GE Zi-n nd CHEN Lin,b,c

ABSTRACT The simulated sea surface temperature anomaly (SSTA) over the tropical Pacific during El Ni?o—Southern Oscillation (ENSO) is investigated in three representative coupled models:CESM1-CAM5, FGOALS-s2, and FGOALS-g2. It is found that there is a significant westward shift bias in reproducing the zonal distribution (ZD) of the ENSO-related SSTA in CESM1-CAM5 and FGOALS-s2, whereas the SSTA-ZD simulated by FGOALS-g2 is relatively realistic.Through examining the SSTA-ZD during both warm and cold phases of ENSO separately,the authors reveal that the SSTA-ZD simulation bias during the ENSO cycle mainly lies in the bias d′uring the warm phase. It is noted that both the simulated zonal wind stress anomaly (τx) and shortwave heat flux (SW) anomaly exhibit westward shift biases in CESM1-CAM5 and FGOALS-s2, while the counterparts in FGOALS-g2 are relatively reasonable. The westward shift biases in representing τ′x and the SW anomaly (SWA) are attributed to the westward-shifted precipitation anomaly (PrA). It is suggested that the mean SST cold bias over the cold tongue region is the key factor behind the westward-shift bias in simulating the El Ni?o-related PrA, which leads to the westward-shifted τ′x and SWA. Collectively, the aforementioned anomaly fields, including the dynamic part (τ′x) and thermodynamic part(SWA), contribute to the westward-shift bias in simulating the El Ni?o-related SSTA. This study provides clues for understanding the ZD simulation biases of ENSO-related fields;however, further in-depth investigation with more model simulations, especially the incoming CMIP6 simulations, is still needed to fully understand the ENSO SSTA-ZD simulation bias in coupled models.

KEYWORDS ENSO; SSTA zonal distribution; coupled models;simulation bias

1. Introduction

The coupled general circulation model (CGCM) has become a powerful tool for examining El Ni?o—Southern Oscillation (ENSO) dynamics, because of its ability to depict the complex interactions of dynamic and thermodynamic feedbacks in the coupled atmosphere—ocean system (Collins et al. 2010). The accuracy of CGCMs in simulating the ENSO phenomenon has been widely evaluated (e.g., AchutaRao and Sperber 2006; Guilyardi et al.2009). However, despite the progress that has been made, CGCMs still exhibit significant biases in simulating basic ENSO features (Guilyardi et al. 2009; Bellenger et al.2014), which critically restricts their climate predictive skills.

Many studies have reported that current CGCMs have a problem with reasonably simulating ENSO’s amplitude,asymmetry, period, irregularity, spatial pattern, and so on(e.g., Bellenger et al. 2014). Among the various aspects of ENSO, realistically representing the spatial distribution of the ENSO-related sea surface temperature anomaly (SSTA)is still challenging in model simulation (Bellenger et al.2014). Previous studies have documented the bias in simulating the meridional structure of the ENSO-related SSTA (Zhang et al. 2013). In terms of the bias in representing the SSTA zonal distribution (hereafter, SSTA-ZD) in ENSO simulation, it is noticed that the modeled SSTA-ZD in the tropical Pacific exhibits a significantly westwardshift bias in the models included in phase 3 of the Coupled Model Intercomparison Project (CMIP3) (Leloup,Lengaigne, and Boulanger 2008; Yu and Kim 2010). The westward-shift bias exhibited in the CMIP3 models still prevails in the CMIP5 models (Zhang and Sun 2014).However, little attention has been paid to the physical processes that are responsible for this simulation bias of the ENSO SSTA pattern in CGCMs.

It is suggested that inaccurate duplication of the ENSO-related atmospheric and oceanic feedback processes is responsible for the ENSO simulation bias (e.g.,Chen, Yu, and Zheng 2016). As ENSO involves various atmosphere—ocean interactions, the collaboration and competition among different atmosphere—ocean feedback processes result in the different ENSO behavior(Bjerknes 1969; Jin 1997). To the first order, the primary ENSO-related dynamic feedbacks include the thermocline feedback, zonal advection feedback, and Ekman feedback, which are all associated with the zonal wind stress anomaly (hereafter, τ′x) during the ENSO cycle; and the primary ENSO-related thermodynamic feedback is the shortwave feedback (Sun, Yu, and Zhang 2009;Chen, Yu, and Sun 2013).

Previous studies have pointed out that both the magnitude and zonal structure simulation biases in the modeled τ′xlead to ENSO amplitude simulation biases(Watanabe et al. 2011; Kim and Jin 2011; Zhang and Sun 2014). This indicates that the bias in simulating the SSTA-τ′xcoupling (which is the essential element of the ENSO dynamic feedbacks) may have a profound influence on the simulation bias in the ENSO-related SSTA-ZD. For the thermodynamic feedbacks, the biases in simulating the zonal structure of the shortwave heat flux (SW) anomalies in CGCMs are also obvious (Chen et al. 2019a). Yu and Kim (2010) decomposed the SSTA spatial pattern in CMIP3 models into Eastern Pacific (EP)and Central Pacific (CP) ENSOs and argued that the cold bias in the simulated EP ENSO type is responsible for the westward-shift bias in the total SSTA.

A number of modeling studies have proposed that the mean state bias is a key factor behind ENSO simulation biases. Chen et al. (2019a) and Chen, Yu, and Sun(2013) pointed out that the westward-shift biases in the simulated atmospheric anomaly fields, such as SW and the associated precipitation during the ENSO cycle, arise from the mean SST cold bias over the Pacific cold tongue in CGCMs. Watanabe et al. (2011) and Kim and Jin (2011)argued that the simulation bias of τ′xcan be attributed to the mean precipitation simulation bias over the eastern Pacific. Zhang and Sun (2014) showed that the westward-shift bias in simulating the τ′xin CMIP5 models is caused by the stronger-than-observed climatological zonal winds.

The aim of this study is to investigate the air—sea feedback processes responsible for the westward-shift bias in simulating the ENSO-related SSTA in CGCMs and provide some preliminary clues for the simulation bias.

2. Data and models

To understand why some models have severe bias in simulating the ENSO-related SSTA-ZD while some models show minor simulation bias, we employed three representative CGCMs from CMIP5 to conduct the analysis. The representative CGCMs used in this study include FGOALS-s2 (Bao et al. 2013), FGOALS-g2 (Li et al. 2013), and CESM1-CAM5 (Hurrell et al. 2013). Both CESM1-CAM5 and FGOALS-s2 show obvious biases in simulating the ENSO-related SSTA-ZD, while FGOALS-g2 presents relatively reasonable simulation (see section 3 for details).

The specific reason for choosing these three CGCMs in this preliminary study, as well as descriptions of the observational and reanalysis datasets, is introduced in detail in the Supplementary Material.

3. Results

The equatorial profile of the ENSO-related SSTA-ZD is shown in Figure 1(a). Here, we use the standard deviation of the monthly SSTA during the entire time series to represent the distribution of the ENSO-related SSTA. In the observation, most of the ENSO-related SSTA is located across the eastern equatorial Pacific (150°W-80°W). For the three representative CMIP5 models, their average shows an obvious westward-shift bias compared to the observation. In particular, the variability of the observed ENSO-related SSTA peaks at around 100°W and maintains at a high value farther east to the Peruvian coast, but that of the modeled SSTA peaks at around 130°W and exhibits a dramatic decline to the east of its maximum. Specifically,the ENSO-related SSTA simulated by FGOALS-g2 is closer to the observation than its counterparts in FGOALS-s2 and CESM1-CAM5. A recent study (Chen et al. 2018) found that the anomaly fields during the ENSO cycle exhibit different spatial patterns between the warm phase (i.e., El Ni?o)and cold phase (La Ni?a) of ENSO, and thus we further examine the spatial distribution of the SSTA through separating the ENSO cycle into two phases. The distributions of the El Ni?o-related SSTA and La Ni?a-related SSTA along the equator are presented in Figure 1(b,c). It is found that the El Ni?o-related SSTA exhibits a clear westward-shift bias (Figure 1(b)), while the La Ni?a-related westward-shift bias is relatively minor (Figure 1(c)). This indicates that the bias in simulating the ENSO-related SSTA-ZD primarily lies in the bias of the El Ni?o-related SSTA. Thus, we mainly focus on the simulation bias and the corresponding factors during the ENSO warm phase.

Figure 2 shows the composite SSTA pattern during the El Ni?o mature phase (ND[0]JF[+1]) for the observation and the three representative models. Here, and throughout the paper, the upper-case letters represent months (e.g., N = November, J = January etc.), and the numbers in square brackets indicate the year (i.e.,[0] = the current year, [+1] = the following year). In the observation, the pronounced positive SSTA during the El Ni?o mature phase is located in the eastern equatorial Pacific, with the maxima over the far eastern Pacific east of 120°W (Figure 2(a)). As shown in Figure 2(b,c), the El Ni?o-related SSTA patterns simulated by CESM1-CAM5 and FGOALS-s2 exhibit significant westward-shift biases compared to the observation. In contrast, FGOALS-g2 produces a relatively more realistic SSTA pattern.Furthermore, we calculated the equatorial profile of the SSTA normalized by the magnitude of the SSTA averaged over a broad box in the central—eastern equatorial Pacific (CEEP; 5°S—5°N, 180°—80°W). The result (Figure 2(e)) confirms that the westward bias in simulating the El Ni?o-related SSTA is apparent in CESM1-CAM5 and FGOALS-s2 but not significant in FGOALS-g2.

Moreover, we propose a Zonal Distribution Index(hereafter, ZDI) to quantitatively measure the ENSO-related SSTA-ZD. The results based on the ZDI show consistent SSTA-ZD simulation bias; that is, CESM1-CAM5 and FGAOLS-s2 show westward bias in simulating the El Ni?o-related SSTA, whereas FGOALS-g2 yields a more reasonable El Ni?o-related SSTA (see section 1 in the Supplementary Material for details).

Based on the assessment above, we examined the ENSO dynamic and thermodynamic feedbacks to demonstrate the physical processes responsible for the westward-shift biases in the three representative CGCMs. In the ENSO cycle, the dominant dynamic feedbacks affecting the evolution of the SSTA include the thermocline (TH), zonal advection (ZA), and Ekman (EK)feedbacks. As described in the introduction, these three ocean dynamic feedbacks (TH, ZA, and EK) all depend on the SSTA-τ′xfeedback. As boreal autumn is the season of the strongest coupled ENSO instability for ENSO SSTA development (Philander et al. 1996; Li 1997), we thus next examine the performance of the simulated τ′xduring the El Ni?o developing phase (SOND[0]J[+1]) in the three coupled models. Figure 3 displays the composite τ′during the El Ni?o developing phase for the observaxtion and three CMIP5 models. In the observation, the equatorial τ′xstretches farther east to 120°W, with the maxima located to the east of the dateline, and it seems that there is no large deficiency in the ZD of the simulated τ′xin the three coupled models. To facilitate the comparison, we present the normalized τ′xalong the equatorial Pacific (Figure 3(e)), in a manner similar to that utilized in Figure 2(e). As shown in Figure 3(e), the τ′simulated by CESM1-CAM5 and FGOALS-s2 show sigxnificant westward-shift biases, while its counterpart in FGOALS-g2 does not exhibit westward-shift bias. The ZDI for τ′xis also consistent with this result. Considering the fact that the zonal bias of the SSTA mainly occurs in CESM1-CAM5 and FGOALS-s2, it is suggested that the westward-shifted τ′xfavors the westward shift of the positive SSTA’s growth. This is because the farther westward τ′xleads to farther westward ocean dynamic feedbacks (TH, ZA, and EK), naturally causing the growth and development of the El Ni?o-related SSTA positioning to the west, relative to the observation. Overall, the westward bias in simulating the τ′xcontributes to the westward bias in simulating the El Ni?o-related SSTA.

Figure 1. Equatorial profile (averaged over 2°S—2°N) of the standard deviation of the monthly SSTA (units: K) for (a) the whole time series (indicating the entire ENSO cycle), (b) the El Ni?o situation (only positive SSTA considered), and (c) the La Ni?a situation (only negative SSTA considered).

Figure 2. The horizontal pattern of the composite SSTA (units: K) during the El Ni?o mature phase (ND[0]JF[+1]) for (a) the observation, (b) CESM1-CAM5, (c) FGOALS-s2, and (d) FGOALS-g2. Here, and throughout the paper, the upper-cse letters represent months (e.g., S = September, O = October etc.), the numbers in square brackets indicate the year (i.e., [0] = the current El Ni?o year,[+1] = the following year); and the SSTA averaged over the Ni?o3.4 region (5°S—5°N, 170°—120°W) exceeding 0.5 standard deviations for five consecutive months is considered as an El Ni?o event. (e) The equatorial profiles (averaged for 5°S—5°N) of the El Ni?o-related SSTA normalized by the SSTA averaged over a broad box in the central-eastern equatorial Pacific (5°S—5°N, 180°—80°W). The vertical dashed lines indicate the ZDI longitudes of the observed and simulated SSTA-ZD.

The SSTA—SW feedback is another key factor affecting the evolution of ENSO. As shown in Figure 4(a), the observed SW anomaly (SWA) exhibits a basin-wide cooling over the equatorial Pacific, with the maxima located to the east of the dateline during the developing phase of El Ni?o. Clearly, the negative SWA could suppress the development of the positive SSTA over the central equatorial Pacific, which may partly contribute to the fact that the El Ni?o-related SSTA mainly locates in the eastern equatorial Pacific. For CESM1-CAM5 (Figure 4(b)), the maximum SWA is located at 170°E, and that simulated by FGOALS-s2 (Figure 4(c)) is located at 160°E, so both simulations have an obvious westward shift (by 25°—35°)compared to the observation (centered around 165°W).The ZDI for the SWA in CESM1-CAM5 and FGOALS-s2 also shows westward-shift biases, by 20°. For CESM1-CAM5 (Figure 4(b)) and FGOALS-s2 (Figure 4(c)), the simulated SWA locates too far west compared to the observation, and thus the effect of suppressing the positive-SSTA development is marginal to the east of the dateline. In contrast, FGOALS-g2 shows no significant bias in simulating the ZD of the SWA anomaly. To further clarify the role of the SWA simulation bias in affecting the bias in simulating the SSTA-ZD, Figure 4(f) shows the deviation of the simulated SWA from the observation. As presented by the dashed lines in Figure 4(f), the positive/negative difference value at a certain longitude means the cooling effect due to the SWA is weaker/larger than the observed, which leads to the unfavorable/favorable condition for warm SSTA development over the particular longitude. From this viewpoint, we can clearly see that the westward-shifted SWA in CESM1-CAM5 can induce an easier growth and development of the warm SSTA over the location to the west of 120°W than the location to the east of 120°W. This would partly contribute to the westward distribution of the El Ni?o-related SSTA in CESM1-CAM5 relative to the observation (in which the maximum El Ni?o-related positive SSTA locates to the east of 120°W). Similarly, the more severe bias in simulating the negative SWA in FGOALS-s2 could falsely overheat the equatorial zonal band extending from 170°W to 120°W compared with the region to the east of 120°W, which partly contributes to the westwardshift bias in simulating the El Ni?o-related SSTA-ZD. In contrast, the ZD of the SWA simulated by FGOALS-g2 does not show too significant a bias compared to the observation, which may partly contribute to its relatively reasonable simulation of the El Ni?o-related SSTA-ZD.Overall, the westward bias in simulating the SWA is another key contributor for the westward bias in simulating the El Ni?o-related SSTA.

In short, the westward biases in simulating both theand SWA match well with the westward biases in simulating the El Ni?o-related SSTA. In fact, the bias in simulating both theand SWA can be attributed to the bias in simulating the atmospheric convection (Sun, Yu,and Zhang 2009; Chen, Yu, and Sun 2013; Li, Wang, and Zhang 2014, 2015; Ferrett, Collins, and Ren 2018). Hence,we examined the precipitation anomaly (PrA) during the El Ni?o developing phase for the observation and three CMIP5 models. As shown in Figure S1, the overall performance in simulating the PrA is consistent with the performance in simulating theand SWA. That is, the PrA simulated in CESM1-CAM5 and FGOALS-s2 exhibit westward biases, whereas FGOALS-g2 generally fits the observation. The westward-shifted atmospheric convection directly leads to the westward-shift bias of the surface wind convergence distribution and cloud cover anomaly, which further induce the westward-shift biases in simulating theand SWA, respectively.

The investigation above shows that the simulation bias in representing the El Ni?o-related SSTA-ZD is associated with the bias in simulating theSWA and PrA. However,it seems that we cannot directly distinguish the causal relationship among the biases in simulating the aforementioned anomaly fields, because ENSO is a product of air—sea interaction, and the simulation outputs were derived from the models that have reached their own equilibrium states. As previous studies have suggested that the mean state simulation bias may be the root physical cause for the biases in simulating the ENSO-related anomaly fields (e.g., Guilyardi et al. 2009), we next examined the performance of simulating the tropical mean states in the three coupled models. Among the tropical mean state biases (Figure S7), it is found that the leading factor causing the westward-shift biases in simulating the equatorial PrA is the excessive cold tongue over the CEEP region. As shown in Figure 4(g,h), the mean SST over the cold tongue region simulated by CESM1-CAM5 and FGOALS-s2 exhibits an obvious cold bias compared to the observation. However, the mean SST cold bias in the FGOALS-g2 simulation is relatively mild.As we know, when the El Ni?o-related SSTA occurs in the CEEP and reaches a certain magnitude, it will induce convection (Graham and Barnett 1987). However, when the El Ni?o-related SSTA with the same magnitude occurs in the CEEP in a CGCM that suffers from a mean SST cold bias, the total SST may not exceed the convection threshold, and thus the appearance of the convection will ultimately shift towards the warmer western equatorial Pacific region. Therefore, it is demonstrated that the simulated colder mean SST over the Pacific cold tongue region pushes the PrA to shift towards the warmer western equatorial Pacific, and then the westward-shifted PrA directly leads to the westward-shifted τ′xand SWA. These westward-shift biases collectively contribute to the westward-shift bias in simulating the El Ni?o SSTA through SSTAτ′xfeedback and SSTA—SW feedback, as discussed above.

Figure 3. The horizontal pattern of the composite zonal wind stress anomaly (units: 10-2 N m-2) during the El Ni?o developing phase (SOND[0]J[+1]) for (a) the observation, (b) CESM1-CAM5, (c) FGOALS-s2, and (d) FGOALS-g2. (e) The equatorial profiles(averaged for 5°S—5°N) of the El Ni?o-relatednormalized by the value of theaveraged in the Ni?o4 region (5°S—5°N, 160°E—150°W). The vertical dashed lines indicate the ZDI longitudes of the observed and simulated equatoria patterns.

Figure 4. The horizontal pattern of the composite shortwave heat flux anomaly (units: W m-2) during the El Ni?o developing phase(SOND[0]J[+1]) for (a) the observation, (b) CESM1-CAM5, (c) FGOALS-s2, and (d) FGOALS-g2. (e) The equatorial profiles (averaged for 5°S—5°N) of the El Ni?o-related SWA normalized by the value of the SWA averaged in the Ni?o4 region (5°S—5°N, 160°E—150°W).The vertical dashed lines indicate the ZDI longitudes of the observed and simulated equatorial SWA patterns. (f) The difference between the model simulations and the observation. (g) Equatorial profile of mean SST (units: K) during the El Ni?o developing season (SONDJ). (h) As in (g) but for the mean SST bias.

4. Summary

In this study, we evaluated the performance of simulating the ENSO-related SSTA-ZD over the equatorial Pacifci in three representative CMIP5 models (CESM1-CAM5,FGOALS-s2, and FGOALS-g2), and investigated the contributing factors for the simulation bias through analyzing the corresponding ENSO dynamic and thermodynamic feedbacks. The main fnidings can be summarized as follows:

(1) The simulated ENSO-related SSTA in CESM-CAM5 and FGOALS-s2 exhibits a significant westwardshift bias compared to the observation, whereas FGOALS-g2 produces a relatively realistic SSTA pattern. By examining the SSTA-SD simulation during the warm and cold phases separately, we find that the westward-shift bias in simulating the ENSO-related SSTA mainly lies in the warm phase.

(2) Through examining the ENSO dynamic and thermodynamic feedbacks, we found that the simulated dynamic aspect () is westward shifted compared to the observation, which could lead to the El Ni?o-related SSTA growth tending towards the west relative to its observed counterpart. From the thermodynamic perspective, the simulated equatorial SWA also exhibits an obvious westwardshift bias in CESM1-CAM5 and FGOALS-s2. Such a westward bias in simulating the negative SWA could falsely overheat the equatorial zonal band extending from 170°W to 120°W compared to the observed El Ni?o-related SSTA center (i.e., to the east of 120°W), which also contributes to the westward-shift bias in simulating the El Ni?o-related SSTA-ZD. The westward biases in simulating both theand SWA can be traced back to the westward-shift bias in simulating the PrA.

(3) As the simulation results analyzed here are derived from models that have reached their equilibrium states, it is plausible that we cannot distinguish the causal relationship among the aforementioned ENSO-related anomaly fields.Since previous studies have suggested that the mean state simulation bias is the key factor behind the ENSO-related anomaly simulation biases, we examined the tropical mean state simulation biases. It was found that the mean SST cold bias over the cold tongue region is the key factor for the westward-shift bias in simulating the El Ni?o-related PrA. Because the cold SST bias over the central-eastern equatorial Pacific prevents the total SST from exceeding the convection threshold, the convection and the associated τ′xand SWA shift westwards to the warmer western equatorial Pacific. Thus, the corresponding τ′xand SWA cause the El Ni?o-related SSTA to develop towards the west, and ultimately the overall El Ni?o-related SSTA appears relatively farther west in the equatorial Pacific compared to that in the observation.

Additionally, we show in the Supplementary Material the criteria for selecting El Ni?o cases and provide our thinking on whether ENSO’s diversity (or the so-called CP El Ni?o) will or will not influence the current results. For the former issue, we show the spread of the composite results among the cases is minor and argue that our results are not sensitive to how we select the El Ni?o cases. For the latter issue, our strategy in the current study is to employ the same criteria to select the El Ni?o cases for both the observation and the CGCMs, which to varying extents have difficulty in reproducing El Ni?o’s diversity (e.g.,Feng et al. 2020). Thus, we argue that such a strategy is reasonable when comparing the model simulations with the observation.

It is worth mentioning that this study provides only a preliminary analysis for the simulation biases of the ENSO-related SSTA-ZD, and it is necessary to examine more model simulations, especially the incoming CMIP6 model results, as well as conduct some more in-depth investigations to confirm the preliminary conclusions in this study.

Disclosure statement

No potential conflict of interest was reported by the authors.

Funding

This work was jointly supported by the National Key Research and Development Program of China [Grant No. 2019YFC1510004],Natural Science Foundation of Jiangsu Province [Grant No.BK20190781], the General Program of Natural Science Foundation of Jiangsu Higher Education Institutions [Grant No.19KJB170019], the open fund of State Key Laboratory of Loess and Quaternary Geology [Grant No. SKLLQG1802], and the LASG Open Project.