Skip to content

A Note on 20th Century Equatorial Pacific Sea Surface Temperatures

T.R. Knutson1, A. Kaplan2, and N. A. Rayner3

1Geophysical Fluid Dynamics Laboratory/NOAA, Princeton, New Jersey

2 Lamont-Doherty Earth Observatory, Palisades, New York

3 Hadley Centre for Climate Prediction and Research, Meteorological Office, Bracknell, Berkshire, United Kingdom

August 26, 1999

1. Introduction

In a recent study, Knutson and Manabe (1998; hereafter KM98) have presented time series of 10-year running mean NINO3.4 SST anomalies from two different reconstructed data sets, along with a Darwin sea level pressure series (see their Fig. 9). NINO3.4 is an index of equatorial Pacific SSTs, averaged over the region 5°N-5°S, 170°W-120°W. The two SST reconstructions included the LDEO (Lamont-Doherty Earth Observatory) data set (Kaplan et al. 1998) and the GISST2.3 version of the UKMO (United Kingdom Meteorological Office) Global Sea-Ice and Sea Surface Temperature data set (Rayner et al. 1996). The purpose of this note is to examine SSTs from two more recent versions of the UKMO series of reconstructions: GISST2.3b and HadISST1 (Rayner, 1999). The GISST2.3 version used in KM98 was not widely distributed in the climate community and is no longer recommended for use, given the availability of the updated versions such as GISST2.3b and HadISST1. The results presented here will show that the early 20th century NINO3.4 anomalies in the more recent UKMO data sets are much closer to the LDEO reconstruction than those for version GISST2.3 used in KM98.

Comparing methodologies of the LDEO and more recent UKMO data sets, the GISST2.3b was analyzed using simple EOF projection, whereas HadISST1 uses a reduced space optimal interpolation which takes into account expected random data errors and is similar to the LDEO method, but without temporal optimal smoothing. In HadISST1, unlike in LDEO, separate account was taken of the leading global SST EOF, which encapsulates long-term trends, to avoid bias towards climatology in data-sparse areas. In addition, reconstructed fields in HadISST1 were augmented by in situ SST superobs which had been subjected to extra quality controls to reduce the influence of sampling and measurement errors.

2. Results

Ten-year running mean NINO3.4 anomalies (relative to 1961-90) derived from the LDEO, GISST2.3, GISST2.3b, and HadISST1 data sets are shown in Fig. 1. The differences in the early 20th century between decadal means in the LDEO and either the GISST2.3b or HadISST1 data sets are on the order of half or less of the magnitude of the differences between LDEO and the earlier UKMO version (GISST2.3). The GISST2.3b curve lies below the LDEO curve for all periods prior to about 1950, whereas the HadISST1 curve lies below the LDEO mainly prior to about 1920. For values centered in the 1920s, the HadISST1 curve is notably closer to the LDEO than to the earlier UKMO versions (GISST2.3 or GISST2.3b). The linear trends and 95% confidence limits obtained using ordinary least squares regression on the monthly anomalies for the common period 1900-1991 are -0.23 +/- 0.17 °C 100yr-1 for LDEO, +0.08 +/- 0.17 °C 100yr-1 for GISST2.3b, and -0.12 +/- 0.16 °C 100yr-1 for HadISST1. Thus the NINO3.4 trends in these three data sets are not statistically distinguishable (with overlapping 95% confidence limits) for this period according to this test. Although not shown here, a 10-yr running mean anomaly curve for an even earlier version of GISST (GISST2.2) is similar to that for GISST2.3b in terms of long term trends, but with the GISST2.2 exhibiting somewhat smaller variability of decadal means prior to 1950.

Another important aspect of 20th century and future climate change to be considered is the regional structure of SST trends in the tropical Pacific. A future change in which the long-term-mean SST climate became more “El Nino-like” (with a local warming maximum in the region of the equatorial Pacific with the strongest interannual variability) or “La Nina-like” (with a local minimum in warming or a local cooling in the region) could have significant societal implications. Most global climate models to date forced by greenhouse gases tend to show the climate evolving toward a more El Niño-like mean state (Meehl and Washington 1996; KM98; Timmermann et al. 1999). On the other hand, Clement et al. (1996) and Seager and Murtugudde (1997) have proposed that the equatorial Pacific may become more La Niña-like owing to ocean dynamical effects not adequately resolved by the global models. Seager and Murtugudde’s analysis was based upon the simulated response to a uniform surface heating of a regional Pacific ocean GCM coupled to an atmospheric mixed layer model.

Examining the spatial structure of trends in reconstructed SSTs over the 20th century, Cane et al. (1997) noted that the LDEO data for 1900-1991 exhibits a cooling trend in the central equatorial Pacific (i.e., suggesting a trend toward a more La Nina-like mean state, in line with the enhancement of the SST zonal gradient). KM98 presented a spatial map of normalized SST trends (1900-1994) over the Pacific basin based on the GISST2.3 version (their Fig. 4a) which showed a fairly pronounced warming over much of the equatorial Pacific region, particularly around 115°-140°W, although with a relative minimum warming or “saddle” structure near 180°-160°W extending toward slightly negative values off-equator. A map of trends for the period 1900-1994 using the GISST2.3b data shows some notable differences with the features in Fig. 4a of KM98, including a broad local minimum (though still positive values) in the vicinity of the central equatorial Pacific. The GISST2.3b trend map for the same period as the Cane et al. analysis (1900-1991) even shows some areas of slightly negative trends in the central equatorial Pacific. Thus both the GISST2.3b and LDEO data indicate that during the 20th century the east-west equatorial SST gradient between the NINO3.4 region and the warm pool region west of the Dateline has had an increasing trend. This result and the analysis of additional SST data sets in Cane et al. (1997) suggest that the positive trend in the east-west equatorial Pacific gradient during the 20th century may be a more robust cross-analysis feature of various SST data sets than the SST trends for individual regions. Comparison maps for GISST2.3b and LDEO are not presented here, since a comparison figure showing maps of 20th century trends for these data sets, along with a more detailed discussion of several SST reconstructions, is being presented elsewhere (Hurrell and Trenberth, 1999).


We conclude that with regards to equatorial Pacific SST reconstructions for the first half of the 20th century, the GISST2.3b and HadISST1 data reconstructions are much closer to the LDEO reconstruction than was the GISST2.3 data presented in KM98. Although differences in decadal means are still readily apparent (Fig. 1) between the LDEO and the more recent UKMO reconstructions (e.g., GISST2.3b, HadISST1), the trends over the period 1900-1991 in NINO3.4 series from these three data sets are not statistically distinguishable. It is not unexpected that HadISST1 is closer to LDEO than are GISST2.3 and 2.3b, due to the similarity in the methods of analysis of the HadISST1 and LDEO data sets. Efforts continue in the analysis, comparison, and refinement of SST reconstructions such as those discussed here and in attempts to reconcile historical SST observations and reconstructions with related indicators of climate variability (e.g., the Southern Oscillation Index) and with climate model simulations.

Acknowledgments. The development of GISST and HadISST was supported by the UK Public Meteorological Service Research and Development Contract and by the UK Department of the Environment, Transport and the Regions contract PECD/7/12/37. We thank our colleagues at GFDL, LDEO, and the Hadley Centre for comments on a draft version of this report.


  • Cane, M.A., A.C. Clement, A. Kaplan, Y. Kushnir, R. Murtugudde, D. Pozdnyakov, R. Seager, and S.E. Zebiak, 1997: 20th century sea surface temperature trends. Science, 275, 957-960.
  • Clement, A.C., R. Seager, M.A. Cane, and S.E. Zebiak, 1996: An ocean dynamical thermostat. J. Climate, 10, 2190-2196.
  • Hurrell, J.W., and K.E. Trenberth, 1999: Global sea surface temperature analyses: multiple problems and their implications for climate analysis, modeling, and reanalysis. Bull. Amer. Meteor. Soc., 80, 2661-2678.
  • Kaplan, A., M. A. Cane, Y. Kushnir, A.Clement, M. Blumenthal, and B. Rajagopalan, 1998: Analyses of global sea surface temperature 1856-1991. J. Geophys. Res., 103, 18,567-18,589.
  • Knutson, T.R. and S. Manabe, 1998: Model assessment of decadal variability and trends in the tropical Pacific Ocean. J. Climate, 11, 2273-2296.
  • Meehl, G.A., and W.M. Washington, 1996: El Nino-like climate change in a model with increased atmospheric CO2 concentrations. Nature, 382, 56-60.
  • Rayner, N.A., E.B. Horton, D.E. Parker, C.K. Folland, and R.B. Hackett, 1996: Version 2.2 of the global sea-ice and sea surface temperature data set, 1903-1994. Clim. Res. Tech. Note 74. Unpublished document available from the Hadley Centre for Climate Prediction and Research, Meteorological Office, London Road, Bracknell, RS12 2SY, U.K.
  • Rayner, N.A., 1999: SST and sea-ice fields for ERA40. (Paper in preparation.)
  • Seager, R. and R. Murtugudde, 1997: Ocean dynamics, thermocline adjustment and regulation of tropical SST. J. Climate, 10, 521-539.
  • Smith, T.M., R.W. Reynolds, R.E. Livezey, and D.C. Stokes, 1996: Reconstruction of historical sea surface temperatures using empirical orthogonal functions. J. Climate, 9, 1403-1420.
  • Timmermann, A., J. Oberhuber, A. Bacher, M. Esch, M. Latif, and E. Roeckner, 1999: Increased El Nino frequency in a climate model forced by future greenhouse warming. Nature, 398, 694-697.