ADVANCES IN ATMOSPHERIC SCIENCES, VOL. 30, NO. 3, 2013, Western Pacific Warm Pool and ENSO Asymmetry in CMIP3 Models SUN Yan 1,2,3 ( ß), De-Zh

Transcription

1 ADVANCES IN ATMOSPHERIC SCIENCES, VOL. 3, NO. 3, 23, Western Pacific Warm Pool and ENSO Asymmetry in CMIP3 Models SUN Yan,2,3 ( ß), De-Zheng SUN 3, WU Lixin 2 (Çá#), and WANG Fan (fi ) Key Laboratory of Ocean Circulation and Waves, Institute of Oceanology, Chinese Academy of Sciences, Qingdao Physical Oceanography Laboratory, Ocean University of China, Qingdao Cooperative Institute for Research in Environmental Sciences, University of Colorado, and NOAA/Earth System Research Laboratory/Physical Sciences Division, Boulder, CO 835, USA (Received 23 July 22; revised 22 October 22) ABSTRACT Theoretical and empirical studies have suggested that an underestimate of the ENSO asymmetry may be accompanied by a climatologically smaller and warmer western Pacific warm pool. In light of this suggestion, simulations of the tropical Pacific climate by 9 Coupled Model Intercomparison Project Phase 3 (CMIP3) climate models that do not use flux adjustment were evaluated. Our evaluation revealed systematic biases in both the mean state and ENSO statistics. The mean state in most of the models had a smaller and warmer warm pool. This common bias in the mean state was accompanied by a common bias in the simulated ENSO statistics: a significantly weak asymmetry between the two phases of ENSO. Moreover, despite the generally weak ENSO asymmetry simulated by all models, a positive correlation between the magnitude of the bias in the simulated warm-pool size and the magnitude of the bias in the simulated ENSO asymmetry was found. These findings support the suggested link between ENSO asymmetry and the tropical mean state the climatological size and temperature of the warm pool in particular. Together with previous studies, these findings light up a path to improve the simulation of the tropical Pacific mean state by climate models: enhancing the asymmetry of ENSO in the climate models. Key words: warm pool, ENSO asymmetry, CMIP3 model, ENSO time-mean effect Citation: Sun, Y., D. Z. Sun, L. X. Wu, and F. Wang, 23: Western Pacific warm pool and ENSO asymmetry in CMIP3 models. Adv. Atmos. Sci., 3(3), , doi:.7/s Introduction Of all regions in the Earth s climate system, the western Pacific warm pool has one unique aspect that deserves our special attention: it has the highest SST among the world s open oceans (Newell, 979; Ramananthan and Collins, 99). As we are increasingly concerned with whether we have reached the point of no return regarding Earth s climate system (Hansen et al., 28), we are especially concerned with model performance in simulating the warmest ocean region on the Earth. The western Pacific warm pool (hereafter simply the warm pool) has also been referred to as a major heat generator in the Earth s climate system (Pierrehumbert, 995). In this region, tropical deep convection is concentrated and latent heat release reaches a broad maximum (Spencer, 993). The latent heat release powers the Walker and Hadley circulations in the atmosphere, which in turn drive the currents in the upper ocean (Held and Hou, 98; Philander, 99, Webster and Lukas, 992; Trenberth and Solomon, 994; Dijkstra and Neelin, 995; Sun and Trenberth, 998). The atmospheric Hadley circulation and its counterpart in the ocean, the meridional branch of the winddriven circulation, extend the influence of the warm pool to the extratropical region (Hou, 998; Lu et al., 998; Sun and Trenberth, 998). This critical role Corresponding author: WANG Fan, fwang@qdio.ac.cn China National Committee for International Association of Meteorology and Atmospheric Sciences (IAMAS), Institute of Atmospheric Physics (IAP) and Science Press and Springer-Verlag Berlin Heidelberg 23

2 NO. 3 SUN ET AL. 94 of the warm pool in dynamics of the climate system puts additional emphasis on the accuracy of our stateof-the-art climate models when simulating the major characteristics of the warm pool. Previous studies have reported some biases in the mean state of the warm pool in coupled climate models. The earliest study using outputs from multiple models was probably the seminal study by Mechoso et al. (995). They noted that an excessive cold tongue was a common feature among the models they examined, implying that the zonal extent of the warm pool in the models may be too confined in the western Pacific. Kiehl (998) directly examined the warm pool simulation by Community Climate System Model version (CCSM) (Boville and Gent, 998; Kiehl, 998) and found the same bias as seen in Mechoso et al. (995) in other models. Kiehl (998) hypothesized that the excessive solar heating reaching the warmpool may force stronger zonal winds and therefore an extended cold tongue (or a smaller warm pool). These studies, however, examined only a single run of the concerned models. As observations comprise a single realization, it is important to examine the spread of an ensemble of runs to draw a conclusion about the biases in the models. We attempted to evaluate the simulations by the climate models of the warm pool and ENSO statistics in a more thorough way. The climate models are in the Intergovernmental Panel on Climate Change 4th Assessment Report (IPCC AR4) and are also in the Coupled Model Intercomparison Project Phase 3 (CMIP3). We not only examined a large set of models [i.e., all of the no flux adjustment IPCC AR4 (CMIP3) models], but we also analyzed the available ensemble runs of individual models. These outputs allowed us to construct Probability Density Functions (PDFs) using the largest data sets available and put the estimate of bias on a stronger statistical footing than studies that were limited to a single model or to a single run of multiple models. The main motivation for this study, however, was to determine whether recent theoretical and empirical predictions regarding a role of ENSO events in determining the mean state of the warm pool are indeed supported by, or at least consistent with, the results from models. Specifically, theoretical and empirical studies have suggested that if models underestimate the nonlinearity in the ENSO dynamics, then the size of the warm pool should be smaller and the mean warm-pool SST should be greater than the observations (Rodgers et al., 24; Schopf and Burgman, 26; Sun and Zhang, 26; Sun and Yu, 29; Sun, 2, 23, 27, 2). We further verified the simulations by the climate models collected by IPCC AR4 (CMIP3) (Meehl et al., 27) with empirical and theoretical findings. We want to determine whether bias of the time mean state of the warm pool and the bias of ENSO in the models had a consistent relationship with the biases implied by empirical and theoretical results. For this purpose, we also attempted to evaluate the simulations by the climate models in CMIP3 of ENSO statistics (ENSO asymmetry in particular) in a more thorough way. Using the same method, we evaluated the climatologic size and temperature of the warm pool to evaluate the statistics of ENSO events. Many studies have evaluated ENSO asymmetry in climate models previously (An et al., 25; Zhang et al., 29; Sun, 2). However, previous studies have also examined the climatology of the warm pool; these studies evaluating ENSO asymmetry examined only a single run of the concerned models. The number of the models examined was also more limited in previous studies. For example, An et al. (25) employed models. Zhang et al. (29) and Sun (2) focused on the National Center for Atmospheric Research models (NCAR) CC- SMs [see the early versions of Community Earth System Model (CESM); see The collective effect of ENSO events on the mean Pacific climate in general and on the warm pool in particular was first suggested by the asymmetry between its warm phase (El Niño) and cold phase (La Niña). The sum of the two do not cancel each other entirely but result in a spatial pattern resembling the anomaly in the warm phase (Burgers and Stephenson, 999; Rodgers et al., 24; Sun and Yu, 29; Sun, 27). This effect has been regarded as the residual of ENSO in these and related studies. Attempting to address the questions of the time-mean effects of ENSO events more rigorously, Sun and Zhang (26) employed a hybrid model (i.e., an empirical atmosphere coupled with an ocean GCM) to contrast the response of the tropical Pacific mean climate to a perturbation in the presence of ENSO events with a case in which ENSO events are surgically suppressed. The results indicated that ENSO events tend to cool the center of the western Pacific warm pool and warm the central Pacific, thus they effectively extend the size of the warm pool but reduce the mean warm pool SST. Sun (2) further reported some preliminary results from forced ocean model experiments in which the strength of the ENSO fluctuations in the surface winds were varied. The preliminary results seem to confirm the findings of Sun and Zhang (26). The paper is organized as follows. Information about the models and the data sets is provided in section 2. The results regarding the biases in the climatological state of the warm pool and ENSO statistics are presented in section 3. A summary and conclusion

3 942 WESTERN PACIFIC WARM POOL AND ENSO ASYMMETRY IN CMIP3 VOL. 3 Table. Color schemes used to denote the data from different models and observations. The number of runs for each model and origin of the countries of the models used are also listed. The dominance of the United States in climate modeling is apparent. Most of the models, including those of the United States, have a very modest number of runs. Note that only models without the use of flux adjustment are included in this study. are provided in section Data and methodology 2. Observations The observational SST data used in this study was the Hadley Centre sea ice and sea surface temperature data set Version (HadISST, Rayner et al., 23). It was developed at the Met Office Hadley Centre, and the monthly data have been available from 87 to the present. The SST field was built from in situ and satellite observations ( grid resolution). Data used in this study covered the period from January 9 to December Models The model data are the results of the 2c3m scenario simulations by the CMIP3 climate models (also in IPCC AR4), available from ftp-esg.ucllnl.org (Meehl et al., 27) a. The analysis was limited to the no-flux adjustment models. The 9 no-flux adjustment models, together with the numbers of runs each individual model in the CMIP3, are listed in Table. A total of 53 models runs was used in our construction of the statistics of the warm-pool climatology and ENSO characteristics. The entire 2th century (January 9 to December 999) was analyzed, though we focused on the last 5 years in the present paper, as the observational data are more reliable during this latter period. 3. Results 3. Western Pacific warm pool simulation in the models The multi-model ensemble runs allowed us to construct a PDF for the warm-pool size. It is shown as thebluecurveinfig.a.theverticallineinblueina The descriptions of all the models listed are available from documentation/ipcc model documentation.php.

4 NO. 3 SUN ET AL. 943 Fig.. (a) Probability density function (PDF, blue curve) for the climatological annual mean warm-pool size. The red vertical line indicates the observed value, and the blue vertical line indicates the averaged value of all model runs. The short colored marks on the horizontal axis indicate the values of individual runs in all models. (b) Time series for the western Pacific warm-pool size (m 2 ) in observations (black) and 9 IPCC AR4 (CMIP3) models (colors) over the last century. Shown in Fig. b are the multi-run ensemble mean values smoothed by a cosine bell window with a width of 49 months. The warm pool is defined as the region where SST is higher than 28 C. Shown are the results using the data from 95 to 999 period, which are considered more reliable. The results based on the data over the entire 2th century are similar to those shown here. Color scheme for identifying models is provided in Table. dicates the multi-model ensemble mean value, that is, the averaged warm pool size simulated by all model runs. The red line indicates the observed value. The short colored bars on the horizontal axis mark the ensemble mean value of the size of the warm pool in each model. Most of the model runs have a warm-pool size that is smaller than that of the observations (Fig. a). Measured by the ensemble mean value of each model, 75% of the models underestimated the size of the warm pool. The multi-model ensemble mean value of the warm-pool size was only 8% of the size of the warm pool as given by observation data (Fig. a). The PDF is also obviously negatively skewed, suggesting that it is more difficult to increase the size of the warm pool in the models than to decrease it. Figure b shows the time series of the ensemble mean value of warm-pool size simulated by each model over the entire period where model runs were available. The majority models underestimated the size of the warm-pool size and did so throughout the entire century. (The few models identified in Fig. a that had a larger warm pool than the observations also did so throughout the entire period). Redrawing Fig. a using different periods of data showed that the underestimate of the warm-pool size did not depend on whether the data were for the entire 2th century, the last 5 years in the 2th century, or the last 3 years in the 2th century were used in the construction of the PDF. A major contributor to the smaller warm-pool size in the models was the tendency of most models to simulate an excessive westward extension of the cold tongue (Fig. 2). The figure shows the time-mean position of the 28 C SST from models and observations. The ensemble mean SST from each model was used to obtain this figure. Another contributor to the smaller size of the warm-pool is that the main part of the warm pool (west of 6 E) in the models is more confined meridionally to the equator, particularly in the northern hemisphere (Fig. 2).

5 944 WESTERN PACIFIC WARM POOL AND ENSO ASYMMETRY IN CMIP3 VOL. 3 Fig. 2. The climatology of 28 C SST in the models (colors) and observations (black). Shown are results for the period Only the results for the ensemble mean of the models are shown in the figure for clarity. Color scheme for identifying models is provided in Table. The variation in warm-pool size among different runs in the same model was slightly relative to their differences with the observation warm pools. An example in Fig. 3 shows the warm-pool size simulated by all the runs of the same model the (a) NCAR CCSM3. (National Center for Atmospheric Research models Community Climate System Model version 3.) and (b) GFDL CM2. (Geophysical Fluid Dynamics Laboratory Climate Models version 2.). Therefore, the intrinsic errors in individual models are responsible for the model observation discrepancies. Although the size of the warm pool in the models was generally smaller than that in the observations, the warm pool in the models was warmer than that in the observations, measured either by the mean SST over the warm pool (Figs. 4a and b) or its maximum SST (Figs. 4c and d). Figures 4a, b and c, d are respectively the same as Figs. a and b except they are for the mean SST of the warm pool (Figs. 4a and b) and for the maximum SST over the warm pool (Figs. 4c and d). The time series of these quantities (Figs. 4b and d) shows that these discrepancies between the model simulations and observations remain relatively constant over the entire period of the simulations. The model observation discrepancies in the maximum SST are particularly striking given that, in the observation, the maximum SST was almost always 3 C; the variability was within.5 C over the entire century (Fig. 4d). The maximum SST in models spread from 27 Cto 35 C in comparison. Twelve of the 9 models simulated a higher maximum SST. The averaged maximum SST over the tropical Indo-Pacific of all model runs was 3.85 C, which was.4 C higher than the observations (3.44 C, Fig. 4c). Also note that the PDF for the maximum SST was highly positively skewed, suggesting that, in the models, it is easier to increase the maximum SST than to decrease it. This assessment reveals systematic biases in both the size and the mean temperature of the Pacific warm pool. Global coupled models tend to simulate a smaller (i.e., zonally too confined in the west and meridionally too confined in the equatorial band) and warmer western Pacific warm pool. This result confirms the findings from previous inter-model comparison studies that diagnosed an exaggerated westward extent of the western edge of the cold tongue (and thus a warm pool too confined in the west) (Latif et al., 2; AchutaRao and Sperber, 22; Davey et al., Fig. 3. The same as Fig. 2, but for all the runs of the same model: (a) NCAR CCSM3. and (b) GFDL CM2.. Note that the variability among the different runs is small in the same model, but the variability is significant between models and observations.

6 NO. 3 SUN ET AL. 945 Fig. 4. (a) The same as Fig. a, but for the mean warm-pool SST. (b) The same as Fig. b, but for the mean warm-pool SST. (c) The same as Fig. a, but for the maximum SST. (d) The same as Fig. b, but for the maximum SST. 22; Hannachi et al., 23; Kug et al., 2). 3.2 ENSO asymmetry in the climate models Using multi-model ensemble runs, we uncovered a common bias among the models: the warm pool was generally smaller and warmer than the observations. This standout bias provides an opportunity to test the theoretical and empirical predictions regarding the nonlinear rectification effect of ENSO activity on the warm pool. If the simulated ENSO statistics do have a distinct bias, and the bias is in the direction that would cause the bias, we have already uncovered the bias in the climatology of the warm-pool in the manner suggested by the aforementioned empirical studies. These results increased our suspicion of a nonlinear rectification effect of ENSO on the mean state. Toward this end, we evaluated ENSO statistics in the models. In particular, we evaluated ENSO asymmetry because it is a measure of the nonlinearity of ENSO dynamics and therefore is a measure of the time-mean effect (Schopf and Burgman, 26; Sun and Zhang, 26; Liang et al., 22; Sun et al., 22 b ). A typical way to measure ENSO asymmetry is the skewness of Niño3 SST (Burgers and Stephenson, 999; An et al., 25). A PDF was constructed for the skewness of the Niño3 SST in the models and observations in the same way as for the warm-pool size and temperature. This PDF is shown in Fig. 5. (The same multi-model ensemble runs were used for Fig. 5 and Fig. a). The figure reveals that the models generally underestimated ENSO asymmetry. The skewness of Niño3 SST anomaly in the observations was positive (redline in the Fig. 5) in contrast to the negative ones Fig. 5. The PDF (blue curve) for the skewness of monthly Niño3 SST anomaly. Data used to construct this figure are the same as for Fig. a. The color scheme for indicating the models is also the same. The red vertical line indicates the value for observations. The vertical blue line is the multi-model ensemble mean. The short color bars on the horizontal axis mark the values for the individual runs in all the models. b Sun, D.-Z., T. Zhang, Y. Sun, and Y. Yu, 22: Rectification of El Niño Southern Oscillation into climate anomalies of longer time-scales: Results from forced ocean GCM experiments. J. Climate submitted.

7 946 WESTERN PACIFIC WARM POOL AND ENSO ASYMMETRY IN CMIP3 VOL. 3 models (Figs. 7c and d). Therefore, the intrinsic errors in individual models were responsible for the model observation discrepancies. 3.3 Biases in the ENSO asymmetry: A cause of a smaller and warmer warm pool? Fig. 6. The frequency (%) distribution of monthly Niño3 SST anomaly. Only the ensemble mean of the models are plotted. Color scheme for identifying the models is provided in Table. in most of the model runs (Fig. 5). Some model runs exhibited positive skewness, but none reached the observed value. The skewness of Niño3 SST anomaly from the observations was.88 based on 5 years data from 95 to 999. The minimum skewness of ENSO anomalies in the model runs was.88, and the maximum was.8. The averaged skewness of Niño3 SST anomaly in all model runs was.2, which was very close to zero, indicating a near-symmetric ENSO in the no-flux-adjustment CMIP3 models (Fig. 5). To show this common model bias using a more traditional method, we plotted the histogram of the Niño3 SST anomaly distribution (Fig. 6). The symmetric nature of ENSO in the models can be readily seen in this figure. In the observations, the cooling events occurred more frequently than the warming events, and the strongest cooling events were weaker than the strongest warm events. In contrast, there were equal occurrences of cooling and warming events in almost all the models. The modeled ENSO events were very nearly symmetric in magnitude. The modeled ENSO asymmetry biases were noted by Leloup et al. (28) and An et al. (25) in a more limited set of climate models. Zhang et al. (29) noted the underestimation of ENSO asymmetry in the NCAR CCSM models and explored its causes by contrasting the differences among the successive versions of the NCAR CCSM (which is now called CESM). The difference in the frequency distribution of Niño3 SST anomalies among different runs by the same model was also small relative to their differences from the observations. Figure 7 shows the frequency distribution of monthly Niño3 SST anomaly simulated by all the runs of the same model, that is, the GFDL models (Figs. 7a and b) and the NCAR The weak ENSO asymmetry in the models may provide an explanation of the biases in the warmpool simulation in the models. As suggested by the aforementioned empirical as well as theoretical studies (Rodgers et al., 24; Schopf and Burgman, 26; Sun and Zhang, 26; Sun and Yu, 29; Sun, 2; Liang et al., 22; Sun et al., 22 b ), the time mean effect of ENSO in the observations was cooling in the center of the warm pool and warming in the central Pacific. In other words, it reduced the maximum SST over the warm pool and expanded the warm-pool size. Judging from the lack of the asymmetry in the modeled ENSO, such a time mean effect of ENSO was either too weak or was nonexistent in the models. This effect caused a warmer bias in the maximum SST and the mean SST over warm pool but a smaller warm-pool size in the models. In the CMIP3 models, stronger ENSO activity measured by variance tended to be accompanied by a stronger ENSO asymmetry in model. The left panel in Fig. 8a shows a slightly positive relationship between the variance and the skewness of Niño3 SST anomalies during the last 5 years of the 2th century. There was also a positive relationship between ENSO asymmetry and the mean western Pacific warm-pool size, which is demonstrated by a comparison of the scatterplots of the mean Western Pacific warm pool size and the skewness of Niño3 SST anomaly (Fig. 8b). This positive relationship appears to be weak, and this may have been because most of the models simulated too-weak or non-existent ENSO asymmetry and thus a weak or nonexistent time mean effect of ENSO. If the models showed large variation among simulations of ENSO asymmetry, a more substantial relationship could be diagnosed. Although the excessive cold tongue was a common feature revealed by previous studies, none linked the biases of ENSO asymmetry to warm-pool simulation. Considering that the definition of the ENSO in models may not be accurate based on the Niño3 index alone, weak ENSO asymmetry and the corresponding weak or non-existent time mean effect of ENSO in model simulations were also subjected to empirical orthogonal function (EOF) analysis. Three models [i.e., GISS AOM (Goddard Institute for Space Studies Atmosphere Ocean Model, Russell et al., 995), GISS MODEL EH (GISS Model version E-H, Schmidt

8 NO. 3 SUN ET AL. 947 Fig. 7. The same as Fig. 3, but for runs from a single model (a) the GFDL CM2., (b) GFDL CM2., (c) NCAR CCSM3., and (d) NCAR PCM. et al., 26) and GISS MODEL ER (GISS Model version E-R, Schmidt et al., 26)] simulated tooweak or nonexistent ENSO. And two models [i.e., BCCR BCM2. (Bjerknes Centre for Climate Research Bergen Climate Model Version 2., Furevik et al., 23) and CNRM CM3 (National Centre for Meteorological Research Climate Model version 3, Salas- Mélia et al., 25)] simulated a stronger ENSO but a very small or non-existent western Pacific warm pool. Among the rest of the models, only two models [i.e., IAP-FGOALS G. (Institute of Atmospheric Physics Flexible Global Ocean-Atmosphere-Land System Model Grid Version., Yu et al., 24) and IPSL CM4 (Institute Pierre Simon Laplace Climate System Model version 4, Goosse and Fichefet, 999)] simulated a stronger ENSO than observations, and the remaining ten models simulated weak or much weaker ENSO than observations. Most of the modeled EN- SOs were confined to the narrow equator band (Fig. 9). The ENSOs in the models showed large discrepancies in both pattern and amplitude compared with the observations. The ENSOs in the models also exhibited weak or non-existent ENSO asymmetry according to the PDF analysis of skewness of the first EOF time series in the models and in the observations (Fig. ). The positive relationship between the time-mean western Pacific warm pool size and skewness was somewhat more distinct than the analysis based on the Niño3 index, but still the result was not robust. The weak ENSO asymmetry and weak or nonexistent time mean effect of ENSO in the CMIP3 models was also seen in the composite analysis of ENSO based on the Niño3 index. In addition to the two models that simulated too-small or nonexistent warm pools and the three CMIP3 models that simulated too-weak or nonexistent ENSO, 9 models (i.e., CSIRO MK3.5 (Commonwealth Scientific and Industrial Research Organization Climate System Model version MK3.5, Gordon et al., 22), GFSL CM2. (Geophysical Fluid Dynamics Laboratory Climate Models version 2., Delworth et al., 26), GDDL CM2. (Geophysical Fluid Dynamics Laboratory Climate Models version 2., Delworth et al., 26), IPSL CM4, MIROC3.2 Hires (Model for Interdisciplinary Research on Climate version 3.2-high-resolution, Nozawa et al., 25), MIROC3.2 Medres (Model for Interdisciplinary Research on Climate version 3.2-mediumresolution, Nozawa et al., 25), NCAR CCSM3. (Kiehl and Gent, 24), UKMO HadCM3 (Met Office of Hadley Centre for Climate Prediction and Research in United Kingdom Climate Model version 3., Gordon et al., 2) and UKMO HadGEM (Met Office

9 x.5.5 VOL. 3 3 b a WPWP Size Skewness of Nino3 SSTA WESTERN PACIFIC WARM POOL AND ENSO ASYMMETRY IN CMIP Variance of Nino3 SSTA Skewness of Nino3 SSTA Fig. 8. (a) Scatterplot between variance ( C2 ) and skewness of Nin o3 SST anomaly from 95 to 999 and (b) the scatter plot between mean western Pacific warm-pool size (m2 ) and skewness of Nin o3 SST anomaly from 95 to 999. The black pentagram represents the observations, and the hexagram represents the average of all CMIP3 model runs. Colors and their data names refer to Table. The same color indicates the same and the different markers in the same color represent the different runs in the same model. Fig. 9. The first EOF mode of the tropical pacific SST from 95 to 999 in observation (HadISST) and in CMIP3 models. The first plot is for the observations, and the rest are for the models. Each model is the result of the multi-run ensemble mean. The same color scale is used to depict the SST anomaly in all plots.

10 NO. 3 SUN ET AL. 949 PDF.5 a Skewness of EOF Time Series Mean WPWP Size 5 x 3 b Skewness of EOF Time Series Fig.. (a) PDF analysis of the skewness of the time series in the first EOF mode for observations and model-ensemble means. The red vertical line indicates the value for observations. The vertical blue line is the multi-model ensemble mean. The short color bars on the horizontal axis mark the multi-run ensemble mean values for the individual models. (b) The scatterplot between mean western Pacific warm-pool size (m 2 ) and skewness of time series of first EOF mode in model simulations and observations from 95 to 999. The black pentagram represents the observations, and the hexagram represents the average of all CMIP3 model-ensemble means. Please refer to Table for colors and their model names. of Hadley Centre for Climate Prediction and Research in United Kingdom Global Environmental Model version, Johns et al., 26) of the 2 models simulated similar patterns but with weak time mean effects of ENSO as measured by the residual of ENSO asymmetry (Fig. ). In observations, the western Pacific warm pool was larger and narrower, and extended farther eastward during El Niño than La Niña. Several models did simulate a larger El Niño warm pool and asmallerlaniña warm pool, which was consistent with the observations. But few models captured all of the features of the ENSO time mean effect both in pattern and amplitude. The differences between the El Niño warm pool and the La Niña warm pool in most model simulations were smaller than in the observations (Fig. ). The weak or nonexistent ENSO asymmetry in nearly all of the CMIP3 models may have contributed to the smaller and warmer western Pacific warm-pool simulations to some extent. The Niño4 and Niño3.4 indices were are also used to evaluate the simulation of ENSO asymmetry in the climate models. As one aspect of ENSO asymmetry, smaller absolute value of skewness, occurred in most of the models. In the observations, the skewness of Niño3 (or Niño3.4) was positive and the skewness of Niño4 was negative. In the models, however, the absolute value of the skewness of all the three indices was smaller, and the average of the skewness in all the model runs was almost zero for all three indices (Fig. 2). The relationship between the biases in warm-pool simulations, especially in warm-pool size and those ENSO asymmetry simulation using Niño4 and Niño3.4 indexes, were also examined. The findings support our conclusions using the Niño3 index. The larger ENSO asymmetry biases (i.e., larger skewness biases in Niño3, Niño3.4, and Niño4) usually corresponded to larger biases in warm-pool size (i.e., smaller warm-pool size) (Fig. 3). Figure 3 shows a positive relationship between the biases of skewness in Niño3 (Niño3.4) and the biases of warm-pool size, while there is a negative relationship between the biases of skewness in Niño4 and the biases of warm-pool size. 4. Conclusions Motivated by recent empirical as well as theoretical results concerning the time-mean effect of ENSO events on the tropical Pacific climatology, we examined the biases in the simulations of the western Pacific warm pool in relation to the biases in ENSO statistics. A common bias in the simulation of the warm pool is a smaller and warmer warm pool in the models than in the observations. A corresponding common bias in the simulation of ENSO is the lack of ENSO asymmetry. The lack of variation among the models in simulating ENSO asymmetry prevents the performance of a robust statistical correlation between the magnitude of the warm-pool bias and the magnitude of the bias in ENSO asymmetry. But given the model diversity and

11 95 WESTERN PACIFIC WARM POOL AND ENSO ASYMMETRY IN CMIP3 VOL. 3 Fig.. The residual time mean effect of ENSO estimated from ENSO asymmetry. The residual effect was calculated using the mean SST anomaly in El Nin o plus the mean SST anomaly in La Nin a. The contours in red represent the 28 C mean SST during El Nin o; the contours in black represent the 28 C mean SST during La Nin a. El Nin o was based on a threshold of +.5 C, and La Nin a was based on a threshold of.5 C for the Nin o3 region (5 N 5 S, 5 27 W). Numbers and their data names are listed in Table. Two models (BCCR-BCM2. and CNRM-CM3) with too-small or no warm pools and three models (GISS-AOM, GISS-MODEL-EH, and GISS-MODEL-ER) with no ENSO were removed from the plots. the consistency in the results based on three indices of ENSO (Nin o3, Nin o3.4, and Nin o4), the chance of such a correspondence between the warm-pool bias and the bias in ENSO asymmetry being random is very small. It follows that the suggested time-mean effect needs to be taken seriously, and that improving ENSO statistics (i.e., ENSO asymmetry in particular) could represent a path toward improving the simulation of the tropical Pacific climatology. It should be emphasized that the biases in the warm-pool climatology caused by a bias in ENSO asymmetry may further enhance the bias in the ENSO asymmetry. Thus, a vicious cycle is perpetuated that is both hard to break and explains the persistence of the biases that have been noted in the two key aspects of the tropical Pacific climate. Fully recognizing this possibility, however, may help us to formulate a more complete strategy to improve the simulation and modeling of tropical Pacific climate on which climate variability modeling over much of the world depends. The importance of dynamical coupling in creating the climatological warm-pool and cold-tongue configurations in the tropical Pacific have long been recognized (Dijkstra and Neelin, 995; Clement et al., 996; Jin, 996; Sun and Liu, 996). The relative roles of clouds and ocean dynamics in creating and maintaining the western Pacific warm pool have also been assessed (Clement et al., 25). But these stud-

12 NO. 3 SUN ET AL a 5 4 c 4 3 e (%) 2 (%) 3 2 (%) Nino Nino Nino3 PDF b Nino4 PDF.5 d Nino3.4 PDF f Nino3 Fig. 2. (a, c, e) The distribution for the Niño4, Niño3.4 and Niño3 index from 95 to 999. (b, d, f) The probability density function analysis for skewness of Niño4, Niño3.4, and Niño3 indices from 95 to 999, respectively. WPWP Size x 3 a Nino4 WPWP Size x 3 b Nino3.4 WPWP Size x 3 c Nino3 Fig. 3. Scatterplot showing the relationship between the biases of warm-pool simulation (western Pacific warm pool size) and the biases of the Niño indices [ (a) Niño4, (b) Niño3.4, and (c) Niño3]. The black hexagram represents the average of all CMIP3 model runs. Please refer to Table for colors and data names. The same model was used; the same color and the different markers in the same color represent the different runs in the same model. ies have not addressed the collective role of ENSO events from the scale-interaction perspective. Thus, the present study, together with earlier studies that followed this line of thinking (e.g., Rodger et al., 24; Schopf and Burgman, 26; Sun and Zhang, 26; Sun and Yu, 29; Sun, 2; Liang et al., 22; Sun et al., 22 b ) extends these theoretical and modeling studies by pointing out the potential importance of rectification of ENSO events into the mean in shaping the size and temperature of the warm pool. Acknowledgements. This work was supported by the Strategic Priority Research Program Climate Change: Carbon Budget and Related Issues of the Chinese Academy of Sciences (Grant No. XDA532), National Natural Science Foundation of China (NSFC) Major

13 952 WESTERN PACIFIC WARM POOL AND ENSO ASYMMETRY IN CMIP3 VOL. 3 Research Project (Grant Nos and 48955), the National Basic Research Program of China for Structures, Variability, and Climatic Impacts of Ocean Circulation and Warm Pool in the Tropical Pacific Ocean (Grant No. 22CB474), China Postdoctoral Science Foudation funded project (22M52378), Chinese Scholarship Council, the Large-scale and Climate Dynamics Program of the US National Science Foundation (Grant Nos. AGS 553 and AGS ). Half of the data analysis work was carried on the computers at Physical Science Division of Earth System Research Laboratory of the National Oceanic and Atmospheric Administration. We would like to thank University of Colorado at Boulder for hosting SUN Yan. We would also like to thank Dr. ZHANG Tao for his help with some programming issues. REFERENCES AchutaRao, K., and K. R. Sperber, 22: Simulation of the El Niño Southern Oscillation: Results from the coupled model intercomparison project. Climate Dyn., 9, An, S.-I., Y.-G. Ham, J.-S. Kug, F.-F. Jin, and I.-S. Kang, 25: El Niño La Niña asymmetry in the coupled model intercomparison project simulations. J. Climate, 8, Boville, B. A., and P. R. Gent, 998: The NCAR climate system model, version one. J. Climate,, 5 3. Burgers, G., and D. B. Stephenson, 999: The normality of El Niño. Geophys. Res. Lett., 26, Clement, A. C., R. Seager, M. A. Cane, and S. E. Zebiak, 996: An ocean dynamical thermostat. J. Climate, 9(9), Clement, A. C., R. Seager, and R. Murtugudde, 25: Why are there tropical warm pools? J. Climate, 8(24), Davey, M., and Coauthors, 22: STOIC: A study of coupled model climatology and variability in tropical ocean regions. Climate Dyn., 8, Delworth, T., and Coauthors, 26: GFDL s CM2 global coupled climate models. Part I: Formulation and simulation characteristics. J. Climate, 9, Dijkstra, H. A., and J. D. Neelin, 995: Oceanatmosphere interaction and the tropical climatology Part II: Why the Pacific cold tongue is in the east. J. Climate, 8, Furevik, T., M. Bentsen, H. Drange, I. K. T. Kindem, N. G. Kvamsto, and A. Sorteberg, 23: Description and evaluation of the Bergen climate model: ARPEGE coupled with MICOM. Climate Dyn., 2, Goosse, H., and T. Fichefet, 999: Importance of iceocean interactions for the global ocean circulation: A model study. J. Geophys. Res., 4, Gordon,C.,C.Cooper,C.A.Senior,H.T.Banks,J.M. Gregory, T. C. Johns, J. F. B. Mitchell, and R. A. Wood, 2: The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Climate Dyn., 6, Gordon, H. B., and Coauthors, 22: The CSIRO Mk3 climate system model. aspendale: CSIRO atmospheric research. (CSIRO atmospheric research technical paper; no. 6). 3pp. [Available online at 2.pdf.] Hannachi, A., D. B. Stephenson, and K. R. Sperber, 23: Probability-based methods for quantifying nonlinearity in the ENSO. Climate Dyn.,. doi:.7/s Hansen, J., and Coauthors, 28: Target atmospheric CO 2: Where should humanity aim? The Open Atmospheric Science Journal, 2, 27 23, doi:.274/ Held, I. M., and A. Y. Hou, 98: Nonlinear axially symmetric circulations in a nearly inviscid atmosphere. J. Atmos. Sci., 37, Hou, A. Y., 998: Hadley circulation as a modulator of the extratropical climate. J. Atmos. Sci., 55, Jin, F.-F., 996: Tropical ocean interaction, Pacific cold tongue, and El Niño Southern Oscillation. Science, 274, Johns, T. C., and Coauthors, 26: The new Hadley Centre climate model (HadGEM): Evaluation of coupled simulations. J.Climate, 9, Kiehl, J. T., 998: Simulation of the tropical Pacific warm-pool with the NCAR climate system model. J. Climate,, Kiehl, J. T., and P. R. Gent, 24: The community climate system model, version 2. J. Climate, 7, Kug,J.-S.,J.Choi,S.-I.An,F. F.Jin,andA.T.Wittenberg, 2: Warm pool and cold tongue El Niño events as simulated by the GFDL 2. coupled GCM. J. Climate, 23, Latif, M., and Coauthors, 2: ENSIP: The El Niño simulation intercomparison project. Climate Dyn., 8, Leloup, J., M. Lengaigne, and J.-P. Boulanger, 28: Twentieth century ENSO characteristics in the IPCC database. Climate Dyn., 3, Liang, J., X.-Q. Yang, and D.-Z. Sun, 22: The effect of ENSO events on the tropical Pacific mean climate: Insights from an analytical model. J. Climate, 25, Lu, P., J. P. McCreary Jr., and B. A. Klinger, 998: Meridional circulation cells and the source waters of the Pacific equatorial undercurrent. J. Phys. Oceanogr., 28, Mechoso, M. R., and Coauthors, 995: The seasonal cycle over the tropical Pacific in coupled ocean-atmosphere general circulation models. Mon. Wea. Rev., 23, Meehl, G. A., C. Covey, T. Delworth, M. Latif, B. McAvaney, J. F. B. Mitchell, R. J. Stouffer, and K. E. Taylor, 27: The WCRP CMIP3 multimodel

14 NO. 3 SUN ET AL. 953 dataset: A new era in climate change research. BAMS, 88, , doi:.75/bams Newell, R. E., 979: Climate and the ocean. Amer. Sci., 67, Nozawa, T., T. Nagashima, H. Shiogama, and S. A. Crooks, 25: Detecting natural influence on surface air temperatures change in the early twentieth century. Geophys. Res. Lett., 23, L279, doi:.29/25gl2354. Philander, S. G., 99: El Niño, La Niña and the Southern Oscillation. Elsevier, New York, 293pp. Pierrehumbert, R. T., 995: Thermostats, radiator fins, and the runaway greenhouse. J. Atmos. Sci., 52, Ramananthan, V., and W. Collins, 99: Thermodynamic regulation of ocean warming by cirrus clouds deduced from observations of the 987 El Niño. Nature, Rayner,N.,A.,D.E.Parker,E.B.Horton,C.K.Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 23: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res., 8, 447, doi:.29/22jd267. Rodgers, K. B., P. Friederichs, and M. Latif, 24: Tropical Pacific decadal variability and its relation to decadal modulation of ENSO. J. Climate, 7, Russell, G. L., J. R. Miller, and D. Rind, 995: A coupled atmosphere-ocean model for transient climate change studies. Atmos. Ocean, 33(4), Salas-Mélia, and Coauthors, 25: Description and validation of the CNRM-CM3 global coupled model, CNRM working note 3. Schmidt, G., and Coauthors, 26: Present-day atmospheric simulations using GISS ModelE: Comparison to in situ, satellite, and reanalysis data. J. Climate, 9, Schopf, P. S., and R. J. Burgman, 26: A simple mechanism for ENSO residuals and asymmetry. J. Climate, 9, Spencer, R. W., 993: Global oceanic precipitation from the MSU during and comparisons to other climatologies. J. Climate, 6, Sun, D.-Z., 2: Global climate change and ENSO: A theoretical framework. El Niño: Historical and Paleoclimatic Aspects of the Southern Oscillation, Multiscale variability and Global and Regional Impacts. H. F. Diaz and V. Markgraf, Eds., 476pp, Cambridge University Press, Cambridge, Sun, D.-Z., 23: A possible effect of an increase in the warm-pool SST on the magnitude of El Niño warming. J. Climate, 6, Sun, D.-Z., 27: The role of ENSO in regulating its background state. Nonlinear Dynamics in Geosciences, J. Elsner and A. Tsonis, Eds., Springer New York, Sun, D.-Z., 2: The diabatic and nonlinear aspects of El Niño-Southern Oscillation: Implications for its past and future behavior. Climate Dynamics: Why Does Climate Vary? AGU Geophysical Monograph, D.- Z. Sun and F. Bryan, Eds., Amer. Geophys. Union, Sun, D.-Z., and Z. Liu, 996: Dynamic ocean atmosphere coupling: A thermostat for the tropics. Science, 272, Sun, D.-Z., and K. E. Trenberth, 998: Coordinated heat removal from the equatorial Pacific during the El Niño. Geophys. Res. Lett., 25, Sun, D.-Z., and T. Zhang, 26: A regulatory effect of ENSO of ENSO on the time-mean thermal stratification of the equatorial upper ocean. Geophys. Res. Lett., 33, L77, doi:.29/25gl Sun, F. P., and J.-Y. Yu, 29: A 5-yr modulation cycle of ENSO intensity. J. Climate, 22, Trenberth, K. E., and A. Solomon, 994: The global heat balance: Heat transports in the atmosphere and ocean. Climate Dyn.,, Webster, P. J., and R. Lukas, 992: TOGA COARE: The coupled ocean-atmosphere response experiment. Bull. Amer. Meteor. Soc., 73, Yu, Y., X. Zhang, and Y. Guo, 24: Global coupled ocean atmosphere general circulation models in LASG/IAP. Adv. Atmos. Sci., 2, Zhang, T., D.-Z. Sun, R. Neale, and P. Rasch, 29: An evaluation of ENSO asymmetry in the community climate system Models: A view from the csubsurface. J. Climate, 22, , doi:.75/29jcli2933..