2006 [blue] an OI dataset [yellow] a model free run [red] our 4D-VAR reanalysis Salinity variances Toyoda et al. (GSOP’08) -- Advantage of our dataset
28
作成した統合データの T,S から見積もった steric height の経年 変化。 海洋環境再現データセット
29
Observed heat content Estimation form 50yr DA exp. Bottom water warming is of particular interest as it can be closely related to changes in the global thermohaline circulation and the warming trend of the global ocean (e.g., Fukasawa et al., 2004). Water temperature difference between WOCE/WOCE-revisit periods at 4000-5500m-depth is O(0.001-0.003K). Bottom water warming seems to be successfully reproduced in our reanalysis field. Bottom water warming in our reanalysis field
30
Dynamical Analysis for Climate Change
31
統合データセットを利用した海洋 貯熱量変動に関する研究(左)、 水塊の経年変動研究(下)。 NPZD モデル概念図 Kouketsu et al. (2010) Toyoda et al. (2010) Reanalysis data allow us to diagnose the real ocean
32
(Masuda et al. 2008) ±8.6 ±8.6 ±14.8 Reanalysis data allow us to reveal the physical mechanism of climate changes.
33
Adjoint Sensitivity Analysis
34
How about the Physical Mechanism? Our aim is to identify the possible causal dynamics, timescales and pathways involved in the observed bottom-water warming. Difference of the heat storage between WOCE-WOCE revisit observational periods. The physical mechanisms governing bottom-water warming are poorly understood since in-situ observations are spatially and temporally sporadic. The changes in heat storage between WOCE-WOCE revisit imply northward running of the warming signal, but… ?
35
Adjoint sensitivity analysis by using a 4D-VAR ocean DA system surface forcing ΔT(Bottom-water warming) The adjoint sensitivity analysis gives the temporal rate of change of a physical variable in a fixed time and space when model variables (e.g., water temperature, salinity, velocity, or surface air-sea fluxes) are arbitrarily changed in the 4-dimensional continuum of one temporal and three spatial coordinates. This is equivalent to specifying the “sensitivity” of a variable to small perturbations in the parameters governing the oceanic state. An adjoint sensitivity analysis moves the ocean representation backward in time!
36
感度解析 ある時刻 t0 ・地点 x0 での熱量 Q を変えるのに何が効いたのかという問題に対して、 時刻 t(<t0) に固定して考えると、時刻 t から t0 までの海面フラックス F と時刻 t での場 X とが候補になる。アジョイント方程式を解くことで各時間ステップtにおける これらのアジョイント変数 (adF, adX= 感度 ) を得る。それぞれの変化量に感度(= アジョイント変数)をかけて足し合わせたものが、熱量の変化量 dQ になる。 dQ = sum (dQ/dF_i) dF_i + sum (dQ/dX_i) dX_i = sum adF_i dF_i + sum adX_i dX_i * 単位の例 種 : dQ [cal] 温度の場合、 adX_i [cal /K]=[cm 3 s], dX_i [K] 熱フラックスの場合、 adF_i [cal /(cal/cm 2 /s)], dF_i [cal/cm 2 /s] 感度計算によって、 adX 、 adF が求められる。 時刻 t 場所 x に温度 1K ( or フラックス 1cal/cm 2 /s )与えたら、それが時刻 t0 における ターゲットの dQ のうち何 cal になるかを定量的に評価できる。 Sugiura (2009)
37
---A contour surface shows bottom-water warming rate when a constant change in water temperature is given--- Results of adjoint sensitivity analysis for a positive temperature anomaly in the abyssal North Pacific After 0-year
38
---A contour surface shows bottom-water warming rate when a constant change in water temperature is given--- Results of adjoint sensitivity analysis for a positive temperature anomaly in the abyssal North Pacific After 5-year
39
---A contour surface shows bottom-water warming rate when a constant change in water temperature is given--- Results of adjoint sensitivity analysis for a positive temperature anomaly in the abyssal North Pacific After 15-year
40
---A contour surface shows bottom-water warming rate when a constant change in water temperature is given--- Results of adjoint sensitivity analysis for a positive temperature anomaly in the abyssal North Pacific After 25-year
41
---A contour surface shows bottom-water warming rate when a constant change in water temperature is given--- Results of adjoint sensitivity analysis for a positive temperature anomaly in the abyssal North Pacific After 35-year
42
---A contour surface shows bottom-water warming rate when a constant change in water temperature is given--- Results of adjoint sensitivity analysis for a positive temperature anomaly in the abyssal North Pacific After 45-year
43
---Shade shows bottom-water warming rate when a constant change in water temperature is given--- Results of adjoint sensitivity analysis for a positive temperature anomaly in the abyssal North Pacific Antarctica New Zealand Antarctica Tasman Sea 170 o E cross-section 160 o E cross- section After 48-year
44
---Contour shows bottom-water warming rate when a constant change in surface heat flux is given--- Results of adjoint sensitivity analysis for a positive temperature anomaly in the abyssal North Pacific After 45-year Source region: Antarctic Sea off Adelie Coast Time scale: 40 year in contrast to the previous estimation of O(multi-centennium)!!
45
WOCE WOCE_rev 底層水温上昇 時間変化 水平移流 鉛直移流 水平拡散 鉛直拡散 (含む地熱効果) Possible mechanism for bottom water warming Masuda et al. (2010)
46
Our scenario should be tested by direct observations… Also, using 4D-VAR data synthesis, sustainable observations is needed to better representation for a global ocean. Argo, revisit cruise is really enhancing the quality of reanalysis products. WOCE sections S04 line P14S line South America Antarctica Australia Africa Pacific Ocean Indian Ocean Atlantic Ocean
47
Toward an Optimal Ocean Observing System
48
観測システム研究 アジョイント感度解析を用いた NINO3 亜表層水温の変動要因の分 布( adT )。 四次元変分法データ同化システムを 利用しアジョイント感度解析を行う ことで ENSO に関する水温変動をひき おこす水温変化の時間スケール、空 間分布を同定した。 0-month -2-month -4-month 0-month
49
3:観測システム研究 アジョイント感度解析から得られた結果を元 に、観測システムシミュレーション (同化) 実験をおこなった。 西部熱帯太平洋の観測網を変化させる (0-100m 0-200m )ことで、東部海域 における水温構造の再現性に差が生じ ることを確認した。 Adjoint source region
50
Summary 4D-VAR data assimilation method is applied to deep ocean reanalysis experiment to obtain a dynamically-self consistent ocean state estimation from surface to bottom (1967-2006). The reanalysis dataset is capable of representation of the recent climate change also in the abyssal ocean. An adjoint sensitivity analysis implies that an increase in the heat input into the Southern Ocean off the Adélie Coast of Antarctica leads to bottom water warming in the North Pacific on a relatively short time scale (within four decades). An adjoint sensitivity analysis was applied to detect an optimal ocean observation system in the equatorial Pacific. Tentative results show further applications in line with this study are promising.
51
Future Work
52
Down-scaling attempt with MOM4 We are trying to implement a 1/16 x 1/16 x 45level regional model with CDA products as boundary conditions (<= Tropical Climate Variability Research Program). Polar region model is also starting; tri-porlar, ice model,… (<= Northern Hemisphere Cryosphere Program) High-resolution data assimilation is now under considerations.
53
Application to Ecosystem model NEMURO is incorporated in our system; collaboration with Environmental Biogeochemical Cycle Research Program
54
END
Similar presentations
© 2025 SlidePlayer.com Inc.
All rights reserved.