90A_2012-A23

Embed Size (px)

Citation preview

  • 7/29/2019 90A_2012-A23

    1/12

    Journal of the Meteorological Society of Japan, Vol. 90A, pp. 385--396, 2012. 385

    DOI:10.2151/jmsj.2012-A23

    NOTES AND CORRESPONDENCE

    Regional Patterns of Wintertime SLP Change over the North Pacific

    and Their Uncertainty in CMIP3 Multi-Model Projections

    Kazuhiro OSHIMA

    Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Japan

    Youichi TANIMOTO

    Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Japan

    Research Institute for Global Change, Japan Agency for Marine-Earth Science and Technology, Yokohama, Japan

    and

    Shang-Ping XIE

    International Pacific Research Center and Department of Meteorology, School of Ocean and Earth Science and

    Technology, University of Hawaii at Manoa, Honolulu, Hawaii

    (Manuscript received 9 May 2011, in final form 14 October 2011)

    Abstract

    Regional patterns of wintertime sea level pressure (SLP) trends over the North Pacific and their uncertaintywere investigated based on the phase 3 of the Coupled Model Intercomparison Project (CMIP3) multi-model pro-jections under the Special Report on Emissions Scenarios (SRES) A1B emission scenario for the 21st century(20002099). While the 24-model ensemble mean of the 100-yr SLP trend over the North Pacific shows a north-ward shift of the Aleutian low (AL), regional patterns of the SLP change vary among the models. Projectedchanges deepen the AL in several models but it shifts northward in some others. The dierent response of theAL results in a large inter-model spread over the North Pacific, which is largest of the Northern Hemisphereand comparable in magnitude to the ensemble mean in the same region. This large spread means a high degreeof uncertainty in the 100-yr SLP trend over the North Pacific.

    For the total uncertainty in the SLP trends over the North Pacific, we examined the relative importance of theinternal climate variability and model uncertainty due to dierent treatments of physical processes and computa-tional scheme. To evaluate each of contributions, a single-realization ensemble using a subset of 10 CMIP3 mod-els is compared to a multi-realization ensemble for the same models in the A1B projections. Additionally the con-trol simulations under preindustrial conditions are examined to evaluate the background internal variability in

    each of the CMIP3 models. Our analysis shows that both the model uncertainty and internal climate variabilitycontribute to the total uncertainty in the 100-yr SLP trends during the 21st century, while the internal climatevariability largely explains the total uncertainty in the 50-yr SLP trends during the first half of the 21st century.

    Corresponding author and present aliation: Kazu-hiro Oshima, Research Institute for Humanity and Na-ture, 457-4 Motoyama, Kamigamo, Kita-ku, Kyoto603-8047, Japan.E-mail: [email protected] 2012, Meteorological Society of Japan

  • 7/29/2019 90A_2012-A23

    2/12

    The changes in surface heat flux and North Pacific subtropical gyre in association with the dierent responseof the AL aect regional patterns of the sea surface temperature trends among models.

    1. Introduction

    To reveal future climate changes in the 21st cen-tury, it is important to examine regional patterns inthe mean state of the atmosphere and ocean underglobal warming. Several studies exist on regionalchanges of North Pacific climate under globalwarming by using the phase 3 of the CoupledModel Intercomparison Project (CMIP3) modelprojections in the 21st century (Hori and Ueda2006; Salathe 2006). They found that the multi-model ensemble mean of sea level pressure (SLP)trend over the North Pacific shows a northward

    shift by 34 latitude and deepening by about0.5 hPa of the Aleutian low (AL) during the 21stcentury. Yamaguchi and Noda (2006) showed thatthe individual CMIP3 model projections under theglobal warming display dierent spatial patternsof the SLP trends over the North Pacific. Becausethe greenhouse gas (GHG) emission scenario wasthe same in these model projections, the dierentregional patterns of the SLP trends may be mostlycaused by the model uncertainty due to dierent re-sponses to the increased GHG depending on dier-ent treatments of physical processes and dierentcomputational schemes (e.g., Deser et al. 2010;Hawkins and Sutton 2009; 2011).

    In addition to the model uncertainty, internal cli-mate variability in individual models can aect re-gional SLP projections. Deser et al. (2010) eval-uated contributions of internal climate variabilityto the total uncertainty in the SLP trends overthe first half of the 21st century based on a 40-member ensemble using the National Center forAtmospheric Research Community Climate SystemModel, version 3 (NCAR CCSM3). They revealedthat the contribution of internal variability wasdominant in the total uncertainty of the SLP trends

    over mid- and high-latitudes in the Northern Hemi-sphere. Hawkins and Sutton (2009, 2011) quantita-tively evaluated each contribution of the model un-certainty, internal climate variability and emissionscenario to the total uncertainty in the trends ofsurface air temperature (SAT) and precipitation(PR), respectively, using the CMIP3 model projec-tions under the global warming. They indicatedthat the relative importance of the three uncertaintyfactors depends on the analyzed variables, selected

    region and projected time period of the trends. Asfor regional SAT trends, internal variability andmodel uncertainty are the dominant contributorsof the total uncertainty in the trends up to 20202030. The model uncertainty is of greater impor-tance in the trends of 20202040. As for regionalPR trends, while the internal variability is the dom-inant contributor to the total uncertainty in manyselected regions up to 2030, the model uncertaintyis generally dominant beyond 2030. In the presentstudy, we attempt to evaluate relative contributionsof the model uncertainty and internal climate vari-ability to the total uncertainty in the regional SLP

    trends over the North Pacific based on the CMIP3multi-model projections.

    Several recent studies showed spatial characteris-tics of sea surface temperature (SST) variability inthe North Pacific under the global warming (Fur-tado et al. 2011; Oshima and Tanimoto 2009; Over-land and Wang 2007; Wang et al. 2009). Thesestudies found that the decadal SST variability inthe 21st century follows the spatial pattern ofthe Pacific Decadal Oscillation (PDO) found in theobserved records and the 20th century climatein coupled models (20C3M) simulations of theCMIP3 models. As for the SST warming trends inthe 21st century, while many studies have examinedregional patterns of the trends in the tropical Pacific(e.g., Collins et al. 2005; DiNezio et al. 2009; Liuet al. 2005; Xie et al. 2010), there are few studies inthe literature that examines regional patterns of theSST warming trend in the mid- and high-latitudeNorth Pacific and its association with the overlyingatmospheric changes. Dierent regional patterns ofthe SLP may induce the dierent regional patternsof SST. The present study further examines the pro-cesses by which regional patterns of SLP trend in-duce regional patterns of SST trend in the North

    Pacific.In the rest of this paper, Section 2 introducesdatasets and analysis methods to study the pro-

    jected trends and their uncertainty. Section 3 showsthe results of the regional dierences in the SLPtrends over the North Pacific in the CMIP3 models.Section 4 evaluates the uncertainty in the SLPtrends. Section 5 discusses the relationship in re-gional spatial patterns between the SLP and SSTtrends. Finally, Section 6 is summary.

    386 Journal of the Meteorological Society of Japan Vol. 90A

  • 7/29/2019 90A_2012-A23

    3/12

    2. Data and analysis method

    We used the global warming projections underthe SRES A1B emission scenario from 2000 to

    2099 and control simulations under preindustrialconditions (PICTL) in the 24 CMIP3 models (Meehlet al. 2007). For simplicity, we named Model A toX as listed in Table 1. Since our focus is on climatechange in and around the North Pacific, we define

    the North Pacific sector as the region of 1570N,120E120W (inset in Fig. 1a). Monthly outputsof SLP, SST, surface zonal wind (UAS), surfaceturbulent heat flux (SHF, sum of latent and sensible

    heat fluxes) and sea surface height (SSH) were ana-lyzed. SST from the Hadley centre sea Ice and SeaSurface Temperature (HadISST, Rayner et al.2003), SLP, UAS and SHF based on the Japanese25-year Reanalysis (JRA25, Onogi et al. 2005) and

    Table 1. A list of the 24 CMIP3 model (group A) projections under the SRES A1B scenario and the PICTL simula-tion used in this study. In the second column, the selected 10 models (group B and C) having more than three ensem-ble members are marked. In the third column, the number of ensemble members in the individual model projectionsunder the A1B scenario is shown. In the fourth and fifth column, models with strong negative and strong positiveNPI (North Pacific index) trends are marked.

    Model

    24

    models

    10

    models

    Number of

    ensemble members

    negative

    NPI trend

    positive

    NPI trendA BCCR-BCM2.0 (Norway) 3 1

    B CGCM3.1_T47 (Canada) 3 3 5

    C CGCM3.1_T63 (Canada) 3 1 3

    D CNRM-CM3 (France) 3 1

    E CSIRO-MK3.0 (Australia) 3 1 3

    F CSIRO-MK3.5 (Australia) 3 1

    G GFDL-CM2.0 (USA) 3 1

    H GFDL-CM2.1 (USA) 3 1

    I GISS-AOM (USA) 3 1J GISS-EH (USA) 3 3 3 3

    K GISS-ER (USA) 3 3 5 3

    L FGOALS-g1.0 (China) 3 3 3

    M INGV-SXG (Italy) 3 1

    N INM-CM3.0 (Russia) 3 1

    O IPSL-CM4 (France) 3 1

    P MIROC3.2_hires (Japan) 3 1

    Q MIROC3.2_medres (Japan) 3 3 3 3

    R ECHO-G (Germany/Korea) 3 3 3 3S ECHAM5_MPI-OM (Germany) 3 3 4

    T MRI-CGCM2.3.2 (Japan) 3 3 5

    U NCAR-CCSM3 (USA) 3 3 7

    V NCAR-PCM1(USA) 3 3 4 3

    W UKMO-HadCM3 (UK) 3 1

    X UKMO-HadGEM1 (UK) 3 1 3

    February 2012 K. OSHIMA et al. 387

  • 7/29/2019 90A_2012-A23

    4/12

    SSH based on the Simple Ocean Data Assimilationreanalysis (SODA, Carton et al. 2000) were used asthe observed references of the 20th century climate.All monthly data were bilinearly interpolated onto

    a common 5 longitude 5 latitude grid for easycomparison. We calculated monthly anomaliesby subtracting the monthly climatologies over the21st century in the A1B scenario. These monthlyanomalies were averaged over boreal winter (DJF;December, January and February) and thensmoothed with the 5-year running mean.

    We calculated linear trends at each grid pointusing the least squares linear regression for all vari-ables from 100-yr (50-yr) DJF mean anomaly re-cords during 20002099 (20002049). Mean andvariance of the SLP trends in the 24 CMIP3 modelsare the multi-model ensemble mean and inter-

    model spread, respectively.As for analysis of the uncertainty, we employed

    the following procedure. For a model with severalensemble members, we assume that the multi-member mean of the model removes the un-certainty due to internal climate variability. Thusthe spread of multi-member ensemble means amongindividual models is considered to be induced onlyby model uncertainty. This statistical treatment ofuncertainties in the climate trend was used in previ-ous studies (e.g., Deser et al. 2010; Yukimoto andKodera 2005). In the CMIP3 model projectionsunder the A1B scenario, only 10 models have threeor more ensemble members (Models B, J, K, L, Q,R, S, T, U and V, Table 1). The mean and varianceof the trends in the multi-member ensembles ofthese 10 models refer to the multi-member ensem-ble mean and inter-member spread, respectively.

    The multi-model ensemble mean and inter-modelspread of the SLP trends are calculated in each ofthe following groups (A, B and C, Table 2). Forgroup A, we chose one simulation each from the

    24 CMIP3 models. Groups B and C are used to es-timate internal variability. Group B consists of threesubsets, each made of 10 runs, one from each of theselected 10 CMIP3 models that contain three or

    more members. The mean of variance from threesubsets of group B represents the uncertainty dueto the internal variability and model uncertainty.For group C, we chose the three-member ensemblemeans of the 10 models. The simulations in thegroup C are the same as in the group B. The inter-model spreads of the SLP trends in the groups Aand B contain the uncertainty both from the inter-nal climate variability and model uncertainty, whilethe inter-model spread in the group C is assumed tocontain only the model uncertainty. For group D,we used 40 members of the model U. Note thateach member of model U was integrated only

    during 20002061 (Deser et al. 2010). The inter-member spread in the group D is also assumed tocontain the uncertainty only due to the internal cli-mate variability.

    As an alternative, we estimate the internal cli-mate variability using the PICTL simulations.Group A of the PICTL simulations consists of 24100-yr records, one each from the 24 models. Forthe 24 models, we calculated the 100-yr (50-yr)SLP trends from 100-yr (first 50-yr) records. GroupB is made 300-yr runs from a subset of 10 models.We calculated 100-yr (50-yr) SLP trends from each300-yr PICTL run. For each model, there are 21such 100-yr samples with the starting years delayedsuccessively by one decade. We used the average ofthe spreads among these 21 of 100-yr (50-yr) trendsas the group-B PICTL simulations. Without GHGincrease, the inter-model spread of the PICTL sim-ulation contains the uncertainty only from the in-ternal climate variability. We assume that theamounts of the internal variabilities are comparableboth in the PICTL and A1B simulations.

    Table 2. Categories of four groups used in the analysis of the uncertainty.

    group EnsembleA Multi-model ensemble based on 24 of single members from each of the 24 CMIP3 models

    B Multi-model ensemble based on three subsets with 10 of single members from the selected 10 CMIP3models having more than three ensemble members (marked in the second column in Table 1);Average over the three subsets is used as group B.

    C Multi-model ensemble based on 10 of multi-member ensemble mean averaged over three members fromthe selected 10 CMIP3 models (marked in the second column in Table 1)

    D Multi-member ensemble based on 40 ensemble members of the model U (NCAR CCSM3)

    388 Journal of the Meteorological Society of Japan Vol. 90A

  • 7/29/2019 90A_2012-A23

    5/12

    3. Regional patterns of SLP trends and

    their spread over the North Pacific

    Regional patterns of the wintertime SLP trendsduring 20002099 are investigated based on the 24CMIP3 model projections (group A) under the A1Bscenario. The multi-model ensemble mean of thegroup A (contours in Fig. 1a) shows negative SLPtrends over the high-latitudes with a minimum of5 hPa 100yr1 over the northern Bering Sea(65N, 180) and a weak local minima over theCanadian archipelago and Barents Sea. PositiveSLP trends are found over the southwest portionof the North Pacific (25N40N, 135E170W),most of the North Atlantic (30N55N, 10W

    60

    W) and around the Mediterranean Sea. The in-ter-model spreads of the SLP trends (gray shades inFig. 1a) over the North Pacific, Arctic and NorthAtlantic sectors are comparable in magnitude tothe multi-model ensemble mean, showing a largeuncertainty in SLP trends over these regions. Wenote that the inter-model spread of the group Acan be induced by both the model uncertainty andinternal climate variability. To further examine thepolarity of the SLP trends over the Northern Hemi-

    sphere, we counted the number of models whoseSLP trend is greater (less) than 0.5 (0.5) hPa100yr1 at each grid point. More than 90% of theCMIP models display the negative SLP trendsaround the Bering Sea and Canadian archipelago(Fig. 1b), while 5070% of the models display thepositive trends over the southwest portion of theNorth Pacific, the North Atlantic and Mediterra-nean Sea. This good agreement in the polarity ofthe SLP trends indicates that the large inter-modelspread over these regions is due to dierences inthe magnitude of the SLP trends among modelsrather than the polarity of the trends.

    Over the North Pacific sector, the multi-modelensemble mean of the group A features a center of

    the negative SLP trends over the Bering Sea, and azonal band of the positive trends that extends fromeast of Japan to the central North Pacific along35N (Fig. 1a). This pattern indicates a northwardshift of the AL consistent with the previous studies(Hori and Ueda 2006; Salathe 2006). The inter-model spread over the North Pacific is largest witha maximum at 12.5 (hPa 100yr1)2 over the north-eastern portion of the sector (50N, 155W, Fig. 1a),where 3040% of the models display positive SLP

    Fig. 1. (a) Multi-model ensemble mean (contours at 1 hPa 100yr1 intervals) and inter-model spread (shadesin (hPa 100yr1)2) of SLP trend during 20002099 in the group A. Supplemental G0.5 hPa 100yr1 con-tours are added and zero contour is omitted. The solid inset region (15N70N, 120E120W) indicatesthe North Pacific sector. (b) Ratio of the number of models with the significant positive (red contours) andnegative (blue contours) SLP trends in the group A. The models with the trend greater (less) than 0.5(0.5) hPa 100yr1 at each grid point were employed for the ratio calculations. Ratios greater than 30%are shown and the contour interval (CI) is 10%.

    February 2012 K. OSHIMA et al. 389

  • 7/29/2019 90A_2012-A23

    6/12

    trends while other 3040% of the models displaynegative ones (Fig. 1b). Thus the large inter-modelspread over the northeastern Pacific is due to dier-ences in polarity of the SLP trends among models.This means that the northward shift of AL found inthe multi-model ensemble mean is not a consistentfeature among group A models.

    To identify the dominant pattern that accountsfor the large inter-model spread of the SLP trendover the North Pacific sector, we performed an em-pirical orthogonal function (EOF) analysis on the100-yr SLP trends of the group A. The first EOFmode (EOF1, Fig. 2a) explains 46.0% of the totalinter-model variance with a center of positivetrends over the northeastern portion of the sector

    (50

    N, 160

    W), close to the climatological centerof the AL (not shown). There are weak centers ofnegative trends over northern Eurasia and the Arc-tic Ocean (Fig. 2a). The scores of the EOF1 arepositively correlated with the trends of the NorthPacific index (NPI, Fig. 2b) among individual mod-els, which is a measure of the AL strength (Tren-berth and Hurrell 1994). The correlation coecientbetween the EOF1 scores and NPI trends is 0.97above 99% significant level.

    4. Sources of uncertainty in SLP trends

    over the North Pacific

    To evaluate relative contributions of the modeluncertainty and internal climate variability to thetotal uncertainty in the SLP trends over the NorthPacific sector, we compared the inter-model spreadsof the SLP trends during 20002099 among thethree groups (A, B, C) of the A1B projections inthe CMIP3 multi-models. The spatial pattern ofthe inter-model spread in the group A over theNorth Pacific sector (Fig. 3a, same as gray shadesin Fig. 1a) is very similar to that in the group B(Fig. 3b), and that in each of the three subsets ofthe A1B group B (not shown). These results indi-

    cate that the inter-model spread is not aectedwhen the number of the models is reduced to 10from 24. We note that the inter-model spread ofthe group B, like that of the group A, contains theuncertainty owing to the model uncertainty andinternal climate variability. Since the contributionof internal variability can be small in the multi-member mean for each of the models, the inter-model spreads in the group C (Fig. 3c) are reducedcompared to the groups A and B (Figs. 3a and 3b).

    Fig. 2. (a) Spatial pattern of EOF1 (46.0% explained variance) for SLP trend during 20002099 in the groupA. CI is 0.5 hPa 100yr1 and the zero contour is omitted. (b) Scatter plot of the EOF1 scores and trends ofNorth Pacific index (NPI), which is defined as an area-weighted mean of SLP anomalies over 3065N,160E140W, in the group A. Labels of A X represent the 24 CMIP3 models (Table 1). Uppercase lettersindicate the 10 models (Models B, J, K, L, Q, R, S, T, U and V) used for the groups B and C. Four modelswith strong negative (Models C, E, V and X) and four models (Models J, K, Q and R) with strong positiveNPI trends are marked with circles. Correlation coecient for this scatter plot is 0.97 above 99% significantlevel.

    390 Journal of the Meteorological Society of Japan Vol. 90A

  • 7/29/2019 90A_2012-A23

    7/12

    Significant decrease (@30%) is found over thenortheastern portion of the North Pacific. On thebasin-scale, the spatial pattern of the inter-modelspread in the groups A and B looks similar to thatin the group C, especially in the western North Pa-

    cific. In the basin-average over the North Pacific,the model uncertainty measured by the inter-modelspread of the group C (row II in Table 3) accountsfor 72% (2.3 (hPa 100yr1)2) of the total uncer-tainty measured by the inter-model spread of the

    Fig. 3. (ac) Inter-model spreads (shades; (hPa 100yr1)2) of the 100-yr SLP trend during 20002099 overthe North Pacific sector in the group (a) A, (b) B, and (c) C. Contours are dierences in the inter-modelspread (5 (hPa 100yr1)2 intervals): (b) group A minus B, (c) B minus C. (dg) As in (ac), but for the50-yr SLP trend (hPa 100yr1)2 during 20002050 in the group (d) A, (e) B, (f ) C, and (g) D. Note thatthe categories of the four groups are described in Table 2.

    February 2012 K. OSHIMA et al. 391

  • 7/29/2019 90A_2012-A23

    8/12

    group B (row I in Table 3), while the internal vari-ability defined as a residual by removing the modeluncertainty from the total uncertainty (row III inTable 3) is 0.9 (hPa 100yr1)2. This result suggests

    that the model uncertainty largely explains the totaluncertainty in the 100-yr SLP trends over most ofthe North Pacific, while internal climate variabilitycontributes to the total uncertainty over the easternNorth Pacific.

    We further examined uncertainty in the SLPtrends for a shorter period in the near-future projec-tion during 20002049 for the same set of fourgroups in the A1B projections. The inter-modelspreads of the 50-yr SLP trends in the groups A, Band C (Figs. 3d, 3e and 3f ) tend to be greater thanthose of the 100-yr SLP trends (Figs. 3ac), whilethe spatial patterns of those spreads still show thecommon feature of a significant center in the north-eastern portion of the North Pacific, as in Figs. 3ac. The dierences in the inter-model spreads of the50-yr trends between the groups A and B (Figs. 3dand 3e) tend to be larger than those of the 100-yrtrends (Figs. 3a and 3b), indicating that the inter-model spread of the 50-yr trends strongly dependson the selection of the models. Such dierences arealso seen in each of the three subsets of the group B(not shown). On the basin-scale, the inter-modelspread of the 50-yr trends in the group C (Fig. 3f )is much reduced than in the group B (Fig. 3e). In

    the area-average over the North Pacific sector, themodel uncertainty (the inter-model spread of thegroup C; row II in Table 3) and the internal vari-ability (the residual; row III in Table 3) accountfor 41% (4.7 (hPa 100yr1)2) and 59% (6.7 (hPa100yr1)2) of the total uncertainty (row I in Table3), respectively. These results indicate the substan-tial contribution of the internal climate variabilityto the total uncertainty in the 50-yr SLP trendscompared to the 100-yr trends. Indeed, the inter-

    member spread among 40 ensemble members ofthe model U (Deser et al. 2010) is very large inthe northeastern portion of the North Pacific(Fig. 3g).

    One may ask whether the number of the en-semble members is sucient to remove the internalvariability in the group C of the A1B projection.Here we employ the PICTL simulations to estimatethe background internal variability of the CMIP3models. Without GHG increase, the inter-modelspreads of the PICTL group B measure the un-certainty from the internal climate variability. Asfor the 100-yr SLP trends in the PICTL group B,the area-average of the inter-model spread over theNorth Pacific sector is about 1.5 times larger inthe PICTL simulation (1.4 (hPa 100yr1)2, row IVin Table 3) than in the A1B projection (0.9 (hPa100yr1)2, row III in Table 3). The spread for the50-yr trends in the group B of the PICTL simula-tion is even larger (7.0 (hPa 100yr1)2, row IV inTable 3). The residuals in the A1B projection (rowIII in Table 3) underestimate the internal variabil-ity. Using the inter-model spread of the group B inthe PICTL simulation as the better estimation ofthe internal variability, the revised model uncer-tainty is 1.8 (hPa 100yr1)2 for the 100-yr trend and4.4 (hPa 100yr1)2 for the 50-yr trend, respectively(row V in Table 3). The inter-model spreads of the24 models (group A) are comparable to those of the

    10 models (group B). This improved statistics showthat both the model uncertainty and internal cli-mate variability contribute to the total uncertaintyin the 100-yr SLP trends during the 21st century,and that the latter contribution is much greater inthe 50-yr trend during the first half of the 21st cen-tury. Based on our analyses of the A1B and PICTLsimulations, we note that a large number of themodels, each with multi-member integrations, isnecessary for a robust near-future projection.

    Table 3. Area-average of the inter-model spread (unit: (hPa 100yr1)2) over the North Pacific sector and related sta-tistics of the 100-yr and 50-yr SLP trends in the A1B and PICTL simulations.

    Analyzedtrend period

    RowNo.

    Statistics and analyzeddataset Uncertainty 100-yr 50-yr

    (I) Spread of group B in A1B Total uncertainty in A1B 3.2 11.4( II ) Spread of group C in A1B Tentative model uncertainty only based on A1B 2.3 4.7( III ) Residual: ( I) minus ( II ) Tentative internal variability only based on A1B 0.9 6.7( IV ) Spread of group B in PICTL Background internal variability in PICTL 1.4 7.0(V) Residual: (I) minus (IV) Better estimation of model uncertainty 1.8 4.4

    392 Journal of the Meteorological Society of Japan Vol. 90A

  • 7/29/2019 90A_2012-A23

    9/12

    5. Relationship in regional patterns between

    SLP and SST trends

    As documented in Section 3, the polarity of the

    SLP trend over the North Pacific is dierent amongthe group A models (Fig. 1b). To examine pro-cesses by which the regional dierences in SLPtrend induce those in SST trend in the North Pa-cific, we calculated composites of the trends in fourmodels with strong negative NPI trend (Models C,E, V and X, listed in Table 1 and marked withcircles in Fig. 2b) and in four models with strongpositive NPI trend (Models J, K, Q and R, listedin Table 1 and marked with circles in Fig. 2b). Asexpected from the significant correlation betweenthe EOF1 scores and the NPI trends (Fig. 2b), theformer (latter) models show the negative (posi-

    tive) SLP trends over the North Pacific (Figs. 4aand 4g).

    In the four models with the strong negative NPItrend, the SLP composite features a negative centerof the minimum of8 hPa 100yr1 over the BeringSea (Fig. 4a). This corresponds to a deepeningtrend of the AL (Fig. 4c). In association with thisdeepened AL, UAS are enhanced by 1.5 m s1

    100yr1 (Fig. 4d). This enhanced westerly jet acts

    to increase the SHF by 1025 W m 2 100yr1 in theextratropical North Pacific (30N45N, Fig. 4e).As a result, the SST trends in the western and cen-tral North Pacific are smaller by 0.21C 100yr1

    than the basin-average SST trend of 2.34C 100yr1(Figs. 4a and b). While an increase of meridionalgradient of the SSH between 30N45N suggestsan intensification of the subtropical gyre in theNorth Pacific (Fig. 4f ), its eect on the SST trendseems weak. Overall, patterns of SST change areconsistent with changes in prevailing wind, with lo-cally enhanced warming over reduced wind speed.

    In the four models with strong positive NPItrend, the SLP composite shows a positive centerwith a maximum of 5 hPa 100yr1 over the centralNorth Pacific (40N, 175W, Fig. 4g). The center ofthe AL moves northward by 5 latitude (Fig. 4i).

    This shift of the AL induces enhanced (reduced)surface westerly trends in 45N60N (20N35N), indicative of a northward shift of the west-erly jet by 5 latitude (Figs. 4g and 4j). The shift ofthe westerly jet is accompanied by an eastwardanomalous jet in ocean current centered at 40N(Fig. 4l) anchoring a locally enhanced warming atthe same latitude in the western basin (Fig. 4h).The enhanced SST warming along 40N takes

    Fig. 4. (af ) Composites of the four models with strong negative NPI trends. (a) SLP trend (contours at1 hPa 100yr1 intervals), regional dierence in the SST trend (colors; C 100yr1), and UAS trend (arrows;m s1 100yr1). Note that the SST trend represents deviation from area-average over the North Pacificsector which is 2.34C 100yr1. (bf ) Meridional plots of zonal mean present-day climatology (blackline), future climatology (red line), and trends (dierences between the future and present-day climatolo-gies, green line). (b) SST, (c) SLP, (d) UAS, (e) SHF and (f ) SSH averaged over 140E160W in the NorthPacific. Note that the axes for the climatologies are at the bottom of the panel, but the axis for the trend isat the top of the panel. (gl) As in (af ), but for the four models with strong positive NPI trends. Area-average of SST trend over the sector is 1.93C 100yr1.

    February 2012 K. OSHIMA et al. 393

  • 7/29/2019 90A_2012-A23

    10/12

    place even though the SHF trend features a positive(i.e., enhanced heat release from the ocean) peak at40N (Figs. 4h and 4k). This indicates that the en-hanced SHF is a result of the SST warming in re-

    sponse to the changes in the subtropical gyre ratherthan a cause of the SST trend. On the horizontalmap, the enhanced SST warming shows a westwardintensification with little change in local westerlywinds (Fig. 4g). This is consistent with an intensi-fied and northward expanded subtropical gyre.

    6. Summary

    The regional patterns of wintertime SLP trendsover the North Pacific and their uncertainty wereinvestigated based on 24 CMIP3 model projectionsunder the A1B scenario and control simulationsunder pre-industrial conditions (PICTL). To evalu-

    ate uncertainty of the SLP trend, we analyzed themulti-model ensemble means and their spread infour groups (groups A, B, C and D, Table 2) ofthe A1B projections and in the two groups (groupsA and B) of the PICTL simulations.

    In the A1B projection, the multi-model ensemblemean of the 100-yr SLP trends during 20002099over the North Pacific based on single memberruns with 24 CMIP 3 models (group A) shows adeepening and northward shift of the AL (Fig. 1a),consistent with previous studies (Hori and Ueda2006; Salathe 2006). Several models in the groupfeature a deepening of the AL while several othersa northward shift of the AL. This large inter-modelspread of the SLP trend over the North Pacific(Figs. 1 and 2) indicates a large uncertainty of SLPchange over this region.

    We have evaluated the relative contributions ofthe model uncertainty and internal climate variabil-ity to the total uncertainty in the SLP trends usingtwo groups from the selected 10 models that havethree or more ensemble members. Group B of theA1B projection, made of three subsets of 10 singleruns from dierent models, is used to estimate thetotal uncertainty, while group C is a multi-member

    ensemble mean from each of these models, wherewe assume that the internal variability is removed.If we assume that the internal variability is su-ciently removed in the group C, the inter-modelspreads in the group C of the A1B projections indi-cate that the model uncertainty dominates the totaluncertainty in the 100-yr SLP trend over the NorthPacific while internal climate variability plays aminor role (Figs. 3b and c). For the shorter projec-tion of the 50-yr trend, the relative contribution

    from internal climate variability increases (Figs.3eg) in agreement with Deser et al. (2010).

    Because of the limited number of ensemble mem-bers in the A1B group C, we additionally employ

    the PICTL simulation to evaluate the backgroundinternal variability of the SLP trend in the CMIPmodels. Without GHG increase, the inter-modelspreads of the PICTL group B measure the un-certainty from the internal climate variability. Ouranalysis of the PICTL simulations shows that thelimited number of A1B projections underestimatesthe internal variability over the North Pacific bothfor the 100-yr and 50-yr trends. The revised esti-mates indicate that both the model uncertainty andinternal climate variability contribute to the totaluncertainty in the 100-yr SLP trends over the NorthPacific, while the internal variability largely ex-

    plains the total uncertainty in the 50-yr trends(Table 3). These results have implications for thediscussion of future climate change over the NorthPacific. Neither multi-member projections with aparticular model nor multi-model ensemble meanof single member runs provide a robust projectionand associated uncertainty. For a reliable future cli-mate projection in the North Pacific, it is necessaryto conduct multi-member ensemble projections us-ing multi-models to reduce the internal variabilityand model uncertainty, respectively.

    The dierent response of the projected SLP trendover the Aleutian Low leads to dierent regionalpatterns of SST trend in the North Pacific (Figs. 4aand 4g). In the four models with the strong negativeNPI trend in the group A, the changes in prevailingwind explains spatial pattern of SST warming, withincreased SHF from the ocean in association withthe deepening of the AL causing reduced SSTwarming in the central North Pacific (Figs. 4a f ).By contrast, in the four models with the strong pos-itive NPI trend in the group A, the northward shiftof the subtropical gyre in the North Pacific, in re-sponse to the similar northward shift of the AL, in-duces a band of enhanced SST warming trend in the

    northwest Pacific along with subtropical-subpoalrgyre boundary (Figs. 4gl). Thus, the inter-modelspread of the SST warming pattern in the NorthPacific is strongly dependent on the dierences inthe SLP trend near the AL.

    Distinctive regional dierences in ocean responseemerge in the North Pacific as shown here and byXie et al. (2010). Besides the gyre-scale adjust-ments, recent studies indicate the importance ofchanges in mode water ventilationa higher verti-

    394 Journal of the Meteorological Society of Japan Vol. 90A

  • 7/29/2019 90A_2012-A23

    11/12

    cal mode processin surface current and SST re-sponse to global warming (Xie et al. 2011; Xu et al.2012). Pacific Ocean adjustments to wind andbuoyancy changes need further investigations.

    The changes in SST and ocean circulation underthe global warming are important not only for cli-mate, but for the marine ecosystem (Wang et al.2010). Recently, Yara et al. (2011) assessed the im-pact of SST warming and its uncertainty on coralsnear Japan based on the CMIP3 multi-model pro-

    jections. Our results about regional dierences inSST warming and changes in the oceanic gyre pro-vide useful information for such assessments of themarine ecosystem change.

    Acknowledgments

    This work was supported by the Global Environ-ment Research Fund (S-5) of the Ministry of theEnvironment, Japan, JAMSTEC, NSF and JSPSInstitutional Program for Young Researcher Over-seas Visits. We acknowledge the modeling groupsfor providing their data for analysis, the Programfor Climate Model Diagnosis and Intercomparison(PCMDI) for collecting and archiving the modeloutput, and the JSC/CLIVAR Working Group onCoupled Modeling (WGCM) for organizing themodel data analysis activity. The multi-model dataarchive is supported by the Oce of Science, U.S.Department of Energy. We also express our grati-

    tude to the Data Integration and Analysis Sys-tem Fund for National Key Technology from theMinistry of Education, Culture, Sports, Science andTechnology, Japan, for providing us with an in-valuable environment for mass data handling. Weacknowledge the CCSM Climate Variability andClimate Change Working Groups 21st centuryCCSM3 Large Ensemble Project, for providingtheir data. NCAR Command Language (NCL)and GrADS were used for analysis and drawingfigures.

    References

    Collins, M., and The CMIP modeling groups, 2005: ElNino or La Nina-like climate change? Clim. Dyn.,24, 89104.

    Carton, J. A., G. Chepurin, and X. Cao, 2000: A SimpleOcean Data Assimilation analysis of the globalupper ocean 19501995 Part 2: results. J. Phys.Oceanogr., 30, 311326.

    Deser, C., A. S. Phillips, V. Bourdette, and H. Teng,2010: Uncertainty in climate change projections:

    The role of internal variability. Clim. Dyn.,doi:10.1007/s00382-010-0977-x.

    DiNezio, P. N., A. C. Clement, G. A. Vecchi, B. J.Soden, B. P. Kirtman, and S.-K. Lee, 2009: Cli-

    mate response of the equatorial pacific to globalwarming. J. Climate, 22:18, 48734892.Furtado, J., E. Di Lorenzo, N. Schneider, N. Bond, and

    J. Overland, 2011: North Pacific Decadal Variabil-ity and Climate Change in the IPCC AR4 Models.J. Climate, doi:10.1175/2010JCLI3584.1, in press.

    Hori, M. E., and H. Ueda, 2006: Impact of global warm-ing on the East Asian winter monsoon as re-vealed by nine coupled atmosphere-ocean GCMs,Geophys. Res. Lett., 33, L03713, doi:10.1029/2005GL024961.

    Hawkins, E., and R. T. Sutton, 2009: The potential tonarrow uncertainty in regional climate predictions.Bull. Amer. Meteor. Soc., 90, 10951107.

    Hawkins, E., and R. T. Sutton, 2011: The potential tonarrow uncertainty in projections of regional pre-cipitation change. Clim. Dyn., doi:10.1007/s00382-010-0810-6, in press.

    Liu, Z., S. Vavrus, F. Fe, N. Wen, and Y. Zhong, 2005:Rethinking tropical ocean response to globalwarming: The enhanced equatorial warming. J.Climate, 18, 46844700.

    Meehl, G. A., C. Covey, T. Delworth, M. Latif, B. Mc-Avaney, J. F. B. Mitchell, R. J. Stouer, and K. E.Taylor, 2007: The WCRP CMIP3 multimodeldataset: A new era in climate change research. Bull.Amer. Meteor. Soc., 88, 13831394, doi:10.1175/BAMS-88-9-1383.

    Onogi, K., J. Tsutsui, H. Koide, M. Sakamoto, S. Ko-bayashi, H. Hatsushika, T. Matsumoto, N. Yama-zaki, H. Kamahori, K. Takahashi, S. Kadokura,K. Wada, K. Kato, R. Oyama, T. Ose, N. Man-noji, and R. Taira, 2007: The JRA-25 Reanalysis.J. Meteor. Soc. Japan, 85, 3, 369432. doi:10.2151/jmsj.85.369.

    Oshima, K., and Y. Tanimoto, 2009: An evaluation ofreproducibility of the Pacific Decadal Oscillationin the CMIP3 simulations. J. Meteor. Soc. Japan,87, 4, 755770. doi:10.2151/jmsj.87.775.

    Overland, J. E., and M. Wang, 2007: Future climate ofthe North Pacific Ocean, Eos Trans. AGU, 88,178, 182.

    Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Fol-land, L. V. Alexander, D. P. Rowell, E. C. Kent,and A. Kaplan, 2003: Global analyses of SST,sea ice and night marine air temperature sincethe late 19th Century. J. Geophys. Res., 108, 4407,doi:10.1029/2002JD002670.

    Salathe, E. P., Jr. 2006: Influences of a shift in NorthPacific storm tracks on western North Americanprecipitation under global warming. Geophys. Res.Lett., 33, L19820, doi:10.1029/2006GL026882.

    February 2012 K. OSHIMA et al. 395

  • 7/29/2019 90A_2012-A23

    12/12

    Trenberth, K. E., and J. W. Hurrell, 1994: Decadalatmosphere-ocean variations in the Pacific. ClimateDyn., 9, 303319.

    Wang, M., J. E. Overland, and N. A. Bond, 2010: Cli-

    mate projections for selected large marine eco-systems. J. Mar. Syst. 79, 258266, doi:10.1016/j.jmarsys.2008.11.028.

    Xie, S.-P., C. Deser, G. A. Vecchi, J. Ma, H. Teng, andA. T. Wittenberg, 2010: Global warming patternformation: sea surface temperature and rainfall. J.Climate, 23, 966986.

    Xie, S.-P., L.-X. Xu, Q. Liu, and F. Kobashi, 2011: Dy-namical role of mode-water ventilation in decadalvariability in the central subtropical gyre of theNorth Pacific. J. Climate, 24, 12121225.

    Xu, L. X., S.-P. Xie, Q. Liu, and F. Kobashi, 2012: Re-sponse of the North Pacific Subtropical Counter-

    current and its variability to global warming. J.Oceanogr., in press, doi:10.1007/s10872-011-0031-6.

    Yamaguchi, K., and A. Noda, 2006: Global warming

    patterns over the North Pacific: ENSO versus AO.J. Meteor. Soc. Japan, 84, 221241.Yara, Y., K. Oshima, M. Fujii, H. Yamano, Y. Yama-

    naka, and N. Okada, 2012: Projection and un-certainty of the poleward range expansion of coralhabitats in response to sea surface temperaturewarming: A multiple climate model study. Gal-axea, in press.

    Yukimoto, S., and K. Kodera, 2005: Interdecadal ArcticOscillation in twentieth century climate simulationsviewed as internal variability and response to exter-nal forcing. Geophys. Res. Lett., 32. L03707, doi:10.1029/2004GL021870.

    396 Journal of the Meteorological Society of Japan Vol. 90A