Geophysical Research Letters

Shoaling of the off-equatorial south Indian Ocean thermocline: Is it driven by anthropogenic forcing?


  • Wenju Cai,

    1. CSIRO Marine and Atmospheric Research, Aspendale, Victoria, Australia
    2. Wealth from Oceans National Research Flagship, CSIRO, North Ryde, New South Wales, Australia
    Search for more papers by this author
  • Arnold Sullivan,

    1. CSIRO Marine and Atmospheric Research, Aspendale, Victoria, Australia
    2. Wealth from Oceans National Research Flagship, CSIRO, North Ryde, New South Wales, Australia
    Search for more papers by this author
  • Tim Cowan

    1. CSIRO Marine and Atmospheric Research, Aspendale, Victoria, Australia
    2. Wealth from Oceans National Research Flagship, CSIRO, North Ryde, New South Wales, Australia
    Search for more papers by this author


[1] Surface warming since 1950 in the off-equatorial south Indian Ocean (IO) occurs without a consistent surface heat flux trend, and is accompanied by a shoaling thermocline. The associated dynamics have not been fully explored. Using 20th century climate model experiments, we test if the shoaling thermocline is attributable to a transmission from the Pacific, where a similar shoaling occurs, and whether it is climate change-induced. A 22-model average produces no such signal. An average of a subset of models that better simulate El Niño-Southern Oscillation (ENSO) and its transmission produces the right direction of the IO thermocline trends. The shoaling in this subset average, taken as anthropogenically induced, is far weaker than the observed, suggesting a significant multidecadal variability component in the observed changes. The Pacific contribution increases with a stronger model ENSO amplitude and broader meridional structure, highlighting the importance of realistic ENSO simulations in modelling long-term change in the IO.

1. Introduction

[2] Sea surface temperatures (SSTs) in the IO have been rising over the past four decades, contributing to droughts in the African Sahel [Giannini et al., 2003] and the Northern Hemisphere mid-latitudes [Hoerling and Kumar, 2003]. In the southern off-equatorial latitudes, it is accompanied by significant subsurface cooling [Alory et al., 2007] with a shoaling thermocline. Below the thermocline, the cooling is generated by aerosol-induced changes in the Conveyor circulation [e.g., Cai et al., 2007]. At the thermocline depth (or z20), the shoaling occurs along the off-equatorial Rossby wave pathway (Figure 1a) [see also Alory et al., 2007, Figure 2a], in part encompassing the thermocline dome of the south IO, where SST and thermocline coupling is strong. A “heat flux dilemma” [Schott et al., 2008] exists because there is no increase in surface heat flux into the off-equatorial IO based on available data [e.g., Yu et al., 2007] to account for the surface warming. There is a suggestion that the surface warming could be a consequence of a quasi-constant heat flux applied to an increasingly thinner surface layer as the thermocline shoals. Thus, it is essential to understand the cause for the shoaling thermocline. Are climate models able to reproduce the observed thermocline change? To what extent it is driven by climate change?

Figure 1.

Linear trends of 20°C isotherm (a) over the 1958–2001 period based on SODA-POP reanalysis and (b) averaged over 22 IPCC AR4 models over the period of 1951–1999. Units for Figure 1a are metres per 44 years, and for Figure 2b they are metres per 49 years.

[3] We address these issues using a coordinated set of 20th century climate model experiments performed for the Intergovernmental Panel on Climate Change's 4th Assessment Report (IPCC AR4). By averaging over a large set of experiments, the variability component will be largely removed, leaving the anthropogenic signal. An important forcing of the off-equatorial south IO thermocline is the transmission of ENSO-induced thermocline anomalies from the Pacific into the IO as Rossby waves [Wijffels and Meyers, 2004; Cai et al., 2005b]. Using 10 IPCC AR4 models, Alory et al. [2007] examined whether the south IO shoaling is a consequence of a weakening trend in the equatorial Pacific trade winds, which occurs in IPCC AR4 models [Vecchi et al., 2006]. While a relationship between the Pacific trade winds and south IO subsurface temperature appears valid, only 3 out of the 10 models produce a shoaling thermocline.

[4] Indeed a 22-model average of linear trends over the 1951–1999 period produces virtually no such signal in the south IO (Figure 1b), despite a shoaling thermocline over the west Pacific. Why are there such strong differences in the model thermocline trends in comparison to the observed (Figure 1a), particularly in the southern IO? We demonstrate that the Pacific forcing of the off-equatorial south IO shoaling thermocline relies on a model's ability to simulate the Indo-Pacific transmission process.

2. Data and Model and Experiments

[5] The performance of the IPCC AR4 general circulation models (GCMs) have been examined in terms of simulating ENSO [Guilyardi, 2006] and the Indian Ocean Dipole (IOD) [Saji et al., 2006]. In the latter study Saji et al. [2006] showed that the Pacific-IO transmission process is poorly simulated in many GCMs, with a number of deficiencies listed as possible causes. Here, we examine the extent to which the ENSO amplitude and structure affect the GCMs' transmission process. This is carried out here using outputs from one simulation of the late 20th century climate from each of the 22 IPCC AR4 models. Monthly anomalies of SST, z20, and upper 250 m heat content are constructed for the 1951–1999 period by subtracting the 49-year mean. Linearly detrended data are used to examine the transmission process. Then the trends in each model are examined. To benchmark model performance, we utilize the Simple Ocean Data Assimilation with the Parallel Ocean Program (SODA-POP) [Carton and Giese, 2008].

[6] ENSO signals transmit through the ocean to the IO via an equatorial Pacific (EP) wave pathway, in which Rossby waves generated in the central equatorial Pacific become coastally trapped waves where the New Guinea coast intersects the equator in the west Pacific [Wijffels and Meyers, 2004]. The reflected waves then propagate polewards along the northern Western Australia (WA) coast as coastally-trapped waves, radiating into the interior IO. There is also a subtropical North Pacific (NP) wave pathway [Cai et al., 2005b], in which subtropical NP Rossby waves associated with ENSO impinge on the western boundary and move equatorward along the pathway of Kelvin-Munk waves [Godfrey, 1975], and reflect as equatorial Kelvin waves, which then also propagate polewards along the northern WA coast as coastally-trapped waves, joining the EP pathway into the IO.

[7] For each model, a lag-correlation between detrended Niño3.4 anomalies (an ENSO index; average SST anomalies over 170°W − 120°W, 5°S − 5°N) and the upper 250 m oceanic heat content anomalies at various lags is calculated, from Lag −9 (9 months prior to the peak of Niño3.4) to Lag +9 (9 months after the peak). While the ocean resolutions of these models range from 0.2° × 0.3° to 4° × 5°, for the purpose of this study they are regridded onto a common grid (0.8° × 1.9°). From there, the evolution in each model is examined to determine if a model simulates the transmission, and whether both pathways (NP+EP) are involved.

3. Pacific-to-Indian Ocean Transmission

[8] Of the 22 GCMs, four models (CGCM3.1(T47), CGCM3.1(T63), GISS-AOM, GISS-ER) do not simulate a signal transmission from the Pacific to the WA coast. CSIRO-Mk3.0 has signals transmitted mostly to the Sumatra-Java and WA coasts, which is shown to be due to an unrealistic presentation of Timor Island in the model [Cai et al., 2005a]. In MRI-CGCM2.3.2, the transmitted signals arrive at the Sumatra-Java coast, not the WA coast. Overall, six out of 22 models do not properly simulate the transmission process.

[9] An examination of the evolution of the remaining models reveals that there are two distinctive groups: one where the transmission is strong (hereafter referred to as the NP+EP group), and another where the transmission is weaker and occurs predominately through the EP pathway (EP group). The evolution averaged over models in each group is plotted in Figures 2f2j and 2k2o, compared with the observed (SODA-POP) in Figures 2a2e. Models in the NP+EP and EP groups are indicated in Figure 3a by diamonds and circles, respectively. In the NP+EP group (Figures 2f2j), the NP signals lead the WA signal-peak by some 9 months (in terms of maximum correlation), as in the observed. Near the equator, Rossby wave signals in this group resemble the observed and are strong. In the EP group (Figures 2k2o), little off-equatorial NP signals are seen, and the equatorial Rossby waves are weaker and closer to the equator than those in the NP+EP group. In both groups, the EP signals lead the WA peak by about 3 months.

Figure 2.

Correlations at Lags −9, −6, −3, 0, +3 months (a–e) using linearly detrended anomalies between Niño3.4 and upper ocean heat content based on SODA-POP, (f–j) averaged over AR4 models that have a transmission of ENSO signals into the IO arriving at the WA coast through both the EP and NP pathways, (k–o) and through the EP pathway only. Positive lags indicate Niño3.4 leads.

Figure 3.

(a) Scatter plot of Niño3.4 correlation with heat content (at Lag +3) off the WA coast, against 1-standard deviation of Niño3.4 SST for 22 IPCC AR4 models. Models in the NP+EP group are indicated by large diamonds, and in EP group by small circles. Symbols with medium squares indicate models with an incorrect transmission process or without a transmission process. Observations are shown as a triangle. (b) The same as Figure 3a but for scatter plot of Niño3.4 correlation with heat content (at Lag +3) against signal to noise ratio of heat content. See text for a definition of the ratio.

[10] Overall, the evolution sequences in the NP+EP group are more realistic in terms of the strength of transmission and the evolution pattern. This suggests that important dynamic elements, such as equatorial, off-equatorial and Kelvin-Munk waveguides, are better represented in the models of this group.

[11] A realistic representation of the topography in climate models is essential for reproducing these transmission pathways. But once the topography for a model is fixed, the strength of the transmission is determined by the ENSO properties. An important factor that determines the strength of the transmission is the amplitude of ENSO. Figure 3a plots the one standard deviation value of Niño3.4 against the correlation between Niño3.4 and heat content off the northern WA coast (average over 115°E–122°E, 12°S–22°S) at Lag +3 for all 22 models. As such, the majority of the six models (squares, Figure 3a) that show an unrealistic or limited transmission have correlations close to zero. The majority of these models have a weaker ENSO than the observed. It is not clear whether it is the topography or the small ENSO amplitude that leads to the small correlation. Overall, we see a well-defined linkage of a strong transmission with a strong ENSO amplitude. Across all models, the relationship has a correlation indicating statistical significance at a 99% confidence level. Plots at Lag 0 and Lag +6 reveal a similar feature (not shown). The linkage holds for either of the EP and NP+EP groups, although models in the NP+EP group tend to have a greater ENSO amplitude, with a stronger linkage.

[12] In models with a strong ENSO, the tropical Indo-Pacific system is overwhelmed by ENSO signals, meaning that the ratio of “ENSO signal to stochastic noise” is greater than in models with a weak ENSO. To illustrate this, we define “signal” as the one standard deviation value associated with Niño3.4 (e.g. linear regression of grid-point anomalies onto Niño3.4, multiplied by the one-standard deviation value of Niño3.4), and “noise” as the standard deviation of the residual after removing ENSO signals. Figure 3b plots the ratios of heat content off the WA coast against the correlation of coastal WA heat content with Niño3.4 for each model. The ratios are generally much larger for models with a stronger ENSO, and stronger ratios are associated with a stronger transmission signal. This relationship is also statistically significant at a 99% confidence level. The results are therefore consistent with the inter-model variations depicted in Figure 3a, providing an explanation as to why the models with a weak ENSO produce a transmission signal that does not manifest itself above stochastic noise along the transmission pathways [Shi et al., 2007].

[13] Overall, the NP+EP group is more realistic, not only in terms of the transmission pathways, but also the overall ENSO pattern, and generally have a greater ENSO amplitude compared to the EP group. In the 6-model non-transmission group, the ENSO amplitude is small and its transmission to the IO is negligible. Below we show that model trends of the off-equatorial south IO thermocline depend on the simulation of ENSO and its transmission signals.

4. The IO Thermocline Trends

[14] Given that climate change signals tend to project onto existing mode of variability, one expects models that better simulate ENSO and its transmission into the IO will generate more realistic trends of the Indo-Pacific thermocline. This is indeed the case: an average over the six models that generate a poor transmission process produces a thermocline trend that is very unrealistic (Figure 4a) with a deepening in the off-equatorial south IO and little change in the west Pacific. Once these six models are removed (Figure 4b), a shoaling thermocline in the off-equatorial south IO and west Pacific emerges, whilst the NP+EP group shows up the strongest transmission (Figure 4c). A confirmation that the change in the off-equatorial south IO is propagation from the west Pacific, not locally forced, is provided by an ensemble-mean wind stress curl trends. Positive curl trends with an effect of deepening the thermocline are produced over the latitude range of the IO shoaling thermocline (not shown).

Figure 4.

Linear trends over the period of 1951–1999 of 20°C isotherm averaged over (a) 6 IPCC AR4 models with limited/improper ENSO transmission to the WA coast, (b) 16 IPCC AR4 models with transmission to the WA coast, and (c) 8 IPCC AR4 models in the NP+EP group only. Units are metres per 49 years.

[15] Comparisons of trends in the EP group with NP+EP group further highlight the dependence on the simulation of ENSO and the associated transmission process. The IO shoaling thermocline trend is larger in the NP+EP group because the strength of transmission is greater (Figure 4c). In fact, the top five models (ECHAM5/MPI-OM, ECHO-G, GFDL-CM2.0, GFDL-CM2.1, CSIRO-Mk3.5) that produce the strongest shoaling trends are all in this group. In the EP group, although the trends in the west Pacific are comparable, the trends in the off-equatorial south IO are generally weaker. Since there is no shoaling thermocline signal in the NP, the stronger trend in NP+EP is due to the greater strength in transmitting the tropical signal.

[16] The west Pacific shoaling in Figure 4b is weaker than that reported in early studies. Vecchi et al. [2006] showed that since 1860 there is a shallowing of some 10 meters (or about 3.5 m per 50 years) in a Geophysical Fluid Dynamical Laboratory (GFDL) model. They noticed the GFDL model generates a greater weakening in the Walker Circulation and hence a stronger shoaling of the west Pacific thermocline, compared with other models. The observed thermocline shoaling (Figure 1a) is of the order of some 10–20 m per 50 years. The multi-model average thermocline trends in both the Pacific and the IO are far weaker (Figure 4b). Even in the NP+EP group, the maximum trend in the south IO is less than 8m per 50 years, and is confined to a narrower meridional band. Thus, it would seem that the observed large Indo-Pacific thermocline change is a consequence of a small climate change-induced trend superimposed on a far larger variability-induced component.

5. Conclusions

[17] Off-equatorial south IO surface warming shows several intriguing features, including the absence of a trend in surface heat flux and an accompanying shoaling thermocline. The present study focuses on the dynamics of the latter feature and assesses the extent to which the shoaling thermocline is attributable to a transmission from the west Pacific, where a similar shoaling occurs due to a weakened Walker Circulation associated with an El Niño-like warming pattern. A 22-model average produces virtually no shoaling of the IO thermocline. The Pacific forcing of the south off-equatorial IO thermocline change relies on the model's ability to successfully simulate ENSO properties and the ENSO transmission process, which include an EP and a subtropical NP pathway to the northern WA coast, where the signals radiate into the IO. We find that four models fail to simulate transmission through either pathway and two models have signals arriving at a wrong location. Eight models produce transmission via the EP pathway. Only eight out of 22 models produce transmission via both pathways, and these models tend to have a greater ENSO amplitude and broader meridional extent of ENSO anomalies, which is necessary for a NP transmission pathway. An ensemble average of the 16 models with transmission reproduces a trend of thermocline shoaling in the off-equatorial IO; the strongest shoaling is produced by the more realistic models. Comparing the model results to the observed, it seems that the shallowing thermocline across the equatorial Indo-Pacific basin is a result of strong multidecadal variability masking a smaller climate-change component. Our results therefore highlight the importance of reproducing ENSO properties and dynamics, such as the equatorial, off-equatorial and Kelvin-Munk waveguides, in simulating IO climate change signals.


[18] We acknowledge the work undertaken by numerous international modelling groups who provided their model experiments for this analysis. In particular we recognise the significant work that the Program for Climate Model Diagnosis and Intercomparison (PCMDI) has achieved in collecting and storing the model data. This work was supported by the Department of Climate Change of the Australian Government. We are appreciative to the two anonymous reviewers for their helpful comments which improved the manuscript.