Skip to main content
Log in

Long-term variations of surface and intermediate waters in the southern Indian Ocean along 32°S

Journal of Oceanography Aims and scope Submit manuscript

Abstract

Variations of water properties in surface and intermediate layers along 32°S in the southern Indian Ocean were examined using a 50-year (1960–2010) time series reproduced from historical hydrographic and Argo data by using optimum interpolation. Salinity in the 26.7–27.3σθ density layer decreased significantly over the whole section, at a maximum rate of 0.02 decade−1 at 26.8–26.9σθ, for the 50-year average. Three deoxygenating cores were identified east of 75°E, and the increasing rate of apparent oxygen utilization in the most prominent core (26.9–27.0σθ) exceeded 0.05 ml l−1 decade−1. The pycnostad core of Subantarctic Mode Water (SAMW) and the salinity minimum of Antarctic Intermediate Water shifted slightly toward the lighter layers. Comparisons with trans-Indian Ocean survey data from 1936 suggest that the tendencies found in the time series began before 1960. Interestingly, cores of many prominent trends were located just offshore of Australia at 26.7–27.0σθ, which is in the SAMW density range. Spectrum analysis revealed that two oscillation components with time scales of about 40 and 10 years were dominant in the subsurface layers. Our results are fairly consistent with, and thus support, the oceanic responses in the southern Indian Ocean to anthropogenic climate change predicted by model studies.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9
Fig. 10
Fig. 11
Fig. 12
Fig. 13
Fig. 14

Notes

  1. In this paper, neutral density (γn) along the 32°S section, used in previous studies, is converted to potential density (σθ) by use of the function of Bindoff and McDougall (2000): σθ = 0.9125 × (γn − 26.5) + 26.4351.

  2. The model freshening drift of 0.0005 year−1 described by Stark et al. (2006) was removed from the reported value of 0.033 decade−1 (0.1 freshening over 30 years).

References

Download references

Acknowledgments

We thank all members of the Ocean Climate Change Research Program, Japan Agency for Marine-Earth Science and Technology, for their comments, especially S. Hosoda, T. Hasegawa, and T. Ohira for their kind advice on our analysis. Two reviewers gave us valuable comments that have improved the manuscript. This study used data from the international Argo Program (http://www.argo.ucsd.edu). Argo is part of the Global Ocean/Global Climate Observing System. Data from Argo profiling floats are freely available.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Taiyo Kobayashi.

Appendices

Appendix A: Reconstruction of zonal sections of the trans-Indian Ocean hydrographic surveys

Zonal sections gridded at intervals of 1° zonally and 0.05σθ diapycnally were reconstructed for each of the trans-Indian Ocean hydrographic surveys (Fig. 1c) because differences in station intervals among the surveys might distort properties averaged over a sector of the section.

The sectional structures of the isopycnal depth, salinity, oxygen, and AOU were reconstructed by use of an OI technique. The layer thickness was calculated later from the reconstructed isopycnal depths. The OI calculation consisted of two steps, in accordance with Roemmich (1983). Large-scale features of the section were estimated at the first OI using a Gaussian correlation function with e-folding scales of 12° zonally and 0.2σθ diapycnally. The signal-to-noise ratio was set to 1 uniformly for all variables. The initial estimate and the signal for the first OI were the average and SD calculated from each trans-Indian Ocean survey data set. In the second step, finer structures were adjusted from the first estimation using values on e-folding scales of 2° and 0.1σθ and a signal-to-noise ratio of 1.5. Some of the property distributions during the surveys are shown in Fig. 15.

Fig. 15
figure 15

Zonal distributions of the trans-Indian Ocean hydrographic survey data (obtained by bottle sampling) for a layer thickness in 1965, b salinity in 1987, and c AOU in 2009. The white contours represent the depths of isopycnal surfaces. The triangles and dots represent the sampling stations and the subsurface measurements, respectively

Appendix B: Possibility of frequency modulation caused by temporal OI operation

The dominant frequencies determined by spectrum analysis, especially that of approximately 0.1 year−1, may be modulated from the true ones by the 3-year temporal correlation of the OI operation. Here, the possibility of the frequency modulation is examined with an OI simulation in which 50-year (100-year) time series are reproduced from virtual data sets of numerous “observations”.

The virtual data sets for the OI simulation were provided as follows. First, we assumed a true time series of salinity at a position for 50 (100) years which included the oscillation components with periods of 40, 10, and 4 years and amplitudes of 0.05, 0.03, and 0.005, respectively (Fig. 16a). The virtual data sets consisted of numerous “observations” at the position and each “observation” had a random Gaussian error with SD of 0.03. From the numerous observations 50-year (100-year) time series were reproduced by the same OI procedure (described in the section “2.2”) except for no spatial correlations and a constant signal-to-noise ratio (1.0).

Fig. 16
figure 16

Results of OI simulations for 50-year data sets. a Solid line and gray dots represent (anomaly of) the true salinity and its 9000 “observations”. Cyan, red, yellow, green, magenta, and blue lines represent the salinity time series reproduced by the OI procedure from 100 (5 cases), 300 (3 cases), 600 (3 cases), 1000 (2 cases), 3000, and 9000 observations, respectively. Vertical error bars at the legend are the typical OI errors obtained from the averages for the period. b Power spectrum densities of the reproduced salinity time series. The vertical bar represents the range of the 95% confidence interval. A time series and a power spectrum density calculated from a data set of 300 observations (black dots in a) are emphasized with red circles

The virtual data sets for 50 years had 100, 300, 600, 1000, 3000, and 9000 observations and those for 100 years had twice as many for the same temporal data density. For the data set of 300 (9000) data for 50 years, 6 (180) data per year were available for the OI estimation on average, the (temporal) data density of which was comparable with that in 1960–1990 (after 2009) around the 32°S section (Fig. 2) considering the spatial correlation scales (5° zonally and 3° meridionally) for the OI estimation.

The time series reproduced objectively from the 50-year data sets are shown in Fig. 16a. As more data were available for the OI estimations, the OI errors and the deviations from the true salinity became smaller in general. Spectrum analysis for the time series clarified an independent peak at an approximate frequency of 0.1 year−1 for all cases (Fig. 16b). Thus, the frequency of the 10-year oscillation was hardly modulated by the OI procedure. Statistically, the peak with the 10-year period seemed more often separated significantly at the 5% significance level from the broad peak including the 40-year oscillation in the more abundant data set. All of 10 calculations for the data sets with 9000 data showed the significantly independent peak, but the peak was not statistically separated from the peak of the longer-period oscillation in 1 case of 10 calculations for 3000 data, 3 of 20 for 1000 data, 9 of 20 for 600 data, 13 of 20 for 300 data, and 16 of 20 for 100 data. This may be because the reproduced amplitude of the 10-year oscillation component was reduced because of the sparseness of data. In the cases for 100-year time series, the independent peak for the 10-year period was reproduced significantly at the 5% significance level from all calculations except for 4 cases of 20 calculations for 200 data in 100 years (not shown). This suggests that time series longer than in this study (50 years) might be required to identify the oscillation of 10-year period significantly from historical (sparse) data sets in statistics.

A smoothing effect because of the temporal correlation in the OI procedure was readily found for frequencies higher than 0.2 year−1, especially for sparser data sets. In the data sets with 3000 and 9000 observations, the reproduced power spectrum densities of the higher frequencies were almost comparable with the true values and the independent peak for the 4-year period was clearly apparent. For the sparser data sets, however, these components were substantially reduced and the oscillation component for the 4-year period was almost screened out.

Rights and permissions

Reprints and permissions

About this article

Cite this article

Kobayashi, T., Mizuno, K. & Suga, T. Long-term variations of surface and intermediate waters in the southern Indian Ocean along 32°S. J Oceanogr 68, 243–265 (2012). https://doi.org/10.1007/s10872-011-0093-5

Download citation

  • Received:

  • Revised:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10872-011-0093-5

Keywords

Navigation