• Bloom, S. C., , L. Takacs, , A. M. da Silva, , and D. Ledvina, 1996: Data assimilation using incremental analysis updates. Mon. Wea. Rev., 124, 12561271.

    • Search Google Scholar
    • Export Citation
  • Choi, B.-H., , D.-H. Kim, , and J.-W. Kim, 2002: Regional responses of climate in the Northwestern Pacific Ocean to gradual global warming for a CO2 quadrupling. J. Meteor. Soc. Japan, 80, 14271442.

    • Search Google Scholar
    • Export Citation
  • Cox, P. M., , and D. Stephenson, 2007: Changing climate for prediction. Science, 317, 207208.

  • Gouretski, V., , and K. P. Koltermann, 2007: How much is the ocean really warming. Geophys. Res. Lett., 34, L01610, doi:10.1029/2006GL027834.

    • Search Google Scholar
    • Export Citation
  • Hansen, J., , M. Sato, , and R. Ruedy, 1997: Radiative forcing and climate response. J. Geophys. Res., 102, 68316864.

  • Hasumi, H., , and S. Emori, Eds., 2004: K-1 coupled GCM (MIROC) description. K-1 Tech. Rep. 1, Center for Climate System Research, University of Tokyo, Tokyo, Japan, 34 pp.

    • Search Google Scholar
    • Export Citation
  • Ishii, M., , and M. Kimoto, 2009: Reevaluation of historical ocean heat content variations with time-varying XBT and MBT depth bias corrections. J. Oceanogr., 65, 287299.

    • Search Google Scholar
    • Export Citation
  • Keenlyside, N. S., , M. Latif, , J. Jungclaus, , L. Kornblueh, , and E. Roeckner, 2008: Advancing decadal–scale climate prediction in the North Atlantic sector. Nature, 45, 8488.

    • Search Google Scholar
    • Export Citation
  • Lean, J., , J. Beer, , and R. Bradley, 1995: Reconstruction of solar irradiance since 1610: Implications for climate change. Geophys. Res. Lett., 22, 31953198.

    • Search Google Scholar
    • Export Citation
  • Lyman, J. M., , S. A. Good, , V. V. Gouretski, , M. Ishii, , G. C. Johnson, , M. D. Palmer, , D. M. Smith, , and J. K. Willis, 2010: Robust warming of the global upper ocean. Nature, 465, 334337, doi:10.1038/nature09043.

    • Search Google Scholar
    • Export Citation
  • Mantua, N. J., , S. R. Hare, , Y. Zhang, , J. M. Wallace, , and R. C. Francis, 1997: A Pacific interdecadal climate oscillation with impacts on salmon production. Bull. Amer. Meteor. Soc., 78, 10691079.

    • Search Google Scholar
    • Export Citation
  • Mochizuki, T., and Coauthors, 2010: Pacific Decadal Oscillation hindcasts relevant to near–term climate prediction. Proc. Natl. Acad. Sci. USA, 107, 18331837.

    • Search Google Scholar
    • Export Citation
  • Nakicenovic, N., and Coauthors, 2000: Emissions Scenarios. Cambridge University Press, 599 pp.

  • Nozawa, T., , T. Nagashima, , H. Shiogama, , and S. A. Crooks, 2005: Detecting natural influence on surface air temperature change in the early twentieth century. Geophys. Res. Lett., 32, L20719, doi:10.1029/2005GL023540.

    • Search Google Scholar
    • Export Citation
  • Pohlmann, H., , J. Jungclaus, , A. Köhl, , D. Stammer, , and J. Marotzke, 2009: Initializing decadal climate predictions with the GECCO oceanic synthesis: Effects on the North Atlantic. J. Climate, 22, 39263938.

    • Search Google Scholar
    • Export Citation
  • Sato, M., , J. E. Hansen, , M. P. McCormick, , and J. B. Pollack, 1993: Stratospheric aerosol optical depths, 1850–1990. J. Geophys. Res., 98, 22 98722 994.

    • Search Google Scholar
    • Export Citation
  • Shiogama, H., , T. Nozawa, , and S. Emori, 2007: Robustness of climate change signals in near term predictions up to the year 2030: Changes in the frequency of temperature extremes. Geophys. Res. Lett., 34, L12714, doi:10.1029/2007GL029318.

    • Search Google Scholar
    • Export Citation
  • Smith, D. M., , S. Cusack, , A. W. Colman, , C. K. Folland, , G. R. Harris, , and J. M. Murphy, 2007: Improved surface temperature prediction for the coming decade from a global climate model. Science, 317, 796799.

    • Search Google Scholar
    • Export Citation
  • Smith, D. M., , R. Eade, , N. J. Dunstone, , D. Fereday, , J. M. Murphy, , H. Pohlmann, , and A. A. Scaife, 2010: Skilful multi-year predictions of Atlantic hurricane frequency. Nat. Geosci., 3, 846849, doi:10.1038/ngeo1004.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    (a) Five-year running mean VAT300 (vertical averaged temperature upper 300 m; °C) of OBSxbt (the bias corrected observations; solid red) and OBSctl (the uncorrected observations; dashed red) averaged over 60°S–60°N. (b) As in (a), but the anomalies are relative to the individual climatological means for 1961–90. An additional time series for NoAS (model simulation with the prescribed anthropogenic and natural forcing factors with no data assimilations; white) is drawn with 1 standard deviation of the 10 ensembles in NoAS (shades).

  • View in gallery

    (a) RMS differences (°C) of the 5-yr mean VAT300 anomaly between OBSxbt and OBSctl for 1961–2000. (b) As in (a), but for the temperature anomaly in the latitude–depth section along the date line. (c) As in (a), but for differences between the ensemble means of ASMxbt and ASMctl. (d) Correlation coefficients between the 5-yr mean OBSxbt minus OBSctl and the corresponding ASMxbt minus ASMctl. Correlation coefficients of 0.62, 0.71, and 0.83 correspond to the 10%, 5%, and 1% significance levels, respectively.

  • View in gallery

    Standard deviation (°C) of the 5-yr mean VAT300 anomaly of OBSxbt for 1961–2000.

  • View in gallery

    RMS errors (°C) of predicted internal variations of VAT300 averaged over the initial 5 yr in (a) HCSTxbt and (b) HCSTctl, normalized by the standard deviation of the observations. The observed internal variations are used as the truth of the predictions. The red contours indicate significant anomaly correlations between the hindcasts and the observations at the 10% level of the t test. (c) Ratio of the RMS errors of HCSTxbt to those of HCSTctl averaged over the initial 5 yr of hindcasts. Solid and dashed red contours show significant improvement in the anomaly correlation of HCSTxbt over that of HCSTctl at the 10% and 25% levels. (d) Ratio of the RMS differences of the 5-yr mean VAT300 during 1961–2000 between the observations and ASMxbt to those of ASMctl.

  • View in gallery

    (a) Spatial pattern of the leading EOF for the internal variation of NoAS VAT300 from 1961 to 2000 in the Pacific. (b) Time series of the 5-yr mean internal VAT300 anomalies (°C) of OBSxbt (red), the ensemble means of HCSTxbt (solid, blue), and the ensemble means of HCSTctl (dashed blue), projected onto the leading EOF (Fig. 5a). The 5-yr running means of hindcasts within 2.5-yr lead are calculated from a mixture of the assimilated and predicted data. At lead-time 0, the plotted value is the average of assimilated data for 2.5 yr before the initial date and predicted data for 2.5 yr after the initial date. The blue shades represent 1 standard deviation of the 10-member ensemble spread in HCSTxbt. (c) Anomaly correlation coefficients between the observed and predicted 5-yr mean PDO variations shown in Fig. 5b. Lead time represents the center of the running mean period. The blue and green curves indicate the correlation coefficients for the hindcasts and persistence predictions (i.e., keeping the initial conditions, which are 5-yr means centered at hindcast start dates, unchanged during the prediction), respectively. The solid curves denote the predictions with the XBT depth bias correction, while the dashed curves are drawn for those without the correction. The dark and light blue shadings represent the 75% and 90% confidence levels of HCSTxbt, respectively, and orange shadings are for HCSTctl. Blue open circles designate significant improvement in the anomaly correlation of HCSTxbt over that of HCSTctl at the 25% level of the t test.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 4 4 2
PDF Downloads 2 2 2

Influence of XBT Temperature Bias on Decadal Climate Prediction with a Coupled Climate Model

View More View Less
  • 1 Atmosphere and Ocean Research Institute, University of Tokyo, Kashiwa, Japan
  • 2 Japan Agency for Marine-Earth Science and Technology, Yokohama, and Meteorological Research Institute, Japan Meteorological Agency, Tsukuba, Japan
  • 3 Atmosphere and Ocean Research Institute, University of Tokyo, Kashiwa, Japan
  • 4 Japan Agency for Marine-Earth Science and Technology, Yokohama, Japan
  • 5 National Institute for Environmental Studies, Tsukuba, Japan
© Get Permissions
Full access

Abstract

The influence of the expendable bathythermograph (XBT) depth bias correction on decadal climate prediction is presented by using a coupled atmosphere–ocean general circulation model called the Model for Interdisciplinary Research on Climate 3 (MIROC3). The global mean subsurface ocean temperatures that were simulated by the model with the prescribed anthropogenic and natural forcing are consistent with bias-corrected observations from the mid-1960s onward, but not with uncorrected observations. The latter is reflected by biases in subsurface ocean temperatures, particularly along thermoclines in the tropics and subtropics. When the correction is not applied to XBT observations, these biases are retained in data assimilation results for the model’s initial conditions. Hindcasting past Pacific decadal oscillations (PDOs) is more successful in the experiment with the bias-corrected observations than that without the correction. Improvement of skill in predicting 5-yr mean vertically average ocean subsurface temperature is also seen in the tropical and the central North Pacific where PDO-related signals appear large.

Corresponding author address: Sayaka Yasunaka, Atmosphere and Ocean Research Institute, University of Tokyo, Kashiwa 277-8568, Japan. E-mail: y_sayaka@aori.u-tokyo.ac.jp

Abstract

The influence of the expendable bathythermograph (XBT) depth bias correction on decadal climate prediction is presented by using a coupled atmosphere–ocean general circulation model called the Model for Interdisciplinary Research on Climate 3 (MIROC3). The global mean subsurface ocean temperatures that were simulated by the model with the prescribed anthropogenic and natural forcing are consistent with bias-corrected observations from the mid-1960s onward, but not with uncorrected observations. The latter is reflected by biases in subsurface ocean temperatures, particularly along thermoclines in the tropics and subtropics. When the correction is not applied to XBT observations, these biases are retained in data assimilation results for the model’s initial conditions. Hindcasting past Pacific decadal oscillations (PDOs) is more successful in the experiment with the bias-corrected observations than that without the correction. Improvement of skill in predicting 5-yr mean vertically average ocean subsurface temperature is also seen in the tropical and the central North Pacific where PDO-related signals appear large.

Corresponding author address: Sayaka Yasunaka, Atmosphere and Ocean Research Institute, University of Tokyo, Kashiwa 277-8568, Japan. E-mail: y_sayaka@aori.u-tokyo.ac.jp

1. Introduction

A decadal climate prediction that covers the period up to 2035 is a major issue to be addressed in the next assessment report of the Intergovernmental Panel on Climate Change (IPCC; Cox and Stephenson 2007). In decadal climate prediction, we need to take into account internal climate variations as well as climatic responses to external forcing factors, such as greenhouse gas concentrations, solar irradiance, and volcanic activity which are presentably not predictable on the decadal time scale. It was only recently that researchers began performing decadal climate prediction experiments with coupled atmosphere–ocean general circulation models (AOGCMs) that focus on internal decadal climate variations, such as the Atlantic meridional overturning circulation and the Pacific decadal oscillation (PDO; e.g., Smith et al. 2007; Keenlyside et al. 2008; Pohlmann et al. 2009; Mochizuki et al. 2010; Smith et al. 2010). For decadal prediction, the oceanic states of the model, rather than the atmospheric states, have to be initialized using observations because of the large heat content and slowly varying dynamics. Therefore, high-quality long-term oceanographic data are needed to initialize the model and validate the decadal climate predictions.

Recent papers point out that significant positive biases in the ocean subsurface temperatures measured by expendable bathythermographs (XBTs) have degraded historical observational datasets since the late 1960s (e.g., Gouretski and Koltermann 2007; Lyman et al. 2010). Ishii and Kimoto (2009) proposed a method to correct the time-varying depth biases in historical XBT observations, which are assumed to be a major source of the temperature biases in question, and showed that the bias correction yields prominent changes in the global mean upper-ocean heat content on not only climatological mean, but also decadal time scales (Fig. 1a). However, the former decadal prediction studies that are listed above have not taken into account these XBT biases. The aim of this study is to examine what impact the XBT bias correction has on decadal prediction.

Fig. 1.
Fig. 1.

(a) Five-year running mean VAT300 (vertical averaged temperature upper 300 m; °C) of OBSxbt (the bias corrected observations; solid red) and OBSctl (the uncorrected observations; dashed red) averaged over 60°S–60°N. (b) As in (a), but the anomalies are relative to the individual climatological means for 1961–90. An additional time series for NoAS (model simulation with the prescribed anthropogenic and natural forcing factors with no data assimilations; white) is drawn with 1 standard deviation of the 10 ensembles in NoAS (shades).

Citation: Journal of Climate 24, 20; 10.1175/2011JCLI4230.1

2. Model experiment

We used the same AOGCM, called the Model for Interdisciplinary Research on Climate 3 (MIROC3; Hasumi and Emori 2004; formerly called MIROC3.2), and the same initialization scheme for the decadal climate prediction as in Mochizuki et al. (2010). The atmospheric component is a T42 spectral model and has 20 levels on a vertical σ coordinate. The resolution of the ocean component is 1.4° in longitude and 0.56°–1.4° in latitude (finer around the equator) and has 44 vertical levels. Using MIROC3, Nozawa et al. (2005) and Shiogama et al. (2007) performed 10-member ensemble simulations. For the period from 1850 to 2000, they used historical data of natural and anthropogenic forcing toward reproducing the twentieth-century climate, which included greenhouse gas concentrations by Hansen et al. (1997), solar irradiance by Lean et al. (1995), and volcanic activity by Sato et al. (1993). The A1B-type emissions scenario in the Special Report on Emissions Scenarios (SRES; Nakicenovic et al. 2000) with the solar constant 1367 W m−2 was used for the period from 2001 to 2030. We define these 10-member ensembles (referred to as NoAS) as a reference field in data assimilation and verification of hindcast results as described below.

A monthly observational gridded dataset of ocean temperature and salinity with a model grid is prepared by an objective analysis in which departures of temperature and salinity from climatology are calculated with the XBT depth bias correction (Ishii and Kimoto 2009; OBSxbt). Here, the XBT depth bias is given by a linear function of the elapsed time from the moment when the XBT probe touches the sea surface to the time when the probe reaches the depth of the temperature observation. To investigate the impact of the XBT depth bias correction on the decadal predictions, another observational dataset is computed with uncorrected XBT data using exactly the same gridding procedure (OBSctl).

To initialize the coupled model for the prediction, we assimilate only observed anomalies relative to the baseline period from 1961 to 1990. Namely, we exclude the climatological mean differences between the model and the observations. This is because correcting model climatology by data assimilation frequently leads to unrealistic climate drifts during subsequent predictions. With an incremental analysis update method (Bloom et al. 1996), the model temperature and salinity in the upper 3000-m depth are forced to approach the gridded observations. Here, the analysis increments, that is, additional heating and water supply, are updated daily and are defined to be 0.025 of differences in temperature and salinity between the model and the gridded observations interpolated daily. It is confirmed that the initialized model reproduces observed temperatures associated with major climate variations such as El Niño and that there does not appear any serious climate drifts during the predictions. We perform 10-member data assimilation runs by using OBSxbt and OBSctl beginning from the 10 initial conditions of NoAS on 1 January 1945 (ASMxbt and ASMctl, respectively). Then, we conduct 10 sets of 10-yr-long, 10-member ensemble hindcast experiments by using the ASMxbt and the ASMctl snapshots as initial conditions every 5 yr from 1 January 1961 to 1 January 2006 (HCSTxbt and HCSTctl, respectively).

3. Impact of the XBT depth bias correction

Figure 1b shows the global mean vertically averaged temperature anomalies from the sea surface to a 300-m depth (VAT300) relative to the climatological means for the period 1961–90. The VAT300 anomalies of OBSxbt are lower in the 1970s and higher after the 1980s in comparison to those of OBSctl, and the former agrees better with NoAS than with the latter.

Figure 2a shows the spatial distribution of the root-mean-squared (RMS) differences of the 5-yr running mean VAT300 anomalies between OBSxbt and OBSctl, and Fig. 2b shows the temperature anomalies in the latitude–depth section along the date line. The VAT300 in the tropics and the subtropics are affected by the correction (Fig. 2a), because the depth biases lead to significant temperature differences through the large vertical temperature gradients there (Fig. 2b). The RMS differences in the VAT300 are comparable with 30%–50% of the standard deviation in the 5-yr running mean OBSxbt anomalies (Fig. 3), and temporal variations of the differences between OBSxbt and OBSctl are coherent with the global mean differences (see Fig. 1b).

Fig. 2.
Fig. 2.

(a) RMS differences (°C) of the 5-yr mean VAT300 anomaly between OBSxbt and OBSctl for 1961–2000. (b) As in (a), but for the temperature anomaly in the latitude–depth section along the date line. (c) As in (a), but for differences between the ensemble means of ASMxbt and ASMctl. (d) Correlation coefficients between the 5-yr mean OBSxbt minus OBSctl and the corresponding ASMxbt minus ASMctl. Correlation coefficients of 0.62, 0.71, and 0.83 correspond to the 10%, 5%, and 1% significance levels, respectively.

Citation: Journal of Climate 24, 20; 10.1175/2011JCLI4230.1

Fig. 3.
Fig. 3.

Standard deviation (°C) of the 5-yr mean VAT300 anomaly of OBSxbt for 1961–2000.

Citation: Journal of Climate 24, 20; 10.1175/2011JCLI4230.1

The differences in the observations that are caused by the bias correction (Fig. 2a) are retained in the data assimilation results at the low latitudes (Fig. 2c), although the RMS differences along the thermocline become smaller by 50% in the data assimilation than in the observations. The reason for the reduction of the differences is because the artificial decadal changes originated from the XBT bias are not acceptable in ASMctl as discussed in section 4. Moreover, the data assimilation was conducted not so strong as to avoid climate drifts during the predictions. The temporal correlation between the 5-yr mean OBSxbt minus OBSctl and the corresponding ASMxbt minus ASMctl is also significantly high in the tropics and the subtropics (Fig. 2d).

Next, we show the impact of the XBT bias correction on the prediction of the internal decadal climate variations in the Pacific Ocean, verifying HCSTxbt and HCSTctl with OBSxbt and OBSctl, respectively. To discuss the internal components of the climate change below, the ensemble means of NoAS, which approximate the response to external forcing, are removed from the observed, assimilated, and predicted temperatures. This procedure is similar to what Mochizuki et al. (2010) adopted. Figures 4a,b show the RMS errors of the 5-yr mean HCSTxbt and HCSTctl VAT300 internal components normalized by the standard deviations of the observed internal variations (not shown, but quite similar to Fig. 3). Both of the hindcast experiments produce the 5-yr mean VAT300 with smaller RMS errors than the observed standard deviations widely in the Pacific Ocean. In these areas, anomaly correlations between the hindcasts and the observations are significant at 10% shown by red contours in Figs. 4a,b. Figure 4c shows the ratio of the RMS errors of HCSTxbt to those of HCSTctl and improvement of anomaly correlation by the bias correction. In HCSTxbt, the gridwise RMS errors averaged over the whole Pacific are 6% smaller than those in HCSTctl, and more than 20% reductions in the errors due to the bias correction appear in the central to eastern tropics and subtropics, and the central North Pacific. The anomaly correlation is also improved in areas with the large RMSE reductions, although the significant level of the anomaly correlation improvement is limited to about 25%. The improvement of the hindcast skill takes a pattern like the PDO, which is discussed later.

Fig. 4.
Fig. 4.

RMS errors (°C) of predicted internal variations of VAT300 averaged over the initial 5 yr in (a) HCSTxbt and (b) HCSTctl, normalized by the standard deviation of the observations. The observed internal variations are used as the truth of the predictions. The red contours indicate significant anomaly correlations between the hindcasts and the observations at the 10% level of the t test. (c) Ratio of the RMS errors of HCSTxbt to those of HCSTctl averaged over the initial 5 yr of hindcasts. Solid and dashed red contours show significant improvement in the anomaly correlation of HCSTxbt over that of HCSTctl at the 10% and 25% levels. (d) Ratio of the RMS differences of the 5-yr mean VAT300 during 1961–2000 between the observations and ASMxbt to those of ASMctl.

Citation: Journal of Climate 24, 20; 10.1175/2011JCLI4230.1

One of the major decadal climate variations in the Pacific is the PDO (e.g., Mantua et al. 1997). The PDO in the present study is defined by the leading empirical orthogonal function (EOF) of the internal variations of the 10-member NoAS VAT300 from 1961 to 2000 over the Pacific (30°S–60°N, 120°E–60°W). The EOF is in good agreement with the observed counterpart; the sign is positive in the central to eastern tropics and the Gulf of Alaska, whereas it is negative in the midlatitude regions of both hemispheres, which extend from the western tropics (Fig. 5a). Because the model has a large bias in the Kuroshio–Oyashio extension due to the northward shift of the Kuroshio separation latitude (e.g., Choi et al. 2002), we focus on the signal only in the tropics and the subtropics. In these regions, the XBT depth bias correction also has a large impact on the data assimilation and on subsequent predictions (Figs. 2 and 4). Therefore, the projection of VAT300 onto the leading EOF is made limitedly for latitudes from 30°S to 30°N of the original domain. Time series of the PDO projected as above are shown in Fig. 5b. The time series for the observations display decadal variations similar to the so-called PDO index (Mantua et al. 1997), which represent, for example, a prominent phase shift in the mid-1970s and a short-term cold phase around 2000. Time series of the PDO computed from OBSctl is quite similar to that of OBSxbt (not shown). This is because the PDO has an east–west seesaw pattern on the tropics and the subtropics, while the bias has an effect of the same sign in the whole domain.

Fig. 5.
Fig. 5.

(a) Spatial pattern of the leading EOF for the internal variation of NoAS VAT300 from 1961 to 2000 in the Pacific. (b) Time series of the 5-yr mean internal VAT300 anomalies (°C) of OBSxbt (red), the ensemble means of HCSTxbt (solid, blue), and the ensemble means of HCSTctl (dashed blue), projected onto the leading EOF (Fig. 5a). The 5-yr running means of hindcasts within 2.5-yr lead are calculated from a mixture of the assimilated and predicted data. At lead-time 0, the plotted value is the average of assimilated data for 2.5 yr before the initial date and predicted data for 2.5 yr after the initial date. The blue shades represent 1 standard deviation of the 10-member ensemble spread in HCSTxbt. (c) Anomaly correlation coefficients between the observed and predicted 5-yr mean PDO variations shown in Fig. 5b. Lead time represents the center of the running mean period. The blue and green curves indicate the correlation coefficients for the hindcasts and persistence predictions (i.e., keeping the initial conditions, which are 5-yr means centered at hindcast start dates, unchanged during the prediction), respectively. The solid curves denote the predictions with the XBT depth bias correction, while the dashed curves are drawn for those without the correction. The dark and light blue shadings represent the 75% and 90% confidence levels of HCSTxbt, respectively, and orange shadings are for HCSTctl. Blue open circles designate significant improvement in the anomaly correlation of HCSTxbt over that of HCSTctl at the 25% level of the t test.

Citation: Journal of Climate 24, 20; 10.1175/2011JCLI4230.1

Both HCSTxbt and HCSTctl follow the observational tendencies for the first several years of lead time, except for the cases starting from 1971 and 1991 (Fig. 5b). Figure 5c shows the anomaly correlations of the 5-yr mean internal PDO variations. The anomaly correlations for both hindcasts are significantly greater than zero, and also superior to persistence predictions. Namely, the PDOs appear predictable for several years ahead, as already reported by Mochizuki et al. (2010). In addition, the introduction of the XBT depth bias correction improves the PDO prediction. In HCSTxbt, the PDOs are predictable for about 9 yr at 90% confidence levels, whereas the confidence levels for HCSTctl reduce at 75%. The prediction skill of HCSTxbt is significantly higher than that of HCSTctl for about 5 yr of lead time from the second to the sixth years as indicated by blue circles in Fig. 5c. The correlation coefficients of both the hindcast and the persistence prediction present local maxima around the seventh year. These peaks might be related to the periodicity of the PDO, but this is beyond the scope of this paper. There are discontinuities at the 4.5-yr of lead time because the hindcast that starts in 2001 is examined only for 4.5 yr.

4. Discussion and conclusions

This study showed the impact of the XBT depth bias correction, which was introduced by Ishii and Kimoto (2009), on decadal climate prediction. The bias correction yielded prominent differences in the assimilated ocean temperatures along the thermocline in the tropical and subtropical Pacific. In addition, the predicted PDOs were in better agreement with the observations than those without the bias correction. The skill in predicting ocean subsurface temperature is improved over areas where temperature variations related to the PDO appear large, although the significance of grid-wise skill is limited.

The time series of the global mean VAT300 anomalies that were simulated by the AOGCM with the prescribed anthropogenic and natural forcing are more consistent with the bias-corrected observations (OBSxbt) than with uncorrected observations (OBSctl; Fig. 1b). It means the XBT biases raise artificial decadal variations. The data assimilation with the correction (ASMxbt) also agrees better with the observations than that without the correction (ASMctl; Fig. 4d). Although the differences in the observations due to the bias correction are large restrictedly in the low latitudes (Fig. 2a), the use of model dynamics appears to improve the thermal fields almost in the whole Pacific (Fig. 4d). Analysis increments of ASMxbt are also smaller than those of ASMctl (not shown). The small RMS differences and the small analysis increments in ASMxbt imply that the initialized model potentially has high prediction skill because the prediction errors can be expectedly small at the beginning of the prediction.

Is it found that observational data with high reliability improved the skill of the decadal climate prediction relevant to the PDO. Although we focused on differences in the observed anomalies with and without the XBT depth bias correction here, the differences on a climatological time scale, which are not treated in the data assimilation of this study, are also prominent (see Fig. 1a). Assimilating the full range of observed temperature values rather than the anomalies would be affected by the XBT bias more seriously. Efforts should continuously be made not only to develop the initialization techniques and the model components, but also to better arrange the historical observational data to improve the prediction skill further.

Acknowledgments

This work was supported by the Ministry of Education, Culture, Sports, Science and Technology, Japan, through the Innovative Program of Climate Change Projection for the 21st Century. S. Yasunaka was financially supported by the Japan Society for the Promotion of Science (JSPS) as a research fellow. The Earth Simulator at JAMSTEC and an NEC SX-8R at NIES were employed to perform the AOGCM simulations.

REFERENCES

  • Bloom, S. C., , L. Takacs, , A. M. da Silva, , and D. Ledvina, 1996: Data assimilation using incremental analysis updates. Mon. Wea. Rev., 124, 12561271.

    • Search Google Scholar
    • Export Citation
  • Choi, B.-H., , D.-H. Kim, , and J.-W. Kim, 2002: Regional responses of climate in the Northwestern Pacific Ocean to gradual global warming for a CO2 quadrupling. J. Meteor. Soc. Japan, 80, 14271442.

    • Search Google Scholar
    • Export Citation
  • Cox, P. M., , and D. Stephenson, 2007: Changing climate for prediction. Science, 317, 207208.

  • Gouretski, V., , and K. P. Koltermann, 2007: How much is the ocean really warming. Geophys. Res. Lett., 34, L01610, doi:10.1029/2006GL027834.

    • Search Google Scholar
    • Export Citation
  • Hansen, J., , M. Sato, , and R. Ruedy, 1997: Radiative forcing and climate response. J. Geophys. Res., 102, 68316864.

  • Hasumi, H., , and S. Emori, Eds., 2004: K-1 coupled GCM (MIROC) description. K-1 Tech. Rep. 1, Center for Climate System Research, University of Tokyo, Tokyo, Japan, 34 pp.

    • Search Google Scholar
    • Export Citation
  • Ishii, M., , and M. Kimoto, 2009: Reevaluation of historical ocean heat content variations with time-varying XBT and MBT depth bias corrections. J. Oceanogr., 65, 287299.

    • Search Google Scholar
    • Export Citation
  • Keenlyside, N. S., , M. Latif, , J. Jungclaus, , L. Kornblueh, , and E. Roeckner, 2008: Advancing decadal–scale climate prediction in the North Atlantic sector. Nature, 45, 8488.

    • Search Google Scholar
    • Export Citation
  • Lean, J., , J. Beer, , and R. Bradley, 1995: Reconstruction of solar irradiance since 1610: Implications for climate change. Geophys. Res. Lett., 22, 31953198.

    • Search Google Scholar
    • Export Citation
  • Lyman, J. M., , S. A. Good, , V. V. Gouretski, , M. Ishii, , G. C. Johnson, , M. D. Palmer, , D. M. Smith, , and J. K. Willis, 2010: Robust warming of the global upper ocean. Nature, 465, 334337, doi:10.1038/nature09043.

    • Search Google Scholar
    • Export Citation
  • Mantua, N. J., , S. R. Hare, , Y. Zhang, , J. M. Wallace, , and R. C. Francis, 1997: A Pacific interdecadal climate oscillation with impacts on salmon production. Bull. Amer. Meteor. Soc., 78, 10691079.

    • Search Google Scholar
    • Export Citation
  • Mochizuki, T., and Coauthors, 2010: Pacific Decadal Oscillation hindcasts relevant to near–term climate prediction. Proc. Natl. Acad. Sci. USA, 107, 18331837.

    • Search Google Scholar
    • Export Citation
  • Nakicenovic, N., and Coauthors, 2000: Emissions Scenarios. Cambridge University Press, 599 pp.

  • Nozawa, T., , T. Nagashima, , H. Shiogama, , and S. A. Crooks, 2005: Detecting natural influence on surface air temperature change in the early twentieth century. Geophys. Res. Lett., 32, L20719, doi:10.1029/2005GL023540.

    • Search Google Scholar
    • Export Citation
  • Pohlmann, H., , J. Jungclaus, , A. Köhl, , D. Stammer, , and J. Marotzke, 2009: Initializing decadal climate predictions with the GECCO oceanic synthesis: Effects on the North Atlantic. J. Climate, 22, 39263938.

    • Search Google Scholar
    • Export Citation
  • Sato, M., , J. E. Hansen, , M. P. McCormick, , and J. B. Pollack, 1993: Stratospheric aerosol optical depths, 1850–1990. J. Geophys. Res., 98, 22 98722 994.

    • Search Google Scholar
    • Export Citation
  • Shiogama, H., , T. Nozawa, , and S. Emori, 2007: Robustness of climate change signals in near term predictions up to the year 2030: Changes in the frequency of temperature extremes. Geophys. Res. Lett., 34, L12714, doi:10.1029/2007GL029318.

    • Search Google Scholar
    • Export Citation
  • Smith, D. M., , S. Cusack, , A. W. Colman, , C. K. Folland, , G. R. Harris, , and J. M. Murphy, 2007: Improved surface temperature prediction for the coming decade from a global climate model. Science, 317, 796799.

    • Search Google Scholar
    • Export Citation
  • Smith, D. M., , R. Eade, , N. J. Dunstone, , D. Fereday, , J. M. Murphy, , H. Pohlmann, , and A. A. Scaife, 2010: Skilful multi-year predictions of Atlantic hurricane frequency. Nat. Geosci., 3, 846849, doi:10.1038/ngeo1004.

    • Search Google Scholar
    • Export Citation
Save