ENSO hindcast skill of the IAP-DecPreS near-term climate prediction system:comparison of full-field and anomaly initialization

SUN Qin, WU Bo, ZHOU Tin-Jun n YAN Zi-XingCollege of Atmospheri Siene, Chengu University of Informtion Tehnology, Chengu, Chin; Leshn Centrl Sttion of Environment Monitoring, Leshn, Chin; LASG, Institute of Atmospheri Physis, Chinese Aemy of Sienes, Beijing, Chin; College of Atmospheri Siene, Nnjing University of Informtion Siene n Tehnology, Nnjing, Chin

1. Introduction

El Ni?o–Southern oscillation (ENSO) is the most dominant air–sea coupling mode of the climate system and has striking impacts on the global climate system (Aceituno 1992; Alexander et al. 2002; McPhaden, Zebiak, and Glantz 2006; Wang, Wu, and Fu 2000; Webster et al. 1998).Conventional ENSO is characterized by maximum warm sea surface temperature anomalies (SSTAs) in the equatorial eastern Pacific (Harrison and Larkin 1998; Rasmusson and Carpenter 1983). In the last 10 years, a new type of El Ni?o has emerged from the conventional El Ni?o, referred to as El Ni?o Modoki (Ashok et al. 2007), central Pacific (CP)El Ni?o (Yeh et al. 2009), or warm pool El Ni?o (Kug, Jin,and An 2009). The dominant feature of El Ni?o Modoki is maximum warm SSTAs in the equatorial central Pacific around the date line and weak negative SSTAs in the equatorial western and eastern Pacific.

Because of its strong variability and substantial global climate impacts, ENSO has always been a central target of seasonal and interannual climate predictions (see review by Barnston et al. 2012). ENSO predictions can be conducted using dynamical or statistical models. Impressive progress has been achieved in ENSO prediction by dynamical models. Some strong ENSO events can be predicted by dynamic models at 1-yr or even longer lead times, and most moderate and weak ENSO events can also be predicted several months in advance (Anderson et al. 2002;Cane, Zebiak, and Dolan 1986; Jin et al. 2008; Latif et al.1998; Luo and Yamagata 2005). ENSO predictive skill can be improved through the following approaches: ensemble forecasting using an intermediate coupled model (Zheng et al. 2006), assimilating wind observations (Zheng and Zhu 2010), and minimizing the uncertainties of parameterizing the effects of subsurface temperature (Zheng and Zhu 2015). At present, the skills of dynamical models exceed those of statistical models, especially for real-time predictions (Barnston et al. 2012).

An essential step of dynamical prediction is initialization, which obtains an initial model state close to the observation. There are two distinct types of initialization approaches: full-field and anomaly initialization (Smith,Eade, and Pohlmann 2013). Their major difference is that the former initializes the model through assimilating raw observational data, while the latter through assimilating model climatology plus observational anomalies. Full-field initialization effectively corrects model biases by constraining model states to the observations during assimilation processes, and thus obtains more accurate initial conditions than those obtained from anomaly initialization.However, a model initialized by the full-field approach will tend to gradually drift towards its preferred climatology,because it is not constrained by observations during hindcast/forecast integrations. In contrast, anomaly initialization preserves the preferred climatology of the model to a large extent, and thus minimizes the drift during hindcast/forecast integrations (Smith, Eade, and Pohlmann 2013).

Recently, the Institute of Atmospheric Physics (IAP)near-term climate prediction system, referred to as IAPDecPreS, was constructed using the FGOALS-s2 global general circulation model and a new ocean data assimilation scheme. The main aim of this study are to evaluate the skill of the system in ENSO and El Ni?o Modoki prediction,and to compare the differences in skill between the anomaly and full-field initialization approaches. IAP-DecPreS and the observational datasets used are introduced in Section 2. Section 3 evaluates the skill of the system in ENSO prediction from two aspects–the temporal evolution of ENSO indices and large-scale spatial patterns during ENSO mature winter. A summary is given in Section 4.

2. Climate prediction system and observational data

2.1. Climate prediction system

IAP-DecPreS was constructed based on a state-of-the-art coupled global climate model (CGCM), FGOALS-s2, developed by the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics(LASG) at the IAP, Chinese Academy of Sciences (Bao et al.2013; Zhou et al. 2014, supplementary material). A new ocean data assimilation scheme, referred to as ‘ensemble optimal interpolation-incremental analysis update’(EnOI-IAU; supplementary material) was developed for FGOALS-s2. EnOI-IAU is the result of a combination of the EnOI and IAU assimilation schemes. EnOI is used to generate an analysis increment. Then, IAU is used to incorporate the analysis increment into the model (Wu, Zhou, and Sun 2017, supplementary material). Initializations using two different approaches–full-field and anomaly initializations–were conducted separately.

Initiated from the initial states derived from the full-field or anomaly initialization runs, systematic hindcast runs were conducted. They are referred to as full-field or anomaly hindcasts. The hindcast runs were initiated from February, May, August, and November, separately, for each year in the period 1979–2015. Both the full-field and anomaly hindcasts had nine ensemble members, with small perturbations of atmospheric and oceanic initial states.Following Barnston et al. (2012), three-month mean variables were taken as prediction targets. We used ann-month lead time to represent the separation between the initial date and the three-month forecast target period. For both the hindcasts and the corresponding observational references, anomalies were calculated as deviations from their climatology over the period 1979–2015.

2.2. Observational data

The following observational datasets were used to assess the predictive skill: (1) The monthly mean precipitation data provided by the Global Precipitation Climatology Project(GPCP; Adler et al. 2003); (2) The extended reconstructed sea surface temperature (SST) from the National Oceanic and Atmospheric Administration (ERSST.v4; Huang et al.2015); (3) The monthly mean sea level pressure (SLP) and circulation data derived from ERA-Interim (Dee et al. 2011).All the datasets cover the period 1979–2015.

3. Results

3.1. Temporal correlation skill for SSTAs

Figure 1 shows the spatial distributions of the temporal correlation skill scores for SSTA predictions at 1-, 4-, 7-, and 10-month lead time. For the anomaly hindcasts, significant correlations are apparent in most areas of the Pacific, tropical Indian Ocean, and North Atlantic at 1-month lead time(Figure 1(a)). The highest skill scores are in the equatorial central-eastern Pacific (CEP), a key area of ENSO. For the lead times of 4, 7, and 10 months, the ocean areas with significant skill scores gradually shrink (Figure 1(b)–(d)).The correlation skill scores in the equatorial CEP gradually decrease from about 0.6 to 0.3. In contrast, the skill scores in the tropical western Pacific, western Atlantic, and Indian Ocean remain above 0.6 persistently, which may be associated with the deeper thermocline and associated greater thermal inertia in these regions.

Figure 1.Spatial distributions of temporal correlation skill scores for predictions of monthly SSTAs in the period 1979–2015: (a–d)anomaly hindcasts at 1-, 4-, 7-, and 10-month lead time; (e–h) as in (a–d) but for the full-field hindcasts. Dotted areas denote values reaching the 0.05 significance level. (i–l) Differences between (a–d) and (e–h).

For the full-field hindcasts, the correlation skill scores at 1-month lead time are lower than their counterparts in the anomaly hindcasts in some ocean areas by about 0.1–0.2(Figure 1(a) and (e)). The areas with significant skill scores are also much smaller. For the lead times of 4, 7, and 10 months,the skill scores of the full-field hindcasts in the equatorial CEP decrease even faster than in the anomaly hindcasts(Figure 1(f)–(h)). However, the long persistence of high skill scores in the tropical western Pacific, western Atlantic, and Indian Ocean is also seen in the full-field hindcasts.

3.2. Predictive skill for temporal evolution of Ni?o3.4 and El Ni?o Modoki indices

The Ni?o3.4 index, defined as the area-averaged SSTAs in the equatorial CEP (5°N–5°S, 170°–120°W), is commonly used to measure the intensity of conventional ENSO events (Barnston, Chelliah, and Goldenberg 1997). Ashok et al. (2007) defined an El Ni?o Modoki index (EMI) as the area-averaged SSTA over the equatorial central Pacific(165°E–140°W, 10°S–10°N) minus half of the sum of SSTAs in the equatorial far eastern Pacific (110°–70°W, 15°S–5°N)and equatorial western Pacific (125°–145°E, 10°S–20°N),which characterizes the sandwich structure of El Ni?o Modoki. The predictive skill scores of the hindcast runs in the Ni?o3.4 index and EMI are evaluated.

The Ni?o3.4 index values predicted by the anomaly hindcasts are significantly correlated with those in the observation, with correlation coefficients reaching 0.84,0.67, 0.56, and 0.42 at 1-, 4-, 7-, and 10-month lead time,respectively (Table 1). The strongest El Ni?o event in 1997/1998, and La Nina event in 1988/1989, are successfully predicted at the 10-month lead time. Most moderate and weak ENSO events are reproduced at the 4- and 7-month lead times. The intensities of the predicted El Ni?o events are underestimated in the hindcast runs at the 7- and 10-month lead times, while the intensities of the predicted La Ni?a events are close to those in the observation (Figure 2(a)). As a result, the positive skewness of ENSO in the observation, representing that El Ni?o tends to be stronger than La Ni?a (An and Jin 2004), is underestimated in the anomaly hindcasts at the 7-month lead time by about 45%, and even wrongly simulated in sign at the 10-month lead time.

A previous study noted that the predictive skill for ENSO after 2000 has reduced (Barnston et al. 2012), because of the variability reduction of ENSO events during the period(Hu et al. 2013; McPhaden 2012). However, such a decline in skill is not seen in the anomaly hindcasts of IAP-DecPreS.The temporal correlation skill scores for the Ni?o3.4 index after 2000 are even somewhat higher that before 2000.The correlation coefficients are 0.83, 0.66, 0.54, and 0.43(0.85, 0.7, 0.59, and 0.42) at the 1-, 4-, 7-, and 10-month lead time for the period before (after) 2000. We calculated the correlation accuracies of the anomaly initialization runs for sea surface height (SSH) anomalies in the CEP over the periods 1980–2000 and 2001–2015, separately. The correlation accuracy increases from 0.90 to 0.94. It is speculated that the increase in skill in the anomaly hindcasts is associated with more accurate initial conditions due to a considerable increase in assimilated ocean observation records after the implementation of the Argo project (supplementary material).

The correlation skill scores of the full-field hindcasts for the Ni?o3.4 index are 0.85, 0.58, 0.39, and 0.24 at the 1-,4-, 7-, and 10-month lead time, respectively–lower than those of the anomaly hindcasts, except at the 3-month lead time (Table 1). Though the strongest El Ni?o and La Ni?a events are predicted, the full-field hindcasts show much lower skill than the anomaly hindcasts in moderate and weak events (Figure 2(b)).

For the EMI, the correlation skill scores of the anomaly hindcasts are 0.76, 0.62, 0.53 and 0.43 at the 1-, 4-, 7-, and 10-month lead time, respectively (Table 1). The strong ElNi?o Modoki event in 2009/2010 is predicted up to the 10-month lead time, though the predicted intensity is weaker than that in the observation (Figure 2(c)). As for the Ni?o3.4 index, the predictive skill scores for the EMI index after 2000 are somewhat higher than those before 2000.

Table 1. Correlation skill scores of the Ni?o3.4 index and El Ni?o Modoki index at the 1-, 4-, 7-, and 10-month lead time predicted by the anomaly and full-field hindcasts, separately.

The correlation skill scores of the full-field hindcasts for the EMI index are 0.68, 0.38, 0.1, and ?0.14 at the 1-,4-, 7-, and 10-month lead time, respectively–much lower than those of the anomaly hindcasts (Table 1). In addition,the intensities of the EMI predicted by the full-field hindcasts are far weaker than those in the observation and the anomaly hindcasts (Figure 2(d)).

Figure 2. Time series of Ni?o3.4 index values predicted by the (a) anomaly and (b) full-field hindcasts. Black lines denote observations. Red, blue, purple, and green lines denote the 1-, 4-,7-, and 10-month lead times, respectively. (c, d) As in (a, b) but for the El Ni?o Modoki index.

Figure 3. (a–d) Temporal correlation skill scores of time series of Ni?o3.4 index values as a function of forecast lead time for hindcast runs initiated from (a) February, (b) May, (c) August, and (d) November. Blue and red lines denote anomaly and full-field hindcasts,respectively. Bars denote ranges of best and worst skill scores of individual members. Black lines denote persistence predictions. (e–h)As in (a–d) but for the El Ni?o Modoki index.

The temporal correlation skill scores of the Ni?o3.4 index initiated from February, May, August, and November as a function of lead time are shown separately in the left-hand panels of Figure 3. The most prominent feature is that the skill scores of the anomaly hindcasts are higher than the full-field hindcasts for all four initiating months. To objectively evaluate the skill of the dynamical model, persistence predictions were conducted. For example, for the persistence prediction initiated from February, the state in January was taken as a persistent state. The anomaly hindcasts initiated from February, May, August, and November outperform the corresponding persistence predictions at the 3-, 2-, 8-, and 6-month lead time, respectively. The persistence prediction beats the model prediction at short lead times probably due to the inaccuracy of ocean initial conditions (Zhou and Zeng 2001). We calculated the annual cycle of the correlation accuracies of the anomaly initialization runs for SSH anomalies in the equatorial CEP(figure not shown). The accuracies of the initial conditions reached the lowest level in July, which suggests that the initial ocean subsurface states have the largest biases in July and thus cause the lowest skill scores of the hindcast initiated from August relative to the corresponding persistence prediction. There are marked declines in skill scores at the lead times of 1–4, 7–10, and 4–7 months, for both the persistent and model predictions initiated from February, August, and November, respectively (Figure 3(a),(c) and (d)). The declines in skill are associated with the‘spring prediction barrier’ of ENSO (Jin et al. 2008). Finally, it is worth noting that the skill scores of ensemble means are higher than those of most individual ensemble members,suggesting that the multi-member ensemble mean is an effective way to improve the skill and reduce the uncertainty of ENSO predictions (Zheng and Zhu 2016).

For the EMI, the skill scores of the anomaly hindcasts initiated from any month are higher than their counterparts in the full-field hindcasts (Figure 3(e)–(h)). The anomaly hindcasts initiated from February, May, and November outperform the corresponding persistence predictions at the 3-, 4-, and 6-month lead time, respectively (Figure 3(e),(f), and (h)). The skill scores of the hindcasts initiated from August are lower than those of the persistence predictions at all lead times (Figure 3(g)). Compared with the Ni?o3.4 index, both the model and persistence predictions of the EMI do not show a clear prediction barrier feature.

3.3. Predictive skill for spatial patterns of ENSO during boreal winter

Figure 4. (a–c) Boreal winter-mean SST (shading; units: K) and SLP (contours; units: Pa) anomalies regressed against the simultaneous normalized Ni?o3.4 index for the (a) observation and (b, c) anomaly hindcasts at (b) one-season and (c) two-season lead time. (d–f) As in (a–c) but for the precipitation (shading; units: mm d?1) and 850-hPa wind anomalies (vectors, units: m s?1).

Typical conventional ENSO events show a strong phase-locking feature and tend to reach mature phase during boreal winter (Rasmusson and Carpenter 1983).The relationship between El Ni?o Modoki and the seasonal cycle is much more complicated than that of conventional El Ni?o. Boreal winter is one of two dominant peak phases of El Ni?o Modoki (Weng et al. 2007). The predictive skill of IAP-DecPreS in terms of large-scale SST, precipitation and low-level circulation anomalies during ENSO (El Ni?o Modoki) peak winter are evaluated specifically. Above,we demonstrated that the anomaly hindcasts have much greater skill than the full-field hindcasts in terms of the temporal evolution of both ENSO and El Ni?o Modoki.Hence, we focus on the former in this section.

For conventional El Ni?o, both the warm SSTAs in the equatorial CEP and V-shaped cold SSTAs to the west are reproduced well by the hindcasts at the 1- and 4-month lead time (Figure 4(a)–(c)). The basin-wide warming of the tropical Indian Ocean is also reproduced. The major discrepancies of the hindcasts are that the warm SSTAs along the western coast of North America, the South China Sea, and the Kuroshio extension are not reproduced. The predicted warm SSTAs in the equatorial CEP extend excessively westward compared with those in the observation.

During El Ni?o mature winter, precipitation over the equatorial CEP is enhanced by underlying warm SSTAs(Figure 4(d)). The positive precipitation anomalies stimulate twin Rossby-wave-like cyclonic circulation anomalies to the west, symmetric about the equator in terms of the Gill model. The tropical western North Pacific is dominated by an anomalous anticyclone, referred to as the WNPAC.The WNPAC increases the precipitation over southeastern China. The convection over the tropical eastern Indian Ocean is suppressed by the remote forcing from the equatorial CEP, though the SSTAs in the tropical Indian Ocean evolve to the basin-wide warming. The extratropical eastern North Pacific is dominated by an anomalous cyclone(low pressure; Figure 4(a) and (c)), which is associated with the Pacific North American teleconnection pattern(Wallace and Gutzler 1981). All these features are reproduced by the anomaly hindcasts at the 1- and 4-month lead time (Figure 4(b), (c), (e), and (f)). The major discrepancies are that: (1) the predicted positive precipitation anomalies over the equatorial CEP are shifted westward relative to those in the observation, corresponding to the underlying excessively extended warm SSTAs; and (2) the intensities of the low-pressure anomalies over the extratropical eastern North Pacific in the hindcasts are much weaker than those in the observation, which is associated with the fact that over the midlatitude North Pacific the atmospheric circulation anomalies are modulated by unpredictable internal high-frequency variability (Pierce 2001).

For El Ni?o Modoki, the major features of the large-scale SST, precipitation, and low-level circulation anomalies in winter are reproduced well by the anomaly hindcast at the 1- and 4-month lead time (supplementary material).

4. Conclusions

In this study, we evaluated the performances of the IAP’s near-term climate prediction system, IAP-DecPreS, which is based on the CGCM FGOALS-s2 and the EnOI-IAU initialization scheme, in ENSO prediction. The skill scores of hindcasts initiated from two distinct initialization approaches–anomaly and full-field initialization–were compared. The major conclusions can be summarized as follows:

(1) The anomaly hindcasts show higher predictive skill than the full-field hindcasts for SSTAs in most global ocean areas at lead times from 1 to 10 months. For both the Ni?o3.4 and El Ni?o Modoki indices, the anomaly hindcasts have higher predictive skill than the full-field hindcast at most lead times. Hence, for the current IAPDecPreS based on FGOALS-s2, anomaly initialization is superior to full-field initialization in terms of ENSO prediction.

(2) The ensemble mean results have predictive skill close to those individual ensemble members with highest skill, for both the Ni?o3.4 and El Ni?o Modoki indices. This indicates that the ensemble mean is an effective way to improve the prediction skill and reduce the uncertainty.

(3) The predictive skill for ENSO is dependent on the initiating month. Both model and persistence predictions for the Ni?o3.4 index initiated from February, August, and November experience declines in skill at the 1–4-, 7–10-, and 4–7-month lead times, respectively, due to the so-called spring prediction barrier of ENSO.

(4) The anomaly hindcasts at the 1- and 4-month lead time reproduce the major features of largescale SST, precipitation, and low-level circulation anomalies during ENSO (El Ni?o Modoki) winter.Impressively, the anomalous anticyclone over the tropical western North Pacific and positive precipitation anomalies over southeastern China are realistically predicted, suggesting that the prediction system has potential in the seasonal prediction of the western North Pacific–East Asian winter monsoon.

The results of this study suggest that, for the current IAP-DecPreS system based on FGOALS-s2, the anomaly initialization method is superior to full-field initiation. It is speculated that this superiority is associated with the following reason: The method of anomaly initialization only assimilates the anomaly field, and thus preserves the model’s preferred climatology, which can effectively reduce the initial shocks in the hindcast/forecast runs.However, this result is model-dependent. For example,Smith, Eade, and Pohlmann (2013) reported that full-field initialized hindcasts are more skillful than anomaly initialized hindcasts. What mechanisms are responsible for the differences in skill between the two initialization approaches deserves further study.

There are many other interesting questions that also deserve further study based on IAP-DecPreS. For example,Zheng, Hu, and L’Heureux (2016) found that the decaying phase of ENSO is more predictable than its developing phase. Thus, an interesting line of research in the future would be to evaluate the impacts of the two different initialization approaches on the predictive skill for ENSO in its different phases.

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 number 2017YFA0604201), the National Natural Science Foundation of China (grant numbers. 41661144009 and 41675089), and the R&D Special Fund for Public Welfare Industry (meteorology)(grant number GYHY201506012).

Aceituno, P. 1992. “El Ni?o, the Southern Oscillation, and ENSO: Confusing Names for a Complex Ocean-atmosphere Interaction.”Bulletin of the American Meteorological Society73(4): 483–485.

Adler, R. F., G. J. Huffman, A. Chang, R. Ferraro, P. P. Xie, J. Janowiak,B. Rudolf, et al. 2003. “The Version-2 Global Precipitation Climatology Project (GPCP) Monthly Precipitation Analysis(1979 Present).”Journal of Hydrometeorology4 (6): 1147–1167.

Alexander, M. A., I. Bladé, M. Newman, J. R. Lanzante, N. C. Lau,and J. D. Scott. 2002. “The Atmospheric Bridge: The Influence of ENSO Teleconnections on Air-Sea Interaction over the Global Oceans.”Journal of Climate15 (16): 2205–2231.

An, S. I., and F. F. Jin. 2004. “Nonlinearity and Asymmetry of ENSO.”Journal of Climate17 (12): 2399–2412.

Anderson, D., T. Stockdale, M. Balmaseda, L. Ferranti, F. Vitart,R. Doblasreyes, R. Hagedorn, et al. 2002. “Comparison of the ECMWF Seasonal Forecast System 1 and 2, including the Relative Performance for the 1997/98 El Nino.”Logic Colloquium211–227.

Ashok, K., S. K. Behera, S. A. Rao, H. Weng, and T. Yamagata. 2007.“El Ni?o Modoki and Its Possible Teleconnection.”Journal of Geophysical Research112 (C11): 505.

Bao, Q., P. Lin, T. Zhou, L. Liu, J. Liu, Q. Bao, S. Xu, and W.Huang. 2013. “The Flexible Global Ocean-atmosphere-land System Model, Spectral Version 2: FGOALS-s2.”Advances in Atmospheric Sciences30 (3): 561–576.

Barnston, A. G., M. Chelliah, and S. B. Goldenberg. 1997.“Documentation of a Highly ENSO-Related SST Region in the Equatorial Pacific.”Atmosphere-Ocean35 (3): 367–383.

Barnston, A. G., M. K. Tippett, M. L. L’Heureux, S. Li, and D. G.DeWitt. 2012. “Skill of Real-time Seasonal ENSO Model Predictions during 2002–2011: Is Our Capability Increasing?”Bulletin of the American Meteorological Society93 (5): 631–651.

Cane, M. A., S. E. Zebiak, and S. C. Dolan. 1986. “Experimental Forecasts of El Ni?o.”Nature322 (6073): 827–832.

Dee, D. P., S. M. Uppala, A. J. Simmons, P. Berrisford, P. Poli, S.Kobayashi, U. Andrae, M. A. Balmaseda, G. Balsamo, and P.Bauer. 2011. “The ERA–Interim Reanalysis: Configuration and Performance of the Data Assimilation System.”Quarterly Journal of the Royal Meteorological Society137 (656): 553–597.

Harrison, D. E., and N. K. Larkin. 1998. “El Ni?o-southern Oscillation Sea Surface Temperature and Wind Anomalies,1946–1993.”Reviews of Geophysics36 (3): 353–399.

Hu, Z.-Z., A. Kumar, H. L. Ren, H. Wang, M. L’Heureux, and F.-F. Jin.2013. “Weakened Interannual Variability in the Tropical Pacific Ocean since 2000.”Journal of Climate26 (8): 2601–2613.

Huang, B., V. F. Banzon, E. Freeman, J. Lawrimore, W. Liu, and T. C. Peterson. 2015. “Extended Reconstructed Sea Surface Temperature Version 4 (ersst.v4). Part I: Upgrades and Intercomparisons.”Journal of Climate28 (3): 911–930.

Jin, E. K., J. L. Kinter, B. Wang, C. K. Park, I. S. Kang, B. P. Kirtman, J. S.Kug, A. Kumar, J. J. Luo, and J. Schemm. 2008. “Current Status of ENSO Prediction Skill in Coupled Ocean–Atmosphere Models.”Climate Dynamics31 (6): 647–664.

Kug, J. S., F. F. Jin, and S. I. An. 2009. “Two Types of El Ni?o Events:Cold Tongue El Ni?o and Warm Pool El Ni?o.”Journal of Climate22 (22): 1499–1515.

Latif, M., D. Anderson, T. Barnett, M. Cane, R. Kleeman, A. Leetmaa,J. O’Brien, A. Rosati, and E. Schneider. 1998. “A Review of the Predictability and Prediction of ENSO.”Journal of Geophysical Research Oceans103393 (30): 375–314.

Luo, J. J., and T. Yamagata. 2005. “Seasonal Climate Predictability in a Coupled OAGCM Using a Different Approach for Ensemble Forecasts.”Journal of Climate18 (21): 4474–4497.

McPhaden, M. J. 2012. “A 21st Century Shift in the Relationship between ENSO SST and Warm Water Volume Anomalies.”Geophysical Research Letters39 (9): 9706.

McPhaden, M. J., S. E. Zebiak, and M. H. Glantz. 2006. “ENSO as an Integrating Concept in Earth Science.”Science314 (5806):1740.

Pierce, D. W. 2001. “Distinguishing Coupled Ocean–Atmosphere Interactions from Background Noise in the North Pacific.”Progress in Oceanography49 (1-4): 331–352.

Rasmusson, E. M., and T. H. Carpenter. 1983. “The Relationship between Eastern Equatorial Pacific Sea Surface Temperatures and Rainfall over India and Sri Lanka.”Monthly Weather Review111 (111): 517.

Smith, D. M., R. Eade, and H. Pohlmann. 2013. “A Comparison of Full-field and Anomaly Initialization for Seasonal to Decadal Climate Prediction.”Climate Dynamics41 (11–12): 3325–3338.

Wallace, J. M., and D. S. Gutzler. 1981. “Teleconnections in the Geopotential Height Field during the Northern Hemisphere Winter.”Monthly Weather Review109: 784–804.

Wang, B., R. Wu, and X. Fu. 2000. “Pacific–East Asian Teleconnection: How Does ENSO Affect East Asian Climate?”Journal of Climate13 (9): 1517–1536.

Webster, P. J., V. O. Maga?a, T. N. Palmer, J. Shukla, R. A. Tomas,M. Yanai, and T. Yasunari. 1998. “Monsoons: Processes,Predictability, and the Prospects for Prediction.”Journal of Geophysical Research: Oceans1031 (C7): 14451–14510.

Weng, H., K. Ashok, S. K. Behera, S. A. Rao, and T. Yamagata. 2007.“Impacts of Recent cp El Ni?o on Dry/Wet Conditions in the Pacific Rim during Boreal Summer.”Climate Dynamics29 (2):113–129.

Wu, B., T. J. Zhou, and Q. Sun. 2017. “Impacts of Initialization Schemes of Oceanic States on the Predictive Skills of the IAP near-Term Climate Prediction System.”Advances in Earth Science32 (4): 342–352.

Yeh, S. W., J. S. Kug, B. Dewitte, M. H. Kwon, B. P. Kirtman, and F. F.Jin. 2009. “El Ni?o in a Changing Climate.”Nature461 (7263):511.

Zheng, F., and J. Zhu. 2010. “Coupled Assimilation for an Intermediated Coupled ENSO Prediction Model.”O(jiān)cean Dynamics60 (5): 1061–1073.

Zheng, F., and J. Zhu. 2015. “Roles of Initial Ocean Surface and Subsurface States on Successfully Predicting 2006–2007 El Ni?o with an Intermediate Coupled Model.”O(jiān)cean Science11(1): 187–194.

Zheng, F., and J. Zhu. 2016. “Improved Ensemble-mean Forecasting of ENSO Events by a Zero-mean Stochastic Error Model of an Intermediate Coupled Model.”Climate Dynamics2016: 1–15.

Zheng, F., J. Zhu, R. H. Zhang, and G. Q. Zhou. 2006. “Ensemble Hindcasts of SST Anomalies in the Tropical Pacific Using an Intermediate Coupled Model.”Geophysical Research Letters33(19): 318–37.

Zheng, Z., Z. Z. Hu, and M. L’Heureux. 2016. “Predictable Components of ENSO Evolution in Real-time Multi-model Predictions.”Scientific Reports6: 35909.

Zhou, G., and Q. Zeng. 2001. “Predictions of Enso with a Coupled Atmosphere-Ocean General Circulation Model.”Advances in Atmospheric Sciences18 (4): 587–603.

Zhou, T. J., Y. Q. Yu, Y. M. Liu, and B. Wang. 2014. “Flexible Global Ocean-atmosphere-land System Model: A Modeling Tool for the Climate Change Research Community.”Springer-Verlag,Berlin Heidelberg483: 217–224. doi:10.1007/978-3-642-41801-3.