Geophysical Research Letters

Improvements in the CMIP5 simulations of ENSO-SSTA meridional width

Authors

  • Wenjun Zhang,

    1. Key Laboratory of Meteorological Disaster of Ministry of Education, College of Atmospheric Sciences, Nanjing University of Information Science and Technology, Nanjing, China
    2. Department of Meteorology, School of Ocean and Earth Science and Technology, University of Hawaii at Manoa, Honolulu, Hawaii, USA
    Search for more papers by this author
  • Fei-Fei Jin

    Corresponding author
    1. Department of Meteorology, School of Ocean and Earth Science and Technology, University of Hawaii at Manoa, Honolulu, Hawaii, USA
      Corresponding author: F.-F. Jin, Department of Meteorology, School of Ocean and Earth Science and Technology, University of Hawaii at Manoa, 2525 Correa Rd., Honolulu, HI 96822, USA. (jff@hawaii.edu)
    Search for more papers by this author

Corresponding author: F.-F. Jin, Department of Meteorology, School of Ocean and Earth Science and Technology, University of Hawaii at Manoa, 2525 Correa Rd., Honolulu, HI 96822, USA. (jff@hawaii.edu)

Abstract

[1] The recent study demonstrated the existence of a systematical narrow bias in the simulated El Niño-Southern Oscillation (ENSO) meridional width of surface temperature anomaly (SSTA) of ENSO by the models participating in Phase 3 of the Coupled Model Inter-comparison Project (CMIP3). The current models developed for Phase 5 of the CMIP (CMIP5) still have this narrow bias in ENSO width relative to the observation, but with a modest improvement over previous models. The improvement can partly be attributed to a better simulation in trade wind, and partly to a better simulation in ENSO period. It has also been demonstrated that the models with a better performance in ENSO width tend to simulate the precipitation response to ENSO over the off-equatorial eastern Pacific more realistically.

1. Introduction

[2] The accuracy of coupled models in simulating El Niño-Southern Oscillation (ENSO) phenomenon, including such basic features as ENSO period and amplitude, serves as one of useful tests of climate model's performance. ENSO behaviors in coupled models have been widely evaluated [e.g.,Neelin et al., 1992; Delecluse et al., 1998; Latif et al., 2001; Davey et al., 2002; AchutaRao and Sperber, 2002, 2006; Capotondi et al., 2006; Guilyardi, 2006; Zhang et al., 2010]. Although a steady progress has been made, coupled models still exhibit large biases in modeling the basic features of ENSO [e.g., Guilyardi et al., 2009].

[3] Among these biases, earlier studies have mentioned that ocean models produce sea surface temperature (SST) anomaly (SSTA) too tightly confined to the equator [e.g., Stockdale et al., 1998; Kang et al., 2001]. Recently, the meridional structure of ENSO SSTA was further investigated and it was shown that the ENSO meridional width in the models participating in Phase 3 of the Coupled Model Inter-comparison Project (CMIP3) is only about two thirds of what is observed [Zhang et al., 2012]. The systematical narrow bias in ENSO width is attributable partly to a weak ocean meridional current which transports less effectively the equatorial SSTA towards the off-equator, and partly to a short ENSO period which allows less time to spread the SSTA away from the equator. Does the systematical narrow bias in ENSO width still exist in current models developed for Phase 5 of the CMIP (CMIP5)? Is there any improvement in the ENSO width? In this study, we assess the ENSO meridional width simulated by the CMIP5 models and compare them with the previous simulations from the CMIP3 models. The impacts of ENSO width on precipitation over the eastern tropical Pacific are also discussed.

2. Data and Methods

[4] We chose to use 15 coupled models from the CMIP5 in the “historical” simulations, which were driven by historical climate forcing including changing atmospheric composition, solar forcing, and land use from 1850 to 2005. The 15 CMIP5 models are CanESM2, CNRM-CM5, GFDL-ESM2M, GISS-E2H, GISS-E2R, HadCM3, HadGEM2-CC, HadGEM2-ES, INM-CM4, IPSL-CM5A-LR, IPSL-CM5A-MR, MIROC5, MPI-ESM-LR, MRI-CGCM3, and NorESM1-M. For comparison with the CMIP5 simulations, 15 coupled models were also chosen from the CMIP3 in the 20th century run (20C3M). These 15 models include BCCR-BCM2.0, CCSM3, CNRM-CM3, ECHAM5-MPI, ECHO-G, FGOALS-g1.0, GFDL-CM2.0, GFDL-CM2.1, GISS-EH, INGV-SXG, IPSL-CM4, MRI-CGCM2.3.2, PCM1, HadCM3, and HadGEM1. Model references and details can be obtained from the CMIP3 and CMIP5 website athttp://www-pcmdi.llnl.gov/ipcc/about_ipcc.php and http://cmip-pcmdi.llnl.gov/cmip5, respectively. All the models selected can reasonably simulate ENSO features, although some models, such as the FGOALS-g1.0, exaggerate the ENSO amplitude. The data variables used in this study are monthly mean fields including SST, zonal wind stress, and precipitation. The zonal wind stress is available in 13 CMIP3 and 10 CMIP5 models. The simulations over the period of 1900–1999 were used to analyze the meridional width of ENSO and its impact on precipitation.

[5] The observation-based data are monthly SST (1900–1999) from the Hadley Center Sea Ice and Sea Surface Temperature data (HadISST) [Rayner et al., 2003]. The zonal wind stress and precipitation is taken from the Simple Ocean Data Assimilation (SODA) data set (1900–1999) [Carton et al., 2000; Giese and Ray, 2011] and the Climate Prediction Center Merged Analysis of Precipitation (CMAP) (1979–1999) [Xie and Arkin, 1996], respectively. A 2–7-year band-pass filter is applied to the SSTA since the ENSO variability is focused.

3. Results

[6] To examine the spatial patterns of SSTA of ENSO, Figure 1 shows the tropical SSTA patterns regressed upon the SSTA over the central and eastern equatorial Pacific (3°S–3°N, 180°–90°W) in the observation and in the CMIP3 and CMIP5 models (Figure 1). In the observation, a typical SSTA pattern of ENSO shows its maximum center at the equator with decay towards both sides of the equator. The SSTA extends to about 13°N and 20°S based on a 0.2 contour or to about 7°N and above 10°S based on a 0.4 contour. It is interesting that the SSTA extends farther to the south than to the north. The SSTA pattern of ENSO simulated by both generations of the models resembles the observation to a significant degree. However, the previously mentioned biases in the CMIP3 models [e.g., Capotondi et al., 2006], still remain in the CMIP5 models. For example, the SSTA variability extends farther westward at the equator. As for the meridional width of ENSO, the CMIP3 and CMIP5 models both show a narrow bias compared to the observation based on either a 0.2 or 0.4 contour. The SSTA in the CMIP3 models spreads only towards about 6° (5°) N and 10° (7°) S based on the 0.2 (0.4) contour. The CMIP5 models tend to simulate a wider SSTA of ENSO than the CMIP3 models. The SSTA reaches about 7° (6°) N and 15° (7°) S based on the 0.2 (0.4) contour. The significant improvement appears mainly over the southeastern tropical Pacific (see Figure S1 in the auxiliary material). To clearly illustrate the meridional scale of ENSO, Figure 1d shows the meridional structure of SSTA averaged over 180°–90°W. Within 5 degrees latitude of the equator, the simulated SSTA structure is almost same as the observation; however, outside of 5°N and 5°S, the SSTA decreases with increase of latitude at a much faster rate in the models than in the observation. Compared to the CMIP3 models, the decreasing rate of SSTA out of the equator is slightly slower in the CMIP5 models. It is notable that the models show a large diversity in the meridional structure of SSTA indicated by the spread (a standard deviation).

Figure 1.

SSTA pattern regressed upon SSTA over the central and eastern equatorial Pacific (3°S–3°N, 180°–90°W) in the (a) HadISST data and in multi-model ensemble from the (b) CMIP3 and (c) CMIP5 models. Contour interval in Figures 1a–1c is 0.2. (d) Meridional pattern of zonal mean (180°–90°W) SSTA in Figures 1a–1c for the HadISST data (black line), the CMIP3 models (blue line), and the CMIP5 models (red line). The dashed blue and yellow lines denote the spread of the CMIP3 and CMIP5 simulations, respectively.

[7] To quantitatively measure the ENSO width, an e-folding width of ENSO is defined as the width by which the meridional profile of ENSO SSTA reaches 1/e. The meridional profile is calculated based on the same method as that used inFigure 1d. Based on this definition, the ENSO widths for the observation and the CMIP3 and CMIP5 models are shown in Figure 2. The observed meridional width of ENSO is about 18 degrees of latitude, whereas, the simulated ENSO widths in all models are narrower than the observation. Different models display very different behaviors in simulating the ENSO width. Some models, such as the two IPSL-CM5A and two HadCM3 models, simulate a relatively large meridional width for ENSO, whereas some other models, such as the ECHO-G, FGOALS-g1.0 and GISS-EH models, simulate a very small ENSO width. The ensemble mean of ENSO width is 13 and 14 degrees of latitude in the CMIP3 and CMIP5 models, respectively; and their difference is statistically significant at the 0.1 confidence level using a Student'st-test. Five out of 15 CMIP3 models have the ENSO width less than 12 degrees of latitude, whereas only one of 15 CMIP5 model (GISS-E2H) has such an extremely narrow width. Even though the ENSO width simulated by the GISS-E2H model is only 11.5 degrees of latitude, it demonstrates a more accurate behavior compared to its previous version (GISS-EH). Similarly, only two CMIP3 models can simulate ENSO width equal to or larger than 15 degrees of latitude, while five CMIP5 models have this ability. The CMIP5 models show a better performance in the meridional width of ENSO SSTA compared to the CMIP3 models. It should be noted that our measure for ENSO width should be independent of ENSO amplitude. Indeed as shown in Figure S2 in theauxiliary material, the ENSO width and amplitude almost has no correlation.

Figure 2.

The ENSO meridional widths (unit: degree in latitude) for the HadISST (grey), CMIP3 (green) and CMIP5 (yellow). The green and orange lines indicate the standard deviations among the CMIP3 and CMIP5 models, respectively.

[8] Our recent study demonstrated that climatic zonal wind stress and ENSO period are important factors influencing the meridional width of ENSO SSTA through observational and theoretical analyses [Zhang et al., 2012]. Here, these two factors are also examined in the CMIP3 and CMIP5 models. Figure 3a shows the dependence of the ENSO meridional width on the climatic zonal wind stress over the central to eastern equatorial Pacific (5°S–5°N, 180°–90°W). The simulated ENSO width has a significantly negative correlation (−0.46) with the intensity of the zonal wind stress, suggesting that the stronger trade winds favor an occurrence of the wider ENSO SSTA. This is because the stronger trade winds, corresponding to the stronger wind stress curl off the equator (see Figure S3 in the auxiliary material), lead to the stronger meridional currents which more effectively spread the equatorial SSTA away from the equator, a mechanism as noted in previous studies [Niiler et al., 2004; Zhang et al., 2009]. Most models tend to simulate a weak zonal wind stress with respect to the observation. This is one of causes of the systematical narrow bias in ENSO width. Comparing these two generations of the models, the trade winds simulated in the CMIP5 models tend to be slightly stronger than those in the CMIP3 models, which favors a wider ENSO.

Figure 3.

(a) Scatter plot of climatic zonal wind stress over the central to eastern equatorial Pacific (5°S–5°N, 180°–90°W) and the ENSO meridional width in the observation (grey), CMIP3 (blue), and CMIP5 (red). (b) Scatter plot of the ENSO period and the ENSO meridional width. Pluses and circles denote different models and ensemble mean, respectively. The dashed line is the regression line for all models.

[9] ENSO period is another factor that affects the ENSO meridional width. Our recent study demonstrated that a longer ENSO period allows more time to spread the SSTA away from the equator by the meridional advection and thus the ENSO meridional width becomes wider [Zhang et al., 2012]. This is also indicated by a significantly positive correlation between the ENSO period and width (Figure 3b). The ENSO period is estimated from the frequency corresponding to the maximum spectral power of the SSTA over 5°S–5°N, 180°–90°W. Relative to the observation, most models show a shorter ENSO cycle, which tends to cause a narrower ENSO width. The CMIP5 models tend to simulate the ENSO period slightly more realistically than the CMIP3 models, a tendency which contributes to a more realistic ENSO width. Some studies argued that a wide zonal wind stress response to ENSO favors an occurrence of a long ENSO cycle [e.g., Kirtman, 1997; Capotondi et al., 2006]. Therefore, the positive relation shown in Figure 3b may partly reflect the impact of ENSO period on its width.

[10] The meridional pattern of ENSO SSTA deserves more attention because it has a significant impact on climate. Based on the ENSO width in Figure 2, the ten widest and ten narrowest simulations are selected for composite analyses, respectively. The precipitation responses to ENSO in both wide and narrow simulations exhibit a double-intertropical convergence zone (ITCZ)-like pattern: a strong precipitation response off the equator but a weak precipitation response at the equator (Figure 4). The weak precipitation response over the eastern equatorial Pacific may be due to a cold bias and a strong east-west gradient in the simulated background SST [e.g.,Latif et al., 2001; Lin, 2007]. The simulated ENSO SSTA tends to induce two bands of precipitation anomaly off the equator over the equatorial side of the ITCZ and South Pacific convergence zone (SPCZ). This is because SSTA of ENSO can effectively modulate convention and thus precipitation over the warm water underlying the IPCZ and SPCZ. In comparison with the wide ENSO simulation, the precipitation anomalies off the equator shift further towards the equator in the narrow ENSO simulation (Figures 4a and 4b). The differences in precipitation anomaly between the two groups clearly show two belts: one near 5°N elongates zonally from the dateline to 100°W; another over 5°–10°S tilts southeastward and stays to the west of 130°W (Figure 4c). Furthermore, the regression of the precipitation response to ENSO upon the ENSO width shows a similar pattern for the influence of the ENSO width on the precipitation (Figure 4d). Therefore, the ENSO width has an important effect on the off-equatorial precipitation response over the eastern tropical Pacific.

Figure 4.

Composite ENSO precipitation response (mm/d) (a) in ten widest ENSO simulations and (b) in ten narrowest ENSO simulations from all CMIP3 and CMIP5 models. The ENSO precipitation response is derived from the regression upon the central to eastern equatorial Pacific SSTA (5°S–5°N, 180°–90°W). The purple contours indicate the observed precipitation response (mm/d) to ENSO. (c) Difference in ENSO precipitation response between ten widest and ten narrowest simulations. Contour in Figures 4a–4c is 0.5 mm/d. (d) ENSO precipitation response regressed upon the ENSO width in all models. Contour interval is 0.1 mm/d/latitude. The light, moderate, and dark shading in Figure 4d indicates the values above 0.2, 0.1, and 0.05 confidence level (Student's t-test), respectively.

4. Concluding Remarks

[11] The meridional width of ENSO SSTA simulated by CMIP5 models is assessed and compared with that simulated in CMIP3 models. A systematic narrow bias in ENSO meridional width remains in the CMIP5 models, but is reduced compared to that in the CMIP3 models. The improvement in simulated ENSO width is partly due to a stronger trade wind and partly due to a longer ENSO period. The ENSO meridional width deserves attention because it has a significant impact on the ENSO precipitation response over the eastern tropical Pacific. Models with the wide ENSO simulation tend to have a better performance in simulating the ENSO precipitation anomaly over the eastern tropical Pacific than models with the narrow ENSO simulation.

Acknowledgments

[12] This work is supported by National Science Foundation grants ATM 1034798, NOAA grand NA10OAR4310200, DOE grant DESC0005110, and by the National Nature Science Foundation of China (41005049), the Special Fund for Public Welfare Industry (Meteorology) (GYHY201206016), and the research fund of NUIST.

[13] The Editor thanks the anonymous reviewers for their assistance in evaluating this paper.