Interannual variability in the Indian Ocean using altimeter and IX1-expendable bathy-thermograph (XBT) data: Does the 18-month signal exist?

Authors

  • Irina V. Sakova,

    1. CSIRO Marine and Atmospheric Research, Hobart, Tasmania, Australia
    2. Also at School of Geography and Environmental Studies, University of Tasmania, Hobart, Tasmania, Australia.
    Search for more papers by this author
  • Gary Meyers,

    1. CSIRO Marine and Atmospheric Research, Hobart, Tasmania, Australia
    Search for more papers by this author
  • Richard Coleman

    1. CSIRO Marine and Atmospheric Research, Hobart, Tasmania, Australia
    2. Also at School of Geography and Environmental Studies, University of Tasmania, Hobart, Tasmania, Australia.
    3. Also at Antarctic Climate and Ecosystems CRC, Hobart, Tasmania, Australia.
    Search for more papers by this author

Abstract

[1] The dominant frequency bands in altimeter sea surface height (SSH) variability over the whole Indian Ocean and in temperature profiles near the Sumatra-Java coast are identified and analysed using spectral analysis techniques. We find that in most regions of the Indian Ocean, the low-frequency part of the SSH spectra (corresponding to signals with periods from six months to six years) is concentrated in five frequency bands separated by substantial spectral gaps: semi-annual, annual, 18–20 months, 3 years, and 4–6 years. The existence of semi-annual, annual, 2–3-year, and 4–6-year periodical signals is well known; however, the 18–20-month signal has not previously been described. Further investigation of temporal and spatial characteristics of this later signal point to its relationship with the Indian Ocean dipole events: the signal is particularly strong between 1994 and 2000; it develops near the Sumatra coast and propagates to the Bay of Bengal and into the Indonesian Throughflow.

1. Introduction

[2] The existence of annual and semi-annual processes in the Indian Ocean (IO) is well known; they have been the subject of numerous studies [e.g., Clarke and Liu, 1993]. In the last decade there has also been a growing interest in studying the interannual variability in the tropical IO [e.g., Perigaud and Delecluse, 1993; Masumoto and Meyers, 1998], particularly after the discovery of a basin-wide anomalous pattern of sea surface temperature (SST) in the equatorial IO — the Indian Ocean Dipole (IOD), or Indian Ocean Zonal Mode [Saji et al., 1999; Webster et al., 1999].

[3] Analysis of anomalous SST conditions over the IO has produced several theories, such as that the “dipole mode” is an independent mode not associated with the El-Nino Southern oscillation (ENSO) [Saji et al., 1999; Webster et al., 1999; Iizuka et al., 2000]; that the mode is part of an ENSO cycle [Baquero-Bernal et al., 2002]; that it can be connected with a southern tropical IO variability [Behera et al., 2000]; or that it is an integral part of the tropospheric biennial oscillation [Meehl and Arblaster, 2002].

[4] Several studies have discussed the timescale of interannual SST and subsurface variability in the IO, but most of them concentrated on processes with periods of two years and longer. The only work that contains a clear indication of a signal with a period between 1 and 2 years is that of McClean et al. [2005]. They analysed the repeat IX1-expendable bathy-thermograph (XBT) data and Parallel Ocean Program model output and calculated spectra of time series of dynamic height anomaly at the northern and southern ends of the XBT line, for both XBT data and model output. Although these spectra clearly show large peaks at 18–20 months for the northern end of the XBT line [see McClean et al., 2005, Figure 5], the authors did not pay attention to this signal because their “focus was on long-period signals”. They conclude that the long-period signal contains two distinct components and “refer to the 2.7 year signal as quasi-biennial and 5.5 year as quasi-pentadal”.

[5] Most of the early studies of IOD were focused on using SST and wind data. Tourre and White [2003] argued that the interannual variability of SST (with periods from 2 to 20 years) over the IO represents the global climate signal that includes the quasi-biennial oscillation (2.1–2.8-year period); the ENSO oscillation (3–7-year period); and the quasi-decadal oscillation.

[6] Analyses of SST were followed by studies of the subsurface interannual variability in the IO [Rao et al., 2002; Vinayachandran et al., 2002; Feng and Meyers, 2003; Loschnigg et al., 2003]. Rao et al. [2002] found that the subsurface temperature consistently has a dipole structure that does not always appear in SST. Using empirical orthogonal functions for analysis of sea surface height (SSH) satellite altimeter data for the period 1993–1999, as well as ACOM 2.0 ocean model output, two dominant modes of the interannual subsurface variability of tropical IO were found. The first mode appears to be governed by the IOD, while the second mode “shows the interesting quasi-biennial tendency” [Rao et al., 2002].

[7] Feng and Meyers [2003] found a two-year time-scale of the upper-ocean evolution in the tropical IO near the Java coast that is unique to the IOD (that is, not directly forced by an atmospheric teleconnection from the Pacific). Their conclusion was based on the timing of the temperature anomalies associated with the strength of upwelling offshore of Java, when “cold anomaly during 1994 and 1997 [was] followed by warm anomaly during 1995/96 and 1998 [see Feng and Meyers, 2003, p. 2268]” and “may be related to biennial oscillation described by Meehl and Arblaster [2002]”. In another work, Loschnigg et al. [2003] found that the ocean dynamics in the tropical IO contributes to the SST anomalies and constitutes an integral part of the tropospheric biennial oscillation (with time scales of 2–3 years) and the monsoon system.

[8] Apart from a possible connection of the interannual processes in the IO with ENSO and long-term atmospheric processes, several authors have studied the role of equatorial and off-equatorial long planetary waves [Masumoto and Meyers, 1998; Rao et al., 2002]. Rao et al. [2002] suggested that the observed quasi-biennial signals (with periodicity of 2–3 years) in the southern tropical IO are associated with Rossby wave propagation and that it dominates over quasi-pentadal (3.5–6 years) signals in some locations.

[9] Overall, the previous studies of the interannual variability of the IO have considered only signals with time scales of two years and longer; however, Sakova et al. [2005] found the existence of a shorter, repetitive signal during the period of XBT and altimeter data collection.

[10] In the present paper we further analyse the interannual variability of the IO for each of the main frequency bands by applying spectral and wavelet analysis methods to satellite altimeter SSH and XBT temperature data.

2. Data and Methods

[11] Two independent data sets are used in this study: gridded (1 degree) SSH weekly-averaged data for the period from October 1992 to August 2004 collected by the ERS/Envisat/TOPEX/Jason-1 satellites (http://www.jason.oceanobs.com/html/donnees/produits/msla_uk.html) and XBT temperature data for the period from January 1989 to December 2002 (http://www.marine.csiro.au/∼pigot/REPORT/overview.html) [Meyers and Pigot, 1999].

[12] The subsurface temperature data investigated below are located in the Java upwelling region and obtained by averaging XBT data in a rectangle with opposite corners at 7.0°S, 104.0°E and 8.0°S, 106.0°E and interpolating to a monthly grid. The resulting time series contain 168 monthly entries. These monthly time series are analyzed by wavelet (Morlet) and Discrete Fourier Transform methods [Emery and Thomson, 1997; Ghil et al., 2002]. Wavelet software was provided by C. Torrence and G. Compo (http://paos.colorado.edu/research/wavelets).

[13] For analysis of two-dimensional SSH fields in the IO, Discrete Fourier Transform Fourier Transform methods are applied to the time series of SSH at each grid point. The resulting gridded spectra are then either presented as power spectral density (PSD) maps or used to select a particular frequency band and conduct Inverse Discrete Fourier Transform to investigate the temporal dynamics of the corresponding signal. Each time series contains 618 weekly entries.

3. Analysis of XBT Data

[14] Figure 1a shows a typical temperature profile near the Sumatra-Java coast. One can see a number of strong upward displacements of the thermocline associated with upwelling that result in the cold water appearing at the sea surface, most noticeably during the 1994 and 1997 IOD events, and possibly also in 1991. There are also a number of more frequent minor displacements that can be seen from the shape of the 20°C (D20) isothermal surface shown by the yellow line. It is well known that the thermocline in this region heaves with a periodicity of two times a year, which is related to the development of the Wyrtki jet [Clarke and Liu, 1993], and these minor displacements can possibly be associated with this semi-annual process. Discrete Fourier Transform methods permit a clear description of the time-scales of this signal.

Figure 1.

(a) XBT temperature data at 6–7°S, 104–106°E, where the yellow line represents the depth of the 20°C isotherm; (b) Power spectrum for depth of the 20°C isotherm, where psd is power spectral density, and Δf is frequency resolution in cycles per month.

[15] Figure 1b shows the power spectrum of the depth of the D20 surface. Interestingly, the spectrum clearly contains a number of well-separated maxima. As expected, one can easily identify the semi-annual component. There are also two surprisingly clear and strong low-frequency maxima corresponding to periods of approximately 18.7 months and 34 months. These low-frequency maxima rise far above the “noise” level in the spectrum and carry more energy than the 6-month component.

[16] To investigate the temporal behaviour of this signal, we conducted a wavelet analysis of the depth of D20 (Figure 2). It showed that the 34-month signal exists between approximately 1991 and 2002, while the 18.7-month signal is seen from 1993 to 2001. The 18.7-month signal is particularly strong during 1996–1999, when its amplitude exceeds that of the 34-month signal.

Figure 2.

Wavelet spectrum of 20°C isotherm at 6–7°S, 104–106°E, where white solid line denotes cone of influence.

[17] Figure 3a shows the low-pass filtered depth of the D20 surface with the cut-off frequency set at approximately 1/14 months. The filtered signal is quite strong, reaching the range of more than 110 meters during 1997–1998, which reflects the fact that most of the energy of the original signal is concentrated in the low-frequency part of the spectrum (Figure 1b). Between 1994 and 1999, the low-passed D20 oscillates with a period close to 18 months; it has maximums during recognised IOD events (1994, 1997) and minimums during recognised negative IOD events (1996, 1998) [Feng and Meyers, 2003; Rao et al., 2002]; however, it also shows strong upward (for example, early 1996) and downward (for example late 1995) motion at other times not associated with the IOD.

Figure 3.

(a) Low-pass filtered depth of 20°C isotherm at 6–7°S, 104–106°E with cut-off frequency at ∼1/14 months (black line) and original 20°C signal (thin blue line); (b) 34-month signal (green line), 18-month signal (red line), and sum of these two signals (black line).

[18] Figure 3b shows separately the signals corresponding to a 34-month (green line) and 18-month (red line) maxima seen in Figure 1b, obtained by band-pass filtering of the original time series. These signals interfere constructively in 1990–1991, 1994 and 1997–1998, with the 1994 and 1997 maxima coinciding with the recognised IOD events.

4. Analysis of SSH Data

[19] Figures 4a–4d show plots of PSD of SSH in a number of different locations in the IO, while Figure 4e represents the mean power spectrum over the large part of the IO. All these spectra have one common feature: they contain a few rather strong and well separated maxima. Apart from the semi-annual (Figure 4d) and annual components (Figures 4a and 4b), there are also three interannual variability modes, two of them already encountered in the spectrum of D20: those corresponding to periods of approximately 18–20 months (Figures 4a–4d) and 3 years (Figures 4a and 4d). There is also a signal with periods of 4–6 years (Figures 4b, 4c, and 4d). Overall, in different parts of the IO, the power spectra of SSH may lack one or other of the identified stand alone maxima, but in all cases, all the well-separated narrow maxima arising in power spectra do belong to one of these five frequency bands.

Figure 4.

Spectral analysis of SSH in different locations in the Indian Ocean (a, b, c, d) and over the ocean (e) - black dashed line, where psd is power spectral density, and Δf is frequency resolution in cycles per month.

[20] Figure 5 presents the plots of the spatial distribution of PSD for 6-, 12-, 18–20 month and 3- and 4–6 year bands. The signals in each frequency band have substantially different spatial distributions; all of them except the 18–20-month signal have already been described in the literature; however, to the best of our knowledge the spatial distributions of PSD of the SSH for each frequency band have not been published before.

Figure 5.

Spatial spectral density of SSH for signals: (a) 6 months; (b) one year; (c) 18–20 months; (d) 3 years; and (e) 4–6 years, where psd is power spectral density, and Δf is frequency resolution in cycles per month.

[21] Figure 5a presents the spatial distribution of PSD for the semi-annual signal. It shows a strong signal in the upwelling region offshore of Indonesia; north-propagating coastally trapped Kelvin waves around the Bay of Bengal; and a U-shaped structure in the western tropical IO with the maxima off the equator and a structure typical of Rossby waves.

[22] The spatial distribution of the 1-year signal shown in Figure 5b has maxima in the Red Sea, Arabian Sea, western part of the Bay of Bengal, and to the south-east of Indonesia. The maximum in the Arabian Sea is obviously caused by strong monsoon winds in this region. The maximum south of the equator is due to wind-forced, annual Rossby waves [Masumoto and Meyers, 1998].

[23] Figure 5c shows the spatial distribution of PSD for the 18–20 month signal. It contains strong signals offshore of Indonesia and in the Bay of Bengal caused by northward-propagating coastal waves (see auxiliary material, Movie S3). There are also two strong maxima in the central IO, in between approximately 5°S and 15°S and 60°E and 95°E and a zonally elongated, narrow maximum at 23°–24°S and between 65°E and 98°E.

[24] The spatial distribution of the 3-year signal shown in Figure 5d contains two major maxima in the IO: one offshore of Indonesia and the other in the central western IO. The maximum offshore of Indonesia is much weaker than for the 18–20 month signal, whilst in the central IO the signal is stronger, wider and more westwards. The signal is also present in the Indonesian Throughflow region and the western part of Pacific Ocean. Figure 5e represents the PSD density for the 4–6-year signal. It has strong maxima in the western IO, Indonesian Throughflow, and western Pacific region.

[25] The analysis of spatial distributions of PSD can provide important clues to the physics of the corresponding variability modes; however, the power spectra contain no phase information. To study the temporal variability of signals in different frequency bands, we conducted Inverse Discrete Fourier Transform of the filtered signals and plotted a series of snapshots of the reconstructed signal in the given frequency band. The resulting movies are provided in Movies S1–S5.

[26] It is interesting to compare the dynamics of the 18–20-month and 3-year signals (Movies S3 and S4). The 18–20-month signal is particularly strong between 1993–2000, it first starts developing near the Sumatra coast and then propagates north into the Bay of Bengal and east into the Indonesian Throughflow; later a second maximum of opposite sign develops over the region 96°E, 15°S to 55°E, 4°S and moves south-west as it amplifies, reaches a maximum and then gradually subsides. The 3-year signal is also strong during the 1994–2000 period. The initial maximum offshore of Indonesia develops simultaneously with a strong maximum of the same sign in the western Pacific and western part of the Bay of Bengal; later a second strong and wide maximum of opposite sign develops over the western part of the IO.

[27] The 4–6-year signal (Movie S5) clearly shows propagation of the signal from the Pacific Ocean.

5. Discussion and Summary

[28] We have identified five spectral bands carrying most of the low-frequency SSH variability in the IO: semi-annual, annual, 18–20 months, 3 years, and 4–6 years. The discovery of a strong 18–20-month signal is the key new result of this study. The signal clearly exists in the IO. Interestingly, despite the previous research investigating on the quasi-biennial mode, no major signal with the bi-annual (24-month) period was found. This leaves two possibilities: either the quasi-biennial mode described in the previous studies consists of two distinct 18–20-month and 3-year signals; or it corresponds to the 3-year signal while the 18–20 month signal represents a separate process. At this stage, the degree of interaction between the 18–20-month and 3-year signals is unclear. Both of these signals are particularly strong between 1994–2002, both display maximum variability in the Java upwelling region and in the western tropical IO; however, they evolve in a very different manner (Movies S3 and S4). The 3-year signal near Indonesia develops simultaneously with a signal in the Indonesian Throughflow and western Pacific region, whereas the 18–20-month signal develops near the Sumatra coast and only then propagates east into the Indonesian Throughflow area. The spatial/temporal dynamics of the 18–20-month signal indicate to us that it represents a primarily internal mode of the IO.

[29] Both the 18–20-month and 3-year signals achieve maximum amplitudes during the period between 1994 and 1999 (Figure 3b). The extremely shallow thermocline during IOD events of 1994 and 1997 may be seen as caused by constructive interference of these two signals (Figure 3b, red and green line), as well as the negative IOD event in 1998. This suggests to us that the IOD events may be a result of extreme manifestation of these two continuous processes rather than isolated events triggered by a particular state of the atmosphere-ocean system.

[30] The 4–6-year signal possibly corresponds to the signal identified in the literature as the ENSO signal. The Pacific origin of this mode can be verified by investigation of its temporal dynamics (Movie S5). However, a note of caution with this 4–6-year signal is that, because of the limited time span of the data used in this study (less than 14 years), the analysis of this extremely low-frequency signal is not as reliable as that for the higher-frequency bands. To reach more definite conclusions on the spatial and temporal properties of this signal, one needs to extend the time span of the currently available SSH data or use model output data.

[31] Overall, the use of spectral analysis of the SSH satellite observations in the IO allowed us to find strong, well separated, low-frequency spectral maxima and to investigate the spatial distribution and temporal dynamics of the corresponding variability modes. We believe that this information gives a strong quantitative basis for subsequent investigations of the underlying complex physical processes in the ocean-atmosphere system.

Acknowledgments

[32] The first author is supported by a joint CSIRO-UTAS Ph.D. scholarship in quantitative marine science (QMS) and a top-up CSIRO PhD stipend (funded from Wealth from Oceans National Research Flagship).

Ancillary

Advertisement