SEARCH

SEARCH BY CITATION

Keywords:

  • ENSO;
  • Indonesian throughflow;
  • forecast;
  • coupled model;
  • western boundary

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References

[1] In situ and satellite sea level data sets over 1980–1998 are used to estimate interannual variations of the geostrophic zonal transport from the western Pacific Ocean into the Celebes Sea. Then the transport component due to the Pacific wind-driven and fully reflected equatorial waves is removed. Finally, the residual variations, named Indonesian throughflow correction (ITFC), are used to correct for the model closed western boundary. ITFC inflows/outflows are leading the warm/cold events by a few months. They are then prescribed at the model western boundary over 1980–1998 to compare with closed boundary simulations. An ITFC inflow anomaly makes the eastern Pacific slightly warmer and trade winds in the central Pacific slightly weaker. Indeed, these quantities are simulated by both experiments in very good agreement with observations. More importantly, prescribing the ITFC greatly improves two characteristics of the model that play a key role in the coupled simulations. The first is the RMS variability of sea level versus longitude; the ITFC shifts to the west the position of the minimum. The second is the basin average sea level; the ITFC largely amplifies its fluctuations, stimulating the charge and discharge of the system. Consequently, the ITFC can have a very large impact on El Niño-Southern Oscillation forecasts. For example, the 1982/1983 and 1997/1998 1-year lead forecasts which fail to predict the warm events with closed boundaries are successful with the ITFC. The impact of the ITFC on coupled simulations is not linearly dependent on the ITFC value itself: In addition to the fluctuating conditions at the western boundary, results depend on how close to instability the system is at each time step.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References

[2] With the exception of a few recent studies [Latif et al., 1994; Schneider and Barnett, 1997], most of the coupled models used to study El Niño-Southern Oscillation (ENSO) events in the Pacific rely on a tropical ocean model with a closed western boundary. In reality, this boundary is not closed. The Indonesian throughflow connects dynamically and thermodynamically the Indian and the Pacific oceans [Gordon, 1986; Godfrey, 1996; Schneider, 1998]. Its mean transport contributes to a warming of the eastern Indian Ocean and a cooling of the western Pacific warm pool [Hirst and Godfrey, 1993; Rodgers et al., 1999]. Its variations are thought to be involved in those of the El Niño-Southern Oscillation (ENSO), including remote areas such as the cold tongue. However, little is known today about how the coupled mechanisms actually work. Indeed, much is unknown about the Indonesian throughflow transport itself [Lukas et al., 1996]. Wyrtki [1987] estimated the throughflow variations by taking the sea level differences between Davao in the western Pacific and Darwin in the eastern Indian Ocean and found little correlation with ENSO. Clarke [1991] finds a stronger correlation by developing a seven-island model to account for the leakage of low-frequency energy at the western Pacific boundary. He thus estimates the interannual transport amplitudes using the Darwin sea level only, arguing that the Darwin-Davao sea level difference cancels out a lot of the Darwin signal. Further observations [Meyers, 1996; Fieux et al., 1996] and model simulations [Kindle et al., 1989; Clarke and Liu, 1994; Vershell et al., 1995; Murtugudde et al., 1998] also support that there is a strong relationship between the throughflow and ENSO. The transport from the Pacific to the Indian Ocean tends to be weaker during El Niño and stronger during La Niña. As warm and cold events are well correlated with equatorial westerlies and easterlies in the central Pacific, the throughflow interannual variability is explained to a large extent by the equatorial Pacific winds [Clarke and Liu, 1994; Murtugudde et al., 1998]. However, the correlation between the transport variations and ENSO is not high; other factors than the zonal wind in the equatorial Pacific affect the transport significantly [Murtugudde et al., 1998; Meyers, 1996]. The throughflow corresponds to the pressure exerted by the entire Pacific onto the Indian Ocean and it involves off-equatorial processes as well. The mean throughflow transport is expected to be in balance with the wind curl over the Pacific basin (or the Indian/Atlantic Oceans) as far as 45°S south of Tasmania [Godfrey, 1989; Wajsowicz, 1994; Godfrey, 1996]. Although the Island Rule dynamics applies only to quasi-steady motions, recent studies suggest that processes on decadal-to-multidecadal timescales interact with the climate interannual variability [see, e.g., Lau and Weng, 2001]. One also needs to account for the barotropic fluctuations of the throughflow which respond almost instantaneously to winds along Australia's south coast (Masumoto and Yamagata, 1996). ENSO forecasts are sensitive to sudden fluctuations as much as decadal changes [Perigaud and Cassou, 2000]. Because of incomplete data sets, inaccuracies, and differences in model assumptions, the various estimates of the throughflow transports available today still bear a large uncertainty.

[3] The purpose of our study is to investigate the impact of the Indo-Pacific connection on ENSO. This is done with a tropical Pacific Ocean model coupled to a statistical model of the atmosphere, by prescribing two different types of conditions at the western boundary of the model. Either we assume closed conditions, or we allow inflow/outflow variations across the boundary. The coupled model can be integrated in time either in a “forced” or a “coupled” context. A forced context means that the observed wind stress anomalies from 1980 to 1998 drive the ocean model which delivers the simulated sea level and sea surface temperature (SST) anomalies, the latter being used to force the atmosphere which finally delivers the simulated wind stress anomalies. In the coupled context, the wind is estimated at each time step by the atmosphere model using the simulated SST, and both ocean and atmosphere models are integrated in time without data insertion. Two-yearlong coupled experiments initialized with the conditions provided by the forced simulations at any month in between 1980 and 1998 correspond to “forecasts.”

[4] This paper does not study the role of the Indo-Pacific oceanic connection by simply replacing a closed western boundary with some arbitrarily modified open boundary condition. Our aim is to correct the model for the fact that in reality, there is some mass flow across this boundary. Sea level data during 1980–1998 provide an estimate of the geostrophic transport variations across the section at 130°E where the Pacific Ocean opens to the Celebes Sea. The correction (Indonesian throughflow correction (ITFC)) is obtained by removing from these geostrophic variations the transport anomalies due to the wind and full reflection of equatorial waves in the Pacific. The latter are provided by the model with closed boundary conditions forced by the observed winds over 1980–1998. In order to examine the validity of this approach, the model outputs obtained in the closed boundary case are first compared with various oceanic and atmospheric data. Then, examining if prescribing the ITFC variations during a model integration forced by the same wind improves the model outputs away from the western boundary or not is the crucial step of validation.

[5] This paper is organized as follows. In the next section we describe the satellite and in situ data used to compute the sea level anomalies over the Indian and Pacific Oceans between 1980 and 1998 and compute the geostrophic transport across the opening of the Pacific onto the Indian Ocean. In section 3 we present and validate the results of the forced experiment with closed western boundary conditions (the control run). Section 4 is dedicated to the estimate of the ITFC applied to correct the closed model boundary. In section 5 the ITFC is prescribed during a forced experiment for comparison with the control run and validation with data. Section 6 is devoted to the impact of ITFC in forecast experiments. A summary of results is proposed in section 7.

2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References

[6] Sea level data over the Indian and Pacific Oceans are estimated from expendable bathythermograph sondes (XBT) between January 1980 and March 1998 and from the TOPEX/Poseidon (T/P) satellite between October 1992 and June 1998. The XBT data set consists of the analyzed temperature fields provided by N. R. Smith from the Bureau of Meteorology Research Center [Smith, 1995]. Actually, this data set does not provide sea level. Instead, we compute the surface dynamic height (dynh) relative to 400 m as a proxy. Both data sets are estimated every month with a 1° × 1° resolution.

[7] Monthly varying climatologies are computed from January 1980 to December 1996 for dynh, and from January 1993 to December 1996 for T/P to determine the sea level anomalies (SLA). All the results of this paper, for data and model, are anomalies relative to the period January 1980–December 1996. Because of geoid uncertainty, T/P does not provide the sea level at any given time. It only provides sea level variations relative to the sea level average during the satellite period. But this average is a function of space, which is not known. It is certainly expected to be different from the surface that would be obtained by averaging sea level over 1980–1996. Therefore a reference surface must be added to T/P data. Indeed, the surface of dynh anomalies averaged between January 1993 and December 1996 shows a signal larger than 1 cm (Figure 1a). The eastern and southern Pacific (mostly east of the date line and down to 10°S) is higher over these 4 years compared to the 13 years before. By contrast, the western and northern Pacific (mostly west of the date line and up to 20°N) is lower. These anomalies have amplitudes of 2 to 4 cm and, on the basis of their homogeneity in space, they cannot be noise. All the T/P data are then corrected by adding the surface displayed in Figure 1a. Note that adding such a reference surface to T/P data has a very big impact on the ENSO forecasts that are obtained by coupled models initialized with altimetric observations (Perigaud et al., 2000a).

image

Figure 1. (a) Mean sea level anomaly derived from dynamic height (dynh) over January 1993–December 1996. Units are in centimeters. Dashed lines correspond to negative values. (b) Ratio of the RMS variability of T/P over the RMS variability of dynh. The RMS is computed over October 1992–March 1998. Dashed lines correspond to values smaller than or equal to 1.

Download figure to PowerPoint

[8] The T/P satellite provides a signal which does correspond to sea level, whereas dynh is only a proxy. Errors of several centimeters are made by using dynh as sea level, because the density variations below the reference depth and the salinity changes above are not accounted for in dynh. So the two signals are not expected to perfectly match one another. Since T/P data correspond to accurate sea level measurements, and sea level, or density changes, is the appropriate quantity for validating the model (see the following sections), we choose to retain T/P data when available, and correct dynh as follows: For each data set, we computed the RMS variability between October 1992 and March 1998. Their ratio at each point (Figure 1b) is used to multiply the dynh time series so that the two data sets have the same amplitude during the satellite period. Figure 1b shows that the ratio is close to 1.0 in the equatorial Pacific Ocean between 10°S and 10°N. As anticipated, the ratio reaches larger values elsewhere. Figure 1b suggests that in addition to the two error sources mentioned above, the lack of in situ measurements in some regions is also a source of error for dynh. The off-equatorial Pacific and the Indian Oceans are indeed poorly covered by XBT data in comparison with the equatorial Pacific.

[9] The dynh values over 1980–1998 are then multiplied at each point by this ratio in order to match the amplitude observed by T/P. Time series estimated by dynh and T/P in two regions of the Pacific or the Indian Ocean are presented in Figure 2 before and after applying the correction to examine the agreement of the two data sets. Before the corrections, the RMS difference in the Pacific Ocean is 3 cm for signals which have RMS amplitudes of 8 cm and 9 cm, and the RMS difference in the Indian Ocean is 5 cm for signals of 4 cm and 7 cm. The agreement is improved when the corrections are applied (RMS differences are 2 cm and 3 cm, respectively). In the rest of the paper, the 1980 to 1998 time series correspond to the corrected dynh between January 1980 and September 1992, and to the corrected T/P data between October 1992 and June 1998, and are all called “SLA” for “sea level anomalies.”

image

Figure 2. Time series of sea level anomalies (SLA) derived from dynh (solid line) or from T/P (dotted line) averaged over (a and c) the western Pacific Ocean (135°E–140°E, 0°–5°N); (b and d) the southeast Indian Ocean (100°E–120°E, 14°S–8°S). Figures 2a and 2b are prior to corrections and Figures 2c and 2d are after corrections.

Download figure to PowerPoint

[10] Geostrophic zonal current anomalies are then derived from SLA following the method described by Delcroix et al. [1994] for matching the equatorial points with the off-equatorial ones. The solid line in Figure 3 represents the geostrophic transport across 130°E between 2.5°N and 7.5°N. The position and width of this section are chosen to match the latitudes of the ocean opening to the Celebes Sea between Mindanao and Halmahera islands. The longitude is slightly to the east of the actual opening in order to avoid data inaccuracies close to the coast. It is also the section which is retained for correcting the boundary conditions of the model (see section 4). The transport amplitudes in sverdrups have been obtained by assuming that the transport takes place in a homogeneous upper ocean layer of 150 m (this also corresponds to the model assumption; see section 3).

image

Figure 3. Time series of zonal transport anomalies in Sverdrup across the section 130°E (2.5°N–7.5°N), counted positive to the east, compared with SLA in centimeters corresponding to (a) the opposite of the signal observed close to Darwin (122°E, 12°S) and (b) the difference (Darwin-Davao), where the Davao signal is that observed by the SLA at 125°E–5°N. Time series have been smoothed over 6 months.

Download figure to PowerPoint

[11] The geostrophic transport in Figure 3 is compared to two different estimates of the Indonesian throughflow (ITF) on the basis of sea level. One is the SLA observed at 122°E, 12°S, close to Darwin, in the Indian Ocean north of Australia. Clarke [1991] explains that a sea level high (low) at Darwin corresponds to an increase (decrease) of the ITF from the Pacific to the Indian Ocean. Figure 3a indicates that the geostrophic transport anomalies across the western boundary of the Pacific also match the Darwin SLA highs and lows in 1985, 1987, 1989, 1992, and 1994. The time series of Figure 3a are correlated by 0.40. The second estimate used for comparison is the difference between the Darwin signal and the SLA observed at 125°E, 5°N, close to Davao, in the Pacific east of the Philippines. Wyrtki [1987] proposed that the throughflow increases (decreases) to the west when the sea level at Davao is higher than at Darwin, and several authors (i.e., Murtugudde et al., 1998; Potemra and Lukas, 1999] found confirmation in their simulations or observations. In 1980–1981, 1986–1987, and 1996–1997, the geostrophic estimate presented in Figure 3 is eastward, and Darwin is higher than Davao. In 1983–1984, 1988–1989, and end of 1992, the geostrophic estimate is westward and Darwin is lower than Davao. The time series of Figure 3b are correlated by 0.66. Thus the ITF variations estimated by other authors using sea level agree to some extent with the geostrophic transport across the opening of the western Pacific boundary. Indeed, this result makes sense, as this opening corresponds to the main water pathway of the ITF [Nof, 1996]. On the other hand, it is also clear that the three estimates of Figure 3 do not always match each other. Prior to 1982 and past 1990, it even happens that Darwin has a sign opposite to its difference with Davao. Note that the correlation we find in Figure 3b is close to the value of 0.75 found by Murtugudde et al. [1998] over 1980–1995. Indeed, our geostrophic estimate compares reasonably well with the ITF transport across the Indonesian channels simulated by Murtugudde et al. [1998]. Both have maximal westward anomalies of ∼5 sverdrups in 1983–1984 and 1988–1989 and minima of ∼3 Sv in 1986–1987 and ∼5 Sv in 1991. However, we find an eastward anomaly in 1994, whereas they do not. Note that Meyers' [1996] ITF estimate along the line between NW Australia and Java has a westward anomaly in 1994. A careful comparison was performed with Meyers' [1996] data, as we initially intended to use the latter for correcting the model. Results are reported elsewhere. For the present paper, the reader needs to be aware of the following conclusions.

[12] On the one hand, these various estimates of the ITF transport, which are all positively and significantly correlated, do have a large uncertainty. It is possible that our estimate misrepresents the ITF anomalies by several sverdrups. On the other hand, our estimate accurately represents the observed geostrophic transport across the western boundary of the Pacific between 2.5°N and 7.5°N. We know that it is a priori a quantity we can use in order to correct the closed boundary conditions of the model. If the model experiments with the corrected conditions deliver improved simulations, then, we will be sure that this quantity represents realistic inflow/outflow anomalies over 1980–1998. In the next section we compute the difference between the geostrophic transport of Figure 3 and the one simulated by the model with closed boundaries across the same section.

3. Model and Control Run

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References

[13] The model used in this study is an Intermediate Coupled ocean-atmosphere Model (ICM) of the tropical Pacific. It simulates interannual anomalies relative to a prescribed climatology. The ocean model named Trident has a dynamic component [Boulanger, 2001] and a thermodynamic component [Boulanger and Menkes, 2001]. The SST anomalies are governed by the zonal, meridional, and vertical anomalous heat transports. Trident is similar to the ocean component of the ICMs presented by Zebiak and Cane [1987, hereinafter referred to as ZC] or by Cassou and Perigaud [2000]. Compared to these ICMs, Trident is an improved ocean model, a second baroclinic mode is implemented and can optionally be activated, the horizontal friction is spatially variable in the momentum equation, and a term of vertical diffusion is added for temperature. The reader is invited to check Boulanger [2001] and Boulanger and Menkes [2001] for a complete description, analysis, and validation of the model. The reader unfamiliar with these ICMs can find more details in Appendix A, in which the results useful for this paper are summarized.

[14] The atmospheric component used here is a statistical model named Astat, which is the atmospheric component of the “Tsub.Astat” model presented by Cassou and Perigaud [2000]. Astat is constructed by singular value decomposition between the Florida State University (FSU) wind stress and Climate Analysis Center (CAC) SST anomalies between 1970 and 1983. The Astat model forced by the observed SST anomalies over 1980–1998 reproduces well the ENSO variations of the wind, including the zonal component in the equatorial central Pacific and the meridional component in the Intertropical Convergence Zone (ITCZ) and South Pacific Convergence Zone (SPCZ). Trident is coupled to a statistical atmosphere rather than to the ZC atmospheric model in order to avoid errors in the zonal and meridional wind components of the eastern Pacific.

[15] For the present study, we choose to retain only one vertical mode in the baroclinic ocean, although more than one mode is needed to account for the totality of the actual Indonesian throughflow variations. Our choice is justified when considering the large uncertainty of the throughflow transport estimates found in the literature, whether from data or from models. Indeed, studies focusing on the upper layer ocean changes, including recent studies [Vershell et al., 1995; Potemra, 1999], are also based on a single mode. In addition, when performing the experiments with prescribed throughflow transport at the western boundary of the model, retaining a single mode avoids arbitrary projections of the observed SLA on the vertical. Moreover, thanks to the introduction of the variable coefficient of horizontal friction, the simulations with one single baroclinic mode are almost as realistic as those with two modes, given an adequate choice of parameters [Boulanger, 2001]. Accordingly, a phase speed equal to 2.50 m s−1, an upper layer thickness equal to 150 m and a Rayleigh friction of a 12 months time decay, are retained here. In a coupled context, like the other ICMs mentioned above, Trident.Astat has an oscillatory behavior or not, depending on initial conditions and parameters such as the coupling coefficient. It can reproduce ENSO-like oscillations like ZC model, but it simulates a mechanism of equatorial heat content recharge by the off-equator which is very different from ZC and resembles the one simulated by “Tsub.Astat” (see Appendix A).

[16] The first forced experiments consist of forcing the model by observed wind stress anomalies from January 1980 to July 1998 with the closed boundary condition. In such forced experiments, the inputs are the wind stress anomalies; the ocean model is first integrated from 1980 to 1998 and delivers the oceanic outputs like the SLA and SST analyzed in this paper. Then the atmosphere model is forced by the SST outputs from 1980 to 1998 to deliver the wind stress ouputs. So, although the latter are not used during the model integration like in coupled experiments, wind outputs of forced experiments are analyzed for comparison with the wind input. The experiment used as the control run (CR) for this paper is initialized from rest in 1970 and is forced by the stress anomalies named Astat (CAC). They are the stress anomalies estimated by Astat forced by the observed SST anomalies. A second experiment named “run.FSU” is initialized from rest in 1961 and forced by the stress anomalies derived from the FSU data set. Model outputs are monthly averaged. Maps of their variability over 1980–1998 are presented for the control run (Figure 4) and compared with observations (Figure 5). Attention is first focused on reproducing the correct patterns of the oceanic and atmospheric signals. Then, we validate the time series of spatially averaged quantities commonly used like the Niño3 index and examine their sensitivity to the wind used to force the model.

image

Figure 4. Maps of (a) SLA, (b) SST, (c) zonal (TX), and (d) meridional (TY) wind stress variability over 1980–1998 simulated by the control run. Units are in centimeters, °C, and dyne per centimeter square, respectively. Shaded areas correspond to the model sea-land mask.

Download figure to PowerPoint

image

Figure 5. Same as Figure 4 but for observations.

Download figure to PowerPoint

[17] The simulated and observed sea level variability maps (Figures 4a and 5a) present the expected patterns with maxima along the equator east of the date line and in the off-equatorial regions west of the date line, but they also display some discrepancies. Compared to the observed maxima, the simulated maxima are stronger in the southwest (12 cm compared to 8 cm), weaker in the eastern equatorial region (6 cm compared to 8 cm), and located significantly east and south of the observed maximum in the northwest region (near 170°E, 5°N compared to 130°E, 10°N). Also, the simulated minimum of variability in the central Pacific is located to the east of the date line, whereas in the real ocean it occurs near the date line. Part of this discrepancy is explained by the fact that the model has only one baroclinic mode. A minimum of four modes when the horizontal friction is homogeneous is suggested by Dewitte [2000] to achieve the skill of an ocean general circulation model (OGCM) in reproducing the observed sea level variability. However, even with four modes or with OGCMs, the minimum of variability is often located to the east of the date line. In Trident, the friction is a function of distance to the coasts and allows the model with a single mode to be almost as skilled as with two. Other factors than the vertical simplification explain this model deficiency, in particular the fact that the model overestimates the SLA due to reflected equatorial waves in comparison to the wind-driven ones [see Perigaud and Dewitte, 1996]. It is crucial to reduce this deficiency as much as possible in a forced context, because the strong coupling in the eastern Pacific tends to displace the minimum even further to the east. Moreover, the strength and duration of the warm events simulated by the coupled model crucially depend on the location of this minimum. We will return to this issue in section 6.

[18] The simulated and observed SST maps present strong similarities (Figures 4b and 5b). The model reproduces the observed patterns of variability, including the presence of two maxima, one offshore slightly south of the equator with a maximum of 1.3°C at 120°W, and the other along the Peruvian coast with an amplitude of 1.8°C. The agreement between the model and the data is indeed quite remarkable.

[19] The variability of the zonal (TX) and meridional (TY) wind stress simulated by the control run is compared here to the observed anomalies Astat (CAC) used to force Trident in this control experiment. For TX (Figures 4c and 5c), the regions of intense anomalies are located on the equator between 170°E and 120°W, and between 10°S and 5°N. The model simulates the maximum between the date line and 160°W slightly south of the equator. The fact that the model places the wind maximum close to the correct location is another significant advantage of the Astat model compared to the ZC atmospheric model [see Perigaud and Dewitte, 1996]. Values are weaker, though, for the model than for the data (0.018 versus 0.024 Pa). ICMs often simulate an amplitude too weak in the forced context, and it is common to correct for this weakness by choosing a coupling coefficient larger than 1.0 when performing coupled experiments.

[20] For TY (Figures 4d and 5d), the main patterns are reproduced by the model, with the highest values between 180°E and 110°W, north of the equator and south of 10°N. The maximum in the north is located in the central Pacific (near 140°W and 5°N), the maximum in the south is weaker and to the west. Contrary to the zonal component, the meridional wind has stronger values for the model than for the data (0.020 versus 0.018 Pa). Similar characteristics are found for the two wind components simulated by Tsub.Astat. Because both components play an important role in the simulated ENSO, the coupling coefficient must be adjusted in order to reproduce mechanisms that are sufficiently realistic. This coefficient, which is applied in the computation of the wind stress outputs of the forced experiments, is first set to 1.0.

[21] One may question the robustness of this validation because the observed winds are actually obtained from observed SST. As illustrated in Figure 6, the agreement with the FSU observed stress is good for both components in the central Pacific. Note that the FSU stresses used here have not been detrended like the stress used to force the ZC model because since 1980, most ships have been equipped with anemometers, and the low-frequency trends of the FSU winds could be real. Discrepancies between FSU and Astat (CAC) winds are observed in the northwestern Pacific (Figure 6d). In order to examine how this discrepancy affects the control run, the model outputs generated by run.FSU are plotted in Figure 7 and allow further validation of the control run.

image

Figure 6. Time series of zonal (TX) or meridional (TY) wind stress anomalies averaged over (a) Niño4 (160°E–150°W, 5°S–5°N), (b) NiñoN (180°E–140°W, 1°N–9°N), (c) NiñoS (180°E–140°W, 5°S–15°S), and (d) NiñoWN (130°E–170°E, 1°N–9°N). Solid line is the Astat(CAC) stress and dashed line is the FSU stress.

Download figure to PowerPoint

image

Figure 7. Time series of (a) SLA over Niño West (130°E –170°E, 10°S–10°N), (b) SLA over Niño3, (c) SST over Niño3, and (d) zonal wind stress over Niño4. Solid line is for observations, dotted line is for the control run, and dashed line is for run.FSU.

Download figure to PowerPoint

[22] All the results displayed in Figure 7 illustrate that the model reproduces the observed indices between 1980 and 1998 really well in the ocean and in the atmosphere. The biggest disagreement between the control run and the data is found in sea level in the western Pacific in 1983–1984. The control run simulates a trough of 15 cm, whereas a trough of 10 cm is observed (Figure 7a). Obviously, this misfit is due to an error in Astat (CAC) stress. Note that the error is not due to TX, but to TY. This is noteworthy because it is common to assume that the TY role on ENSO is negligible, whereas it is very important, especially in coupled experiments (see Perigaud et al. [2000a] and Appendix A). In 1983, Astat (CAC) does not detect a strong enough southward displacement of the ITCZ (check Figure 6b). Thus the strong TY anomaly of FSU in 1983 is associated with an anticyclonic curl which, given the cyclonic curl associated with TX, explains why the signal simulated in run.FSU is less negative than in the control run. The TY anomaly observed by FSU in 1983 is the strongest of all since 1980. Note that in 1997–1998 it is TX which is the strongest of all and then, the 17 cm sea level trough observed by the satellite is as strong as in the control run (19 cm) or in run.FSU (14 cm). In the eastern Pacific (Figure 7b) the control run does not have an error in 1983 as large as in the west. For both experiments, the first two events are slightly overestimated by the model, whereas the last two are slightly underestimated. Overall, between 1980 and 1998 the agreement is good: The control run and data are correlated by 0.79, and they have an RMS difference of 4 cm in the west and 3 cm in the east. The control run also reproduces the observed SST and TX anomalies remarkably well (Figures 7c and 7d). Correlations with observations are 0.87 and 0.91, respectively, and RMS differences are 0.5°C and 0.006 Pa. The agreement is of similar quality for run.FSU.

[23] In summary, the control run reproduces the sea level, the SST, and the wind stress variability indices of Figure 7 quite well over 1980–1998. Relatively speaking, the biggest model deficiency is found in the sea level. The minimum of variability is not located where it is observed at the date line, and the amplitudes of the anomalies in the western and eastern Pacific do not always match the observations. Part of this discrepancy is due to errors in the wind used to force the control run. However, replacing this wind by the FSU wind does not significantly improve the simulations, except for the sea level in the western Pacific in 1983. In addition to the wind forcing, the sea level is affected by the wave reflection at the boundaries. The closed boundary conditions verified by the model ignore the mass inflows and outflows that take place across the western boundary in reality. Let us now try and estimate such flow variations.

4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References

[24] The baroclinic model is decomposed into Rossby and Kelvin components as shown by Cane and Patton [1984]. To verify the conditions of full reflection at the western boundary, the model integrates the zonal transport due to the Rossby signal between 20°S and 20°N and determines the Kelvin amplitude that guarantees no inflow or outflow across this boundary. We have seen in section 2 that sea level data can be used to estimate the geostrophic transport at the western boundary of the model across the section where the boundary is open in reality. We can now use the model outputs to estimate the transport (Rossby + Kelvin) across this part of the boundary which is due to the wind-driven, propagated, and-fully reflected equatorial waves. If one assumes that the ocean model is perfect except for its closed conditions at the western boundary, the difference between the observed and the simulated estimates (ITFC) corresponds to the transport which is due to the oceanic connection of the Pacific with the Indian Ocean. Studying the impact of the Indo-Pacific connection on the ENSO signals simulated by our Pacific model then consists of prescribing the ITFC inflow-outflow variations at the western boundary, meaning that at each time step, the ITFC is added to the value of the Kelvin wave at the western boundary obtained by the wind and the full reflection of the incoming Rossby signal. So let us first examine the model transport and then determine the ITFC.

[25] The 1980–1998 time series of the geostrophic zonal current anomalies are derived from the simulated sea level following the same method as for the observations in section 2. In fact, with the model SLA it was verified that this method gives the same zonal current anomalies as the zonal baroclinic currents computed by the model. Computations were done for the control run and for run.FSU in order to examine the part of the signal which could be due to wind error. The transports in Figure 8a correspond to the integration of the current anomalies between 2.5°N and 7.5°N at 130°E as in Figure 3b. Other latitudinal bands between 0° and 10°N and other longitudes between 130°E or 140°E have been tested. Transport estimates then slightly differ in amplitude and time, but the results derived by prescribing the ITFC in the forced context (section 5) are not very sensitive to the extent or to the position of the ITFC section. On the contrary, results in the coupled context (section 6) are significantly affected, and the transport which best improves the coupled SST, wind, and SLA corresponds to the 2.5°N–7.5°N section at 130°E. For simplicity, the results presented in this paper all correspond to this section.

image

Figure 8. Time series of (a) geostrophic transport across 130°E (2.5°N–7.5°N) derived from observed (solid line) or simulated SLA (dotted line is for the control run; dashed line is for run.FSU) and (b) ITFC derived from the transport difference (data-control run) in sverdrups and Niño3 SST index in °C (multiplied by 3 for the plot).

Download figure to PowerPoint

[26] The strongest transport anomalies simulated by the control run are to the west in late 1982–early 1983, in 1987, and in 1997 (Figure 8a). This is because the integration of the currents between 2.5°N and 7.5°N accounts more for the Rossby flow than for the Kelvin component which is centered at the equator. As expected, the Rossby currents in the western Pacific are westward and the strongest during the mature phase of the El Niño events. However, each event is unique. Note that in late 1992–early 1993, after the termination of the relatively mild event, both experiments simulate an eastward anomaly which is the strongest of all. The years 1993 to 1998 correspond to the biggest scatter between the two model experiments and the data. It is striking in Figure 8a that the observed transport significantly differs in amplitude and phase with the simulated transports, whereas both simulations are in good agreement. Thus wind error is not the main reason for why the observed transport differs from that of the model. The transport difference (data-model) is possibly representative of the Indo-Pacific connection.

[27] The transport difference for the control run, the ITFC, is presented together with the ENSO signal as observed with the SST Niño3 index (Figure 8b). The correlation between the two signals is positive (0.65). Note that the geostrophic transport of Figure 3b is also positively correlated with ENSO, but the correlation is not as high (0.29), close to the value (0.31) found by Murtugudde et al. [1998]. So the ITFC is more correlated to ENSO than the transport itself. This is because warm events trigger westward upwelled Rossby waves that reflect at the western boundary to erode the warm growth; the model transport is negatively correlated with ENSO (−0.76). Even though the ITFC is more significantly positively correlated to ENSO than the transport itself, what is important in furthering our understanding of the coupling between the Indo-Pacific connection and ENSO is precisely the phase difference between the two series. The sign of the shift indicates whether the ITFC changes are primarily governed by the equatorial SST of the Pacific via the wind forcing, or whether it is rather the SST changes that are influenced by the ITFC fluctuations via oceanic transport. The positive peaks of the ITFC in Figure 8b are clearly leading the SST peaks by 6 to 12 months. Actually, the ITFC is leading the SST almost all the time over 1980–1998; the maximum correlation (0.72) is obtained for a 2 month lead in the ITFC. Hence the ITFC fluctuations could have a triggering role in the warm and cold events observed over 1980–1998. Of course, the ITFC themselves are also a response to atmospheric variations; probably they involve the global atmospheric patterns and the slower oceanic adjustments of the off-equatorial Pacific and the Indian Oceans as proposed by Godfrey [1989, 1996] or Wajsowicz [1995]. We come back to this issue in section 5. Even if the ITFC leads the SST, the lead is far from constant. Figure 8b shows large differences in amplitude between the associated ITFC and SST peaks. Other processes are taking place at the same time. The SST Niño3 is coupled with the wind anomalies in the central equatorial Pacific, and both keep changing and influencing each other, even if there were no Indo-Pacific connection. The ITFC also keeps changing in time, probably in response to larger-scale variations. All these processes are constantly evolving from one event to the next and make each event unique. The ITFC rise is not leading the SST in 1981–1982, whereas it is well in advance for the other events. The 1993 to 1995 period is quite remarkable with its two small ITFC peaks that happen a few months prior to and after the two aborted warm events of this period. Finally, for the last event the ITFC started to rise as early as 2 years prior to the SST warming. Each event needs to be examined separately with appropriate coupled experiments. In order to do this, one needs to initialize the model with realistic conditions at the western boundary. So let us now examine the ITFC role on the ENSO simulated in the forced context.

5. Forced Simulations With a Prescribed ITFC at the Western Boundary

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References

[28] In this section, the values of the ITFC are prescribed at the western boundary during a model experiment named run.ITFC. Even though the model undergoes some inflow-outflow at its boundary, it neither loses mass nor drifts away to a different state during its time integration, since the prescribed values have a zero-mean transport. Except for the western boundary condition, the configuration of the experiment is identical to the control run. Results of the two experiments are compared and validated with the observations that have not been assimilated in the model.

[29] In a forced context the impact of prescribing the ITFC on the baroclinic ocean is fully explained by the correction of the Kelvin amplitude at the western boundary. It means that the amplitude of the thermocline correction is linearly dependent on the ITFC with a systematic lag as it propagates to the east. It reaches the eastern boundary in 2 months without much frictional damping. In the eastern Pacific, because the temperature of the mixed layer is primarily governed by thermocline displacements, the ITFC is expected to have an impact on the SST. The ITFC also affects the zonal current anomalies which play a role in ENSO by advecting warm waters from the western Pacific [see Picaut et al., 1996; Jin and An, 1999]. So one can anticipate that prescribing an ITFC inflow (outflow) at the western boundary will create positive (negative) SLA and zonal current anomalies along the equator, which means a positive (negative) correction within 2 months of the Niño3 SST index. The latter is expected to simultaneously modify the wind anomaly in the central Pacific by a westerly (easterly) correction.

[30] Details in Figure 9 confirm that this is how the ITFC affects the simulated anomalies. In particular, the SST negative anomalies between late 1983 and mid-1986 correspond to the outflow prescribed between early 1983 and late 1985. Similarly, the warmer SST anomalies in mid-1987 happen a few months after the peak of ITFC inflow in 1986–1987. However, the striking result of Figure 9 is that the ITFC impact on these simulated indices is small. The sea level is affected by no more than 5 cm (Figures 9a and 9b). The RMS differences remain equal to 4 cm. Prescribing the ITFC does not significantly change the Niño3 SST index either (Figure 9c). The amplitude of the impact is similar to that found by Murtugudde et al. [1998]. Indeed, even without any western boundary reflection whatsoever, Trident forced by observed winds reproduces reasonably well the Niño3 index observed during the recent event [Boulanger and Menkes, 2001]. Over 1980–1998 the correlation is 0.84 for the CR and 0.87 for run.ITFC; the RMS difference remains the same (0.5°C). Finally (Figure 9d), prescribing the ITFC increases the amplitude of the wind peaks simulated in 1987, 1988, 1992, and 1997 such that they are closer to the observed ones. Overall, the correlation is 0.88 for CR and 0.91 for run.ITFC; the RMS difference is 0.006 Pa and 0.007 Pa, respectively. Given the level of uncertainty in sea level, SST, and wind data, the impact of the ITFC on these indices is negligible.

image

Figure 9. Time series of (a) SLA over Niño West, (b) SLA over Niño3, (c) SST over Niño3, and (d) TX over Niño4. Solid line is for observations, dotted line is for the control run, and dashed line is for run.ITFC.

Download figure to PowerPoint

[31] These results are likely to be specific, though, to the forced context. The small differences between the two experiments are explained by the fact that the ENSO signal in the Pacific is primarily governed by the zonal wind stress anomalies, which are identical in these two experiments. However, in the coupled context, very slight changes of the thermocline can drastically affect the results by putting the model in a self-sustained oscillation mode or in a stable regime. One of the critical aspects that control the coupled behavior is the position of the minimum of sea level variability as a function of longitude along the equator. As was shown in section 4, this position for the control run is not at the date line where it is observed, but in the central Pacific. Figure 10 shows that prescribing the ITFC significantly reduces this model deficiency. It also reduces the simulated variability in the west and increases it in the east, thus in better agreement with observations. In both experiments, the simulated minimum is located to the east of the observed one because the model overestimates the variability due to free and reflected waves in comparison to the wind-driven ones. This has been demonstrated by Perigaud and Dewitte [1996] by testing various friction and reflection coefficients.

image

Figure 10. RMS variability of SLA over 1980–1998 as a function of longitude along the equator.

Download figure to PowerPoint

[32] Let us now explain why adding the ITFC values displaces the minimum to the west. In the case of a warm event at a given time in both experiments, the SLA forced by a westerly wind anomaly in the central Pacific is positive in the east and negative in the west, and the reflection of the negative Rossby SLA into the negative Kelvin SLA contributes to even more negative SLA in the west and to less positive SLA in the east. So with the same wind, prescribing an eastward ITFC corresponds to reflecting less negative SLA. The signal is therefore less negative in the west, and after propagation it is more positive in the east, thus decreasing the variability in the west, increasing it in the east, and locating the minimum less to the east of the forcing region. This positive impact is an additional argument indicating that our ITFC estimate corresponds to an actual process missed by the model when the boundary is closed.

[33] Another critical quantity involved in the coupled behavior is the SLA averaged all across the Pacific between the western boundary and the American continent in the equatorial band (5°S–5°N). This quantity plays a key role in the model oscillations as it needs to be recharged by the off-equator [Jin, 1997; Cassou and Perigaud, 2000] (see Appendix A). Its variations are presented for the two experiments and the observations over 1980–1998 (Figure 11a). The observed RMS amplitude over 1980–1998 is 2.7 cm. The model with closed boundary has much weaker variations. The ITFC has a significant impact on this signal in increasing the RMS amplitude of the simulated signal from 1.3 cm to 2.5 cm. Also, the strong anomalies observed in 1982, 1983–1984, 1988–1989, 1991–1992, and 1996–1997 are fairly well reproduced by the ITFC experiment. Note that as in Figure 9a, the signal simulated with the ITFC tends to lag the observed signal. The correlation with observations is 0.68 for CR and 0.69 for run.ITFC, and the RMS difference is 2.1 cm and 1.8 cm, respectively.

image

Figure 11. Time series of SLA averaged over the equatorial Pacific (130°E–80°W, 5°S–5°N). (a) Plots derived from observations (solid line), from the control run (dotted line), or from run.ITFC (dashed line). (b) SLA average difference between model outputs in centimeters and ITFC transport in sverdrup. (c) Observed SLA average in centimeters and ITFC transport in Sv.

Download figure to PowerPoint

[34] As expected, the difference between the run.ITFC and the control run is well correlated with the ITFC (0.75), with the ITFC leading the sea level correction (Figure 11b). In addition, Figure 11b shows that the difference between the two closed boundary experiments is small in comparison to the correction brought by the ITFC. The experiment forced by FSU simulates a basin-averaged sea level which is very similar to the control run and poorly agrees with data. A surprising result is found by comparing the observed sea level with the ITFC (Figure 11c). The two time series are positively correlated (0.72), but the ITFC does not lead the basin-averaged sea level; it is the latter which leads the ITFC. The correlation is equal to 0.78 with a lead of 2 months for the observed sea level. Thus, in reality, the ITFC does not control the basin-averaged equatorial sea level as it does in this forced experiment. This detail has important consequences for ENSO forecasts and will be examined in the next section. It is also consistent with the fact that the ITFC involves large-scale connections of the equatorial Pacific with the rest of the world ocean.

[35] Finally, let us examine the zonally averaged SLA outside the equatorial band. As expected, for both experiments, the model performance is poorer than in the equatorial band. In the north (Figure 12a) the correlation between model and data is not significantly improved by the ITFC (from 0.42 to 0.47). This is because the north level is highly influenced by the winds in this band. Indeed, it is found that the Astat (CAC) wind misses the wind variability which is not retained by the singular value decomposition, and the experiment forced by FSU recovers pretty well the level changes of the north. In the south (Figure 12b), the ITFC improves the correlation between the model and the data from 0.21 to 0.44. It significantly improves the sea level of the south because the latter is relatively more dependent on the boundary conditions than on the wind. It is found that run.FSU does not simulate the south better than the control run. Observations reveal that the sea level south of the equator is dominated by a regular increase between 1983 and 1997. This trend is completely missed by the two model experiments which have a closed boundary. It is quite remarkable that prescribing the ITFC introduces a trend in the south which also corresponds to a regular increase between 1983 and 1997. It appears that the observed interdecadal trend in the south could be associated with the Indo-Pacific oceanic connection. This could well take place in reality and needs more investigation.

image

Figure 12. Time series of SLA averaged over the Pacific, from 130°E to 80°W (a) from 5°N to 10°N and (b) from 5°S to 10°S. Plots are derived from observations (solid line), from run.ITFC (dashed line), or from the difference between run.FSU and the control run (dotted line).

Download figure to PowerPoint

6. Impact of the ITFC on Coupled Simulations

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References

[36] In this section, we compare various series of two-yearlong forecasts. A first series is generated with the standard configuration, each forecast being initialized with the conditions delivered by the control run at some month out of the 204 months between 1980 and 1998. This series generated by the model with closed boundary is used as our reference series for the rest of this paper. Overall, the predictive skill of Trident.Astat is similar to that of Tsub.Astat [see Perigaud et al., 2000a]. This means that, statistically, over the 204 cases and for all lead times the model predictions in SST are of similar quality to those of ZC, and the predictions in sea level and wind are better. It also means that for lead times larger than 6 months, and in the cases of the 1982–1983 and 1997–1998 events the model fails to predict the warming. For example, the forecasts initialized prior to May 1982 or prior to March 1997 miss the warm events. If initialized past these dates, Trident.Astat is successful in predicting the warm growth, reversal, and termination of the warm events. Trident.Astat is thus no less skilled than the other models used for ENSO forecasts in our community. Like Tsub.Astat, Trident.Astat is more successful in predicting the 1986–1987 event than the other events. These results are illustrated with the April 1982 and April 1986 forecasts in Figure 13.

image

Figure 13. Examples of forecasts of SST Niño3 indices for the model with closed boundary. (a and b) Comparison of forecasts initiated with the CR (dotted) and run.ITFC (dashed). (c and d) Comparison of forecasts with a weak coupling coefficient (1.0, dotted) or a strong one (1.8, dashed). Observed indices correspond to the solid lines. Unit of measure is °C.

Download figure to PowerPoint

[37] A second series of forecasts was generated to examine the impact of the ITFC on forecasts by using the conditions delivered by the forced experiment run.ITFC instead of the control run. It is known that forecasts can be significantly affected by very slight modifications of their initial conditions. For example, changing the initial SST by a few tenths of a degree as proposed by Chen et al. [1995] can improve some of the ZC SST predictions by more than 3°C. This is not the case for Trident.Astat. Taking the initial conditions of run.CR or run.ITFC never significantly affects the forecasts. In particular, the model in April 1982 fails to predict the warm events, whether it is initiated with or without the ITFC (Figure 13a). The initialization with the ITFC does not degrade either the April 1986 predictions (Figure 13b). These results make sense. As shown by Perigaud et al. [2000a], initializing the model with sea level data in addition to the SST data does not significantly affect the forecasts because the control run is already remarkably consistent with the observed sea level, wind, and SST data. So prescribing the ITFC only in the initialization of the model does not improve the forecasts. The forecasts initiated prior to April 1982 or March 1997 all fail to predict the big events.

[38] Note that increasing the coupling coefficient does not improve the 1982 (or 1997) predictions, whereas it significantly increases the amplitude of the 1986 predictions (Figures 13c and 13d). In these examples the coefficient is equal to 1.8 instead of 1.0 in the standard configuration. This choice is indeed not unreasonable if one wants to compensate for the many energy sources missing in this simplified model (i.e., the high- energy wind fluctuations or the positive feedback of latent heat exchanges). Obviously, the correct answer for the 1986 case is in between (the prediction with a coefficient equal to 1.4 is the most realistic one; not shown here). Although this is not visible in Figure 13d because the 1986 event is mild, the reader should be aware that increasing the coupling coefficient lengthens the duration of the simulated event in case of a big event. Thus the message illustrated by Figures 13c and 13d is twofold. On the one side, for those cases where the model does predict a warm event, the amplitude and phase of the predicted event are significantly sensitive to the coupling strength. This must be kept in mind for the tests presented in this section. On the other hand, no matter how strong the coupling coefficient is, there are times when the model stays in quasi-normal conditions and misses the big events.

[39] The next series is obtained by prescribing the ITFC during the forecast time stepping instead of specifying it only in the initial conditions. To make the comparison easier, all forecasts presented below are initialized with the control run. In Figures 14a, 14c, and 14d, the reader can see examples where the impact of the ITFC on the predicted SST is very large, much larger than in the forced context. The reader can also see an example where the impact is small (Figure 14b), even though this is a case where the ITFC anomalies are among the strongest (see Figure 8b). It is certain that the role of the ITFC on the ENSO forecast is not a linear function of the ITFC. Forecasts depend on the ocean preconditioning in addition to the phase and amplitude of the ITFC itself. Let us first examine a few examples initiated at various times. For each example, we present the two forecasts with the ITFC that have been obtained either with the standard coupling coefficient (called “weak”) or the latter equal to 1.8 (called “strong”).

image

Figure 14. Examples of forecasts of SST Niño3 indices for the model with closed boundary (dotted) or with the ITFC prescribed during the forecast (dashed). Forecasts begin in (a) December 81, (b) May 83, (c) April 86, and (d) December 96. The two dashed lines are obtained with a weak coupling coefficient (1.0) or a strong one (1.8). Observed indices correspond to the solid lines. Unit of measure is °C.

Download figure to PowerPoint

[40] For the forecast initiated in December 1981 (Figure 14a), the index stays slightly negative during the 2 years, whereas with the ITFC the model predicts a warm event of almost 3°C in mid-1983. Indeed, all of the nine forecasts initiated between July 1981 and April 1982 miss the warm event with closed boundary conditions, and all of them predict the onset of a warm event with the ITFC. The predictions are not perfect, of course. With the standard coupling coefficient the predicted peak is too small and late by 6 months. With the strong coupling coefficient, the forecast is significantly better. It is quite remarkable that even with the weak coefficient, the ITFC impact on the forecast is large (2°C), considerably larger than in the forced context (0.5°C). This is all the more unexpected as the ITFC anomalies during this forecast have a small amplitude to the east. The forecasts initiated between May 1982 and December 1983 are successful without the ITFC to predict the growth, peak, and decay of the warm event and the return to quasi-normal conditions. Forecasts are presented in Figure 14b for the case initiated in May 1983 because the consecutive 24 months cover a period when the ITFC anomalies are among the strongest to the west. For all cases, the forecasts are reasonably good. They are slightly negative in 1984 and do not erroneously predict a cold event when the westward ITFC is applied. The impact in this case is not much larger than in the forced experiment, not even for the case of strong coupling coefficient. Similar results are found in 1988–1989 when the ITFC is strong to the west, the SST is cold, but the ITFC impact is <1°C (not shown).

[41] The model without ITFC predicts reasonably well the 1986–1988 period. The case initialized in April 1986, which is highly sensitive to the coupling coefficient without ITFC, is examined here with the ITFC. While the warm event predicted with closed boundary decays to normal too soon (see Figure 13d), prescribing the ITFC significantly enhances the predicted warm event in amplitude and duration (Figure 14c). For the case of weak coupling coefficient, the model with the ITFC predicts a warm peak close to the observed one in August 1987. For the case of strong coefficient, the model predicts two warm peaks in January 1987 and in January 1988. This is consistent with the ITFC being eastward for the whole 1987 year. In both cases, the impact is much larger than in the forced experiment.

[42] Finally, the model initialized in December 1996 which misses the big recent event is able to predict a strong warming when the ITFC is prescribed (Figure 14d). Its success is reasonable; the warm peak is predicted without significant delay even for the case of weak coupling. This is consistent with the ITFC anomalies being to the east and fairly strong in 1997. Indeed, even before December 1996 and as early as March 1996 (not shown), the ITFC is eastward and strong, and the model predicts a warm event with a peak in winter 1997–1998. In the March 1996 case, however, the model predicts a slow and regular increase during 20 months from 0°C to 3.5°C peak. The latter 20-month lead prediction matches the observed warm peak in fall 1997, but it does so for erroneous reasons: The observed warming did not start before winter 1997.

[43] In all the cases that we checked, we found that prescribing the ITFC can have a negligible (less than 0.5°C) or a big (more than 1°C) impact, and that in the latter case the ITFC significantly improves the predictions. Indeed the forecasts initiated prior to the 1991–1992 or prior to the 1993 warm anomalies (not shown) are significantly improved. However, giving general statistics about the forecasting performance of the model with or without ITFC is not very informative and can be misleading for reasons illustrated by Perigaud et al. [2000a]. Rather than delivering overall statistics, understanding that each forecast is a unique case brings useful information. The impact of the ITFC during the forecasts strongly depends on initial conditions in addition to what is prescribed at the western boundary. The rest of this section explains why the ITFC impact can be strong in the coupled context, in contrast to the forced context.

[44] In both contexts, the impact on the predicted SST is certainly consistent with the sign of the ITFC prescribed during the previous months, and the mechanisms are the same ones as those explained in the forced context: prescribing an inflow in the west slightly deepens the thermocline by a quantity that propagates as a free Kelvin wave to the eastern Pacific and adds up to the wind-induced displacement of the thermocline. The only difference with the forced context is that the wind is now generated by the simulated SST instead of being prescribed from the observed SST. Thus the coupled feedbacks between the thermocline, the SST, and the wind render the model very sensitive to very small displacements of the thermocline. The deeper the thermocline, the warmer the SST, the stronger the westerlies, and the deeper the thermocline. The relationship between an ITFC anomaly, a thermocline change, and an SST change is far from being linear in a coupled context. It all depends whether the model state is close to an unstable mode or not. The authors do not wish to provide a criterion of stability as a function of the thermocline depth since much more than the latter is involved in determining the limits of the stability regime. As highlighted by several authors [Jin and An, 1999; Cassou and Perigaud, 2000], zonal advection has a positive feedback on the coupled system as well. Indeed, for the case where a statistical atmosphere is used as in the present study, even the meridional wind feedback must be taken into account to explain why the model is sensitive to very slight changes of the thermocline depth in the coupled context, whereas it is not in the forced context [Cassou and Perigaud, 2000]. The reader can find in this reference a detailed description of the various terms involved in the SST equation in the forced or the coupled experiments. With or without ITFC, if the coupled model simulates a warm growth, the mechanisms are the same. The point here is not to give these explanations again, it is rather to highlight that the ITFC variations can significantly contribute to the growth of a warm event. These results suggest that they should not be neglected in the models used for forecasting ENSO.

7. Summary and Perspectives

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References

[45] This study examines the impact of the Indo-Pacific oceanic connection on ENSO using a Pacific ocean-atmosphere model and data between January 1980 and June 1998. Part of this paper is focused on the estimate of the flow across the Pacific western boundary and on its validation with the ENSO signals simulated by the model in a forced context. Results confirm that the variations of this flow are linked to the ENSO signal of the equatorial Pacific. It also confirms that they depend on other factors in the off-equatorial Pacific and Indian Oceans as well. Indeed, one needs a better understanding of these factors to make progress in ENSO forecasting. This suggests the need for more data acquisition (such as measurements across the opening of the Pacific Northwestern boundary) and for model improvements (such as adding baroclinic modes). This study is a first investigation of the role of the Indo-Pacific oceanic connection on the coupled ocean-atmosphere. Given the data uncertainty and the extreme sensitivity of coupled models to initial conditions, parameterizations, and off-equatorial processes, our choice is to rely on sea level data, in particular the accurate TOPEX/Poseidon data set, and on models with physics simplified as much as possible.

[46] It has been mentioned several times in the literature that a proxy to the Indonesian throughflow variations can be monitored from sea level in the southeastern Indian Ocean, or by taking its difference with the sea level in the western Pacific. It is found here that the geostrophic transport across 130°E (2.5°N–7.5°N) in the western Pacific is positively correlated with these proxies. All proxies show significant discrepancies though. The western Pacific geostrophic transport is the estimate retained here to correct the model closed boundary conditions. The transport correction due to the Indo-Pacific connection (ITFC) is obtained by removing from it the component simulated by the wind-driven and fully reflected equatorial waves of the Pacific. It is found that the ITFC correlates better with the SST Niño3 index (0.7) than the transport before removal (0.3), and it is not very sensitive to the differences between the two wind-forcing fields that were tested. More importantly, the ITFC transport is validated by the outputs of run.ITFC. It reduces the model data misfit for quantities that are away from the western boundary. In better agreement with data, the ITFC weakens the sea level variability in the west, increases it in the east, and puts the minimum to the west of where it is simulated with closed boundary conditions. Also, the basin-averaged equatorial SLA, which does not significantly depend on the wind used to force the model, is significantly improved by the ITFC. It is also found that the interdecadal trend observed in the south is partly recovered by prescribing the ITFC.

[47] Most of the models used for ENSO forecasts ignore the mechanisms associated with the Indo-Pacific oceanic connection. In the forced experiments, because the wind is controlled by observations, the ITFC has a negligible impact on the gross parameters of the model such as the SST Niño3 index. However, there is no such control in a coupled context. Results of this paper demonstrate that the ITFC plays a key role in forecast experiments. Thus prescribing the ITFC inflow in 1981–1982 makes the model predict the onset of a warm event, whereas without the ITFC the model stays in a quasi-normal state during the 2 years of forecast. The ITFC impact on the SST in the coupled system is far from being linear, and it is important to avoid general statements. Statistically speaking, there is a positive correlation between ENSO indices and the decrease of the transport from the Pacific to the Indian Ocean.

[48] One may want to apply a rule such as reducing the coefficient of reflection into Kelvin at the western boundary [see Clarke, 1991] to forecast the magnitude of the ITFC. Then one would be able to deliver forecast in a fully predictive context. However, it is not a wise path to follow. The coupled model with a systematically reduced coefficient will easily enter a growth mode where it drifts away from reality by losing or gaining more and more mass and heat, depending on the initial conditions and the parameterization. In addition, rules derived from the Pacific-only model do not take into account the fact that the ITFC is due to other factors than the sole Pacific equatorial wind. The present results highlight the uniqueness of each event, when one looks beyond the first gross features that are well described by theory. Perspectives, rather, consist in further improving the ITFC estimate and the model performance in each case. For example, in the 1981–1982 case, the onset predicted in December 1981 by Trident.Astat with the ITFC is late compared to the observations, whereas in the 1997–1998 case, the onset predicted in October 1996 (not shown) is too early. The model is obviously missing some other physics. We know that adding westerly wind bursts (WWB) greatly improves the 1981–1982 and the 1997–1998 forecasts and does not degrade the 1986–1988 forecasts, because the impact of WWB on the coupled system depends a lot on the charge of the equatorial ocean [Perigaud and Cassou, 2000]. This charge, as demonstrated here, is certainly improved in Run.ITFC. It needs to be further improved, because observations indicate that, in reality, it does not lag but tends to lead the ITFC. In any case, making progress in implementing both WWB and ITFC in coupled models are key steps toward understanding how coupled systems work.

[49] In the forecast community one should no longer neglect the role of the throughflow variations. In both contexts the ITFC variations modify the index via the same mechanisms, and the Kelvin wave amplitude is modified at the western boundary. It adds to the wind-driven component as it propagates to the east, and it affects the position of the thermocline in the east and therefore the SST. However, this Kelvin correction has a much stronger impact when it puts the model in a state where the coupled feedbacks between the SST, the wind, and the thermocline grow as in a warm event. The fact that the ITFC easily shifts Trident.Astat from a stable state to an unstable growth is a priori not limited to the coupled models with simplified physics.

Appendix A: A Summary of Results Obtained by Studying Various ICMs

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References

[50] One of the most widely used ICMs is the model described by ZC. This ICM is called “CZ.” Perigaud and Dewitte [1996] use satellite and in situ observations of the 1980s–1990s to validate the CZ model outputs. They use the CZ code provided by the authors and force its ocean component by FSU winds, then force its atmosphere component by the CZ SST results. They thus identify some important misbehavior in the zonal currents which reverse every 9 months, the SSTs which miss the cold events, and the zonal wind stresses which are located up to 40°E of where they should be.

[51] Perigaud and Dewitte [1996] introduce three “fixes” to CZ. They damp the 9-month oscillation by increasing with a factor 4 the ocean friction which is (3 year)−1 in CZ. They use XBT data to improve the parameterization of the subsurface temperature at 50 m (Tsub) as a function of thermocline depth. They replace the CZ atmospheric component by a statistical relationship (Astat) between the SST and the wind stress anomalies. The results of the three fixes in a model called “Tsub.Astat” are greatly improved in a forced context.

[52] Perigaud et al. [2000b] explore the effects of each of these fixes in a coupled mode. In this validation process, they find out that neither CZ nor any of the modified versions of this model behave like a delayed oscillator. The CZ model does not oscillate if the off-equatorial winds or the Rossby meridional modes superior to 5 is filtered out. A key process necessary to sustain the oscillatory behavior is the heat recharge of the equatorial ocean by the off-equator, as described by the “recharged oscillator paradigm” [ Jin, 1997]. The major problem found in CZ is that the off-equatorial recharge in the north has the wrong sign: After warm events, the north is discharged, whereas observations indicate that it should be recharged. The reasons for this failure are most visible in the wind: CZ simulates strong easterlies off-equator in the eastern Pacific that are not observed. Cassou and Perigaud [2000] find that the Tsub.Astat model which combines the three fixes behaves much more realistically. Tsub.Astat is also sensitive to the off-equatorial processes. In particular, it simulates a recharge in the north after warm events, because it simulates northerlies in the central and eastern Pacific north of the equator as in the observations. They also find that the more to the east the westerlies are located, the stronger the warm events, whereas cold events do not depend on this location. Finally, Perigaud et al. [2000a] analyze various initialization procedures to improve the skills of these ICMs in predicting ENSO. They find that forecasts with a lead time larger than 6 months are very sensitive to the off-equatorial recharge processes.

[53] Initially, starting from the same equations as CZ, Boulanger [2001] and Boulanger and Menkes [2001] write a new code for the ocean dynamics and for its mixed layer thermodynamics, called Trident. They implement more accurate numerical schemes, including a finer grid resolution and a more realistic land-sea mask. They reestimate the climatological fields as well as the various parameterizations, on the basis of observations during the 1970s and 1980s. They add a second baroclinic mode which can be activated or not. They perform experiments over 1993–1998 where Trident is forced by the wind stress estimated from scatterometer and in situ data and analyze the model performance in reproducing the observed sea level, current, and SST anomalies. Boulanger [2001] designs a spatially variable horizontal friction coefficient in the momentum equation and finds that the latter considerably improves the simulation of current anomalies by damping the propagation of long Rossby waves which are often overestimated in linear ocean models. Boulanger and Menkes [2001] add a term of vertical diffusion for temperature in the mixed layer and find that the latter significantly improves the temperature simulations in the eastern Pacific.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References

[54] The research described in this paper was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. The authors thank Victor Zlotnicki and Akiko Ayashi from the Jet Propulsion Laboratory, Pasadena, for providing the processed and gridded TOPEX/Poseidon data, and Neville Smith from Bureau of Research in Meteorology and Climatology, Melbourne, for providing the fields of subsurface temperature over the Indian and Pacific Oceans. They also thank Jean-Philippe Boulanger from Laboratoire d'Oceanographie Dynamique et de Climatologie, Paris, for providing the Trident model and for his input and corrections in the manuscript. Finally, they thank Keith Rodgers from LODYC for improving the manuscript.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Sea Level Data and Geostrophic Transport Across the Western Pacific Boundary
  5. 3. Model and Control Run
  6. 4. Estimation of the Flow Across the Model Western Boundary Due to the Indo-Pacific Connection
  7. 5. Forced Simulations With a Prescribed ITFC at the Western Boundary
  8. 6. Impact of the ITFC on Coupled Simulations
  9. 7. Summary and Perspectives
  10. Appendix A: A Summary of Results Obtained by Studying Various ICMs
  11. Acknowledgments
  12. References