Observed freshening and warming of the western Pacific Warm Pool

Abstract

Trends in observed sea surface salinity (SSS) and temperature are analyzed for the tropical Pacific during 1955–2003. Since 1955, the western Pacific Warm Pool has significantly warmed and freshened, whereas SSS has been increasing in the western Coral Sea and part of the subtropical ocean. Waters warmer than 28.5°C warmed on average by 0.29°C, and freshened by 0.34 pss per 50 years. Our study also indicates a significant horizontal extension of the warm and fresh surface waters, an expansion of the warm waters volume, and a notable eastward extension of the SSS fronts located on the equator and under the South Pacific Convergence Zone. Mixed layer depth changes examined along 137°E and 165°E are complex, but suggest an increase in the equatorial barrier layer thickness. Our study also reveals consistency between observed SSS trends and a mean hydrological cycle increase inferred from Clausius–Clapeyron scaling, as predicted under global warming scenarios. Possible implications of these changes for ocean–atmosphere interactions and El Niño events are discussed.

Appendix

In our study, errors in the SSS fields, in the areas covered by waters fresher than fixed thresholds, and in positions of salinity fronts have to be estimated. For that purpose, we made choices to decide whether a SSS value is a “good” estimate of reality, or if we have enough values in a given time series to confidently compute a long-term trend. In this appendix, these different choices, which are necessarily arbitrary, are described. However, sensitivity tests have been performed for each of these choices, and unless mentioned otherwise, they lead to similar results that do not affect significantly our conclusions.

1.1 Data from different instruments and sampling depth issues

The SSS data originate from different instruments, and procedures of SSS measurements changed during the last decades. The question thus arises as how to mix the different data sets without biasing the computation of salinity trends. As detailed in Delcroix et al. (2005), bucket samples and TSG measurements were corrected, taking CTD measurements as a reference, relying on statistics derived from about 1,100 simultaneous observations performed during research cruises. Another question arises as to what a surface salinity measurement is. While bucket samples (obtained chiefly prior to the 1990s) were collected at the surface of the ocean, it was in the wake of ships steaming at 15–25 knots with drafts of 10–15 m. Hence, we believe that the bucket salinity measurements are probably representative of the 0–10 m averaged salinity. Also, TSG measurements (mostly obtained from 1992 on) were derived from seawater collected at hull intakes located at different depths on merchant ships with the same drafts, and steaming at about the same speed as above. Again, we believe that TSG measurements are also representative of the 0–10 m averaged salinity. We thus selected all the 0–10 m salinity data from Nansen casts (mostly obtained prior to the 1980s), CTD instruments, and TAO/TRITON moorings. Strictly speaking, what we called Sea Surface Salinity in this study should actually be referred to as Near Surface Salinity, representative of the 0–10 m averaged salinity. It is worth noting, however, that the vertical gradient of salinity in the upper 10 m is likely negligible for our purposes. While high resolution measurements in the Warm Pool have revealed the possible occurrence of sharp vertical salinity gradients (as much as 0.5 in the upper 5 m during heavy rainfall), these were very localized in space and time (Soloviev and Lukas 1997). Moreover, a recent study, performed at global scale with a maximum of observations, finds that the salinity difference between 1 and 10 m is 90% of the time less than 0.05 (Hénocq et al. 2008). It is thus unlikely that the mix of instruments and 0–10 depth sampling could bias the present trend analysis given our averaging and gridding procedures.

1.2 Data selection criteria

Our optimal interpolation method provides for each geographical point and for each time step a theoretical estimate of the error variance. This error field is a number comprised between 0 and 1, percentage of the variance of the SSS signal. We consider, somewhat arbitrarily, that we have a “good” SSS (x, y, t) value at one time step t and at one position x, y when the associated error (x, y, t) is lower than 0.36, corresponding to an error lower than 60% of the SSS standard deviation.

To compute the mean SSS, we consider as representative the bins (1° longitude by 1° latitude) where good quality SSS data are available during at least 20% of the 1955–2003 period (i.e., during at least 118 months out of 472). For all our analyses on decadal and long-term trends, only the spatial bins (1° longitude by 1° latitude) containing at least 5 “good” observations during at least 7 of the 10 pentads: 1955–1959, 1960–1964, 1965–1969, and so on till 2000–2004, and during at least 6 of the 9 shifted pentads: 1958–1962, 1963–1967 and so on… till 1998–2002) are considered. The resulting geographic mask is shown in Fig. 13, together with the total number of months of SSS data from 1955 to 2003.

Fig. 13

Number of months of SSS observations from 1955 to 2003. The heavy black line delimits the region where the calculations of trends are reliable (see the Appendix)

To compute a linear temporal trend in one spatial bin, and regression on indices, only “good” SSS data are considered.

1.3 Influence of non-uniform sampling

To compute the surface area covered by waters fresher than a fixed threshold, it can be misleading to consider only the “good” SSS values, as the number of data will vary in time and influence our surface computation. Therefore, we computed at first the maximum size attained by this surface area. Then, we computed the number of spatial bins inside that fixed maximum domain that contain at least one “good” SSS value per year. The temporal evolution of these numbers of bins is represented in Fig. 14 for each threshold, as a percentage of the total number of bins contained in the maximum domain. We consider that it is possible to confidently compute a surface covered by waters fresher than a threshold at a time t when at least 70% of the total domain surface is covered by full bins at this time t. Since the beginning of the 1970s, the number of full bins is high and quite constant in the different areas.

Fig. 14

Left panels, from top to bottom: Surface covered at least one month during 1955–2003 by waters fresher than 34.2, 34.4, 34.8 and 35 pss. Right panels: Number of bins with data inside the corresponding surface, expresses as a percentage of the total surface area

We also computed the error on the fresh pool size with the following method. Where the domain is not well sampled, we created two “error SSS” fields. One with the mean SSS value at this point plus one standard deviation, and one with the mean SSS value minus one standard deviation. We computed the fresh pool sizes on these two error SSS fields, and this gives the error bars in Fig. 6. These errors are obviously overestimated, because it is not probable that all errors are of the same sign. Linear regressions are thus computed on the three fields, giving error bars. The non-uniformity of data coverage and sampling is thus not the cause of the increase in the fresh pool size.

To compute the longitudinal or latitudinal position of a salinity front, a similar method is used. We computed the number of spatial bins inside the different domains (4°S–4°N 140°E–170°W; 18°S–8°S, 165°E–155°W) that contain at least one “good” SSS value per year. As a percentage of the total number of bins contained in the maximum domain. We consider that it is possible to confidently compute the position of a front at a time t when at least 70% of the domain is filled with full bins at this time t.

1.4 Subsurface analyses

Finally, Fig. 15 shows the location of salinity/temperature profiles along 165°E and 137°E, as a function of latitude and time. The BLT is defined as the difference DT₀₂−D_sig. DT₀₂ is the depth where the temperature has decreased by 0.2°C as compared to the 10-m depth temperature, and D_sig is the depth where the potential density has increased from the 10-m depth density by a density threshold equivalent to the same temperature change of 0.2°C (Eq. 1 of de Boyer Montégut et al. (2007)). Any difference between DT₀₂ and D_sig will then only be due to salinity stratification. The MLD is defined as the depth where the density is higher than the 10-m depth density by 0.068 kg m⁻³. This threshold corresponds approximately to a decrease of 0.2°C for Warm Pool waters. The computations of MLD and BLT are thus consistent. A maximized error for the mixed layer depth and barrier layer thickness is computed for each profile as a function of the vertical resolution of the profile. Data are interpolated onto a 0.5° latitude regular grid. To take into account these profile-dependent errors, weighted linear regression is computed, with weights inversely proportional to the errors. Temporal variability is important, and a classical computation of the slope error depending on the length of the time series and on its variability is thus performed (Emery and Thomson 2001). We also computed linear trends on data interpolated with different correlation radii values and different smoothing, and this gives error bars on our results.

Fig. 15

Location of temperature/salinity subsurface profiles collected along the 165°E (top) and 137°E (bottom) longitudes, as a function of latitude and time