Advertisement

A comparison of full-field and anomaly initialization for seasonal prediction of Indian Ocean basin mode

  • Shuai Hu
  • Bo WuEmail author
  • Tianjun Zhou
  • Zhun Guo
Open Access
Article

Abstract

The full-field initialization (FI) and anomaly initialization (AI) are two currently existing initialization techniques for seasonal climate predictions. Knowledges on the skill dependence of seasonal predictions of Indian Ocean basin mode (IOBM) on these two initialization approaches are important but remains unknown. In this study, we try to fill in the knowledge gap by performing hindcasts with a near-term climate prediction system named IAP-DecPreS. We find that anomaly initialized hindcasts (Hindcast-A) show higher skill than full-field initialized hindcasts (Hindcast-F) in the predictions of IOBM. Initialized from November, both the types of hindcasts reasonably predict the Indian Ocean basin-wide warming during ENSO mature winter, but only the Hindcast-A successfully reproduces the persistence of the IOBM to the following summer. The differences in the hindcast skill arise in developing stage of the IOBM, during which the Hindcast-F underestimates the warming tendency in the southern tropical Indian Ocean (TIO). The discrepancy is associated with both ENSO remote forcing and local air-sea interaction processes. ENSOs predicted by Hindcast-F decay faster than those in the observations and the Hindcast-A, leading to an underestimation of ENSO forcing to the Indian Ocean. In addition, both the positive downward shortwave radiative flux and latent heat flux anomalies over the southern TIO are underestimated by the Hindcast-F due to the modulations of the model climatology. These results imply that the FI scheme is unable to maintain the inherent physical processes responsible for the IOBM formation in the coupled model, although it can improve model climatology in the Indian Ocean.

Keywords

Seasonal prediction Indian Ocean basin mode Anomaly initialization Full-field initialization Coupled General Circulation Model 

1 Introduction

Indian Ocean basin mode (IOBM), which features basin-wide warming or cooling in the tropical Indian Ocean (TIO), is the leading mode of the TIO sea surface temperature anomalies (SSTAs) on the interannual time scale (Klein et al. 1999; Alexander et al. 2002; Du et al. 2009; Schott et al. 2009; Tao et al. 2014). The IOBM usually forms in ENSO mature winter, peaks in following spring, and persists through summer (Alexander et al. 2002; Lau and Nath 2003; Schott et al. 2009), which has striking impacts on the regional and global climate. IOBM can prolong El Niño’s influence on the Asian climate during El Niño decaying summer when the equatorial central-eastern Pacific positive SSTAs have returned to normal or changed to negative. It maintains an anomalous anticyclone over the northwestern Pacific through driving eastward propagating atmospheric Kelvin waves (Yang et al. 2007; Wu et al. 2009; Xie et al. 2009). The IOBM can also influences Indian Summer Monsoon precipitation (Yang et al. 2007; Chowdary et al. 2017). Because of its long persistence and important climate impacts, the IOBM was treated as a source of northwestern Pacific and East Asia climate predictability (Chowdary et al. 2011).

The IOBM is driven by the ENSO remote forcing through two different mechanisms, that is the “atmospheric bridge” (Alexander et al. 2002; Li et al. 2003) and the tropospheric temperature (TT) mechanisms (Chiang and Sobel 2002; Chiang and Lintner 2005). The “atmospheric bridge” mechanism suggests that El Niño weakens convection and surface wind over the TIO through modulating the Walker circulation, which produces greater downward solar radiation and less upward latent heat flux, contributing to Indian Ocean basin-wide warming (Lau and Nath 1996; Klein et al. 1999; Alexander et al. 2002). The “TT” mechanism suggests that the signal of El Niño-driven enhanced convective heating propagates eastward through Kelvin waves and thus warm the entire tropical troposphere. The warm atmosphere leads to the SST warming through moist convective adjustment (Chiang and Sobel 2002; Chiang and Lintner 2005). In addition to the ENSO remote forcing, ocean wave dynamics associated with the preceding IOD also play a critical role in reinforcing the eastern Indian Ocean warming in winter (Li et al. 2002; Hong et al. 2010). Besides, both the observational (Du et al. 2009) and prediction studies (Chowdary et al. 2010) suggest that local air-sea interactions in the TIO have contributions to the IOBM persistence into summer. During El Niño mature winter, anticyclonic anomalies over the southeastern TIO force oceanic downwelling Rossby waves. This downwelling Rossby wave propagates westward and suppresses upwelling in the thermocline dome (Masumoto and Meyers 1998; Yu et al. 2005), which produces warming of the southwestern TIO SST (Xie et al. 2002). This warming increases the atmospheric convection in boreal spring and anchors an antisymmetric atmospheric circulation pattern across the equator, causing a second SST warming north of the equator through wind-evaporation-SST (WES) feedback after the southwesterly summer monsoon establishes in May (Du et al. 2009).

Because of the complexity of IOBM formation and maintenance mechanisms, realistic simulations of the IOBM in climate models is still a critical challenge. Saji et al. (2006) showed that most of the Coupled Model Intercomparison Project Phase Three (CMIP3) models can reproduce the IOBM associated with the ENSO with a delay of a few months, and models without the IOBM usually have large biases in ENSO simulations. Du et al. (2013) further investigated and found that only half of the CMIP5 models can capture the key processes of the IOBM, including the local air-sea interactions that account for the IOBM persistence into summer. Biases in simulations of the IOBM and its capacitor effect in CMIP3/CMIP5 models have also been investigated in a recent study (Tao et al. 2016).

A realistic simulation of IOBM is just a first and necessary precondition toward predicting IOBM using CGCMs. Data initialization is another key step in dynamical seasonal forecasts (Doblas-Reyes et al. 2013). Assimilations of observations for model initializations force model far away from its inherent climatology, because of model biases in both the mean state and interannual variability, and their complex interactions (Srinivas et al. 2018). Hence, in free runs for prediction, the model will rapidly return back to its preferred state, which is known as initial shock (Balmaseda and Anderson 2009). The model climate drift induced by initial shock is a severe problem that may contaminate the forecast signal (Smith et al. 2007).

To solve the problem of the initial shock, initialization techniques for seasonal-to-decadal climate predictions fall into two main approaches: full-field initialization (FI) and anomaly initialization (AI) (Smith et al. 2013). The FI obtains the initial model state close to real state, with initial error efficiently reduced. However, due to the model deficiencies, once the model begins the free run for prediction, its trajectory drifts away from the observations and back to its preferred state. Thus, bias correction needs to be conducted to the raw predictions by using the historical forecasts to removing the lead time dependent model biases (CMIP–WGCM–WGSIP Decadal Climate Prediction Panel 2011). However, this correction strategy neglects the nonlinearity between the model biases and predictive anomalies, so residual biases still emerge in the predictions. In addition, whether the bias correction based on the historical forecasts could be applied to the real-time or future forecasts is still unclear. The problem of initial shock can be partly overcome by the AI, in which only observational anomalies are assimilated but model climatology generally remains unaltered (Schneider et al. 1999; Pierce et al. 2004). Although the initial errors of the AI are stronger than those of the FI, the model drift is largely reduced. But the observational anomalies are often not consistent with those of model, it still requires bias correction (Magnusson et al. 2013).

Recently, a near-term climate prediction system IAP-DecPreS was constructed based on a state-of-the-art CGCM, FGOALS-s2 (Wu et al. 2018). Based on the prediction system and the hindcast experiments, we will investigate the differences in the prediction of IOBM between the AI and FI. We try to answer the following three questions. (1) How well do the IAP-DecPreS in the IOBM prediction based on the two distinct initialization approaches? (2) What are fundamental processes casing the difference in the predictive skill for the IOBM between the AI and FI?

This paper is organized as follows. Section 2 introduces the prediction system and analytical method. In Sect. 3, we show the model performance in simulation of the IOBM. Section 4 gives the results of IOBM predictive skills in seasonal hindcasts based on full-field and anomaly initialization. Finally, the causes of these differences are discussed in Sect. 5.

2 Data and methods

2.1 Climate prediction system and experiments

A near-term climate prediction system IAP-DecPreS was constructed based on a state-of-the-art fully coupled global climate model (CGCM), FGOALS-s2, which was developed by the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG) at the Institute of Atmospheric Physics (IAP), Chinese Academy of Sciences (CAS) (Bao et al. 2013; Zhou et al. 2013). In FGOALS-s2, the atmospheric component is the Spectral Atmosphere Model in IAP LASG version 2 (SAMIL2), with a horizontal resolution of approximately 2.81° (longitude) × 1.66° (latitude) and 26 vertical levels extending from the surface to 2.19 hPa (Bao et al. 2013). The oceanic component is the LASG IAP Climate System Ocean Model version 2 (LICOM2), with 30 levels in the vertical direction and a horizontal resolution of approximately 0.5° × 0.5° in the tropics that gradually increases to 1° × 1° in the extratropical zone (Liu et al. 2012). The land component is the Community Land Model version 3 (CLM3, Oleson et al. 2004). The sea ice component is the Community Sea Ice Model version 5 (CSIM5, Briegleb et al. 2004). FGOALS-s2 has been used in the Coupled Model Intercomparison Project Phase 5 (CMIP5) experiments (Zhou et al. 2014, 2018).

The FGOALS-s2 is initialized through a newly designed ocean data assimilation scheme, ensemble optimal interpolation (EnOI) and incremental analysis update (IAU), referred to as EnOI-IAU (Wu et al. 2018). The EnOI-IAU can assimilate observational sea surface temperatures (SSTs) from HadISST version 1.1 (Rayner et al. 2003) and upper-700 m subsurface ocean temperature profiles from the EN4 dataset version 1.1 (Good et al. 2013). Please see Wu et al. (2018) for detail information about EnOI-IAU and IAP-DecPreS.

In the process of the initialization, we used two different approaches, namely, full-field and anomaly initializations. Systematic ensemble seasonal hindcasts were conducted from the initial states derived from the FI (AI) runs from February, May, August and November for each year over the period 1979–2015. These hindcasts are referred to as Hindcast-F (full-field) and Hindcast-A (anomaly). For each hindcast, both Hindcast-F and Hindcast-A have 9 ensemble members and 14-month integrations. The lead time of n months represents the time interval between the initial date and the predictive target. Please see Table 1 for detailed information about the experiment results we used in this study.
Table 1

Description of the experiments used in this study

Exp name

Integration

Initial condition

Member size

External forcing

Historical simulation

1850–2005

PI-control

1

External forcing for the CMIP5 historical runs

Anomaly initialization (AI) run

January 1950–December 2015

January 1 1950 from 3 historical runs

3

As in the historical (RCP4.5) run, before (after) 2005

Full-field initialization (FI) run

As in AI

As in AI

3

As in the historical (RCP4.5) run, before (after) 2005

Hindcast-A

Starts in February, May, August and November in each year from 1979 to 2015, integrated for 14 months

Initial states produced by the AI

9

As in the historical (RCP4.5) run, before (after) 2005

Hindcast-F

As in Hindcast-A

Initial states produced by FI

9

As in the historical (RCP4.5) run, before (after) 2005

2.1.1 Observational data

The following observational and reanalysis datasets were used to assess the predictive skill: (1) observational precipitation data from the Global Precipitation Climatology Project (GPCP, Adler et al. 2003), (2) observational SST from the Met Office Hadley Centre Sea Ice and SST Dataset (HadISST version 1.1, Rayner et al. 2003), (3) circulation and tropospheric temperature data from the ERA-interim reanalysis (Dee et al. 2011). All the datasets are monthly mean and cover the period of 1979–2016.

2.2 Analysis methods

For both the Hindcast-F and Hindcast-A, the predicted anomalies are calculated with respect to the predicted climatology of 1979–2016,
$$\widehat{Y}\left( {i,t} \right) = Y\left( {i,t} \right) - \mathop \sum \limits_{k = 1}^{N} Y\left( {k,t} \right)/N$$
(1)
where \(Y\left( {i,t} \right)\) and \(\widehat{Y}\left( {i,t} \right)\) are the raw and bias-adjusted (anomalous) values for hindcast \(i\) at lead time \(t\), respectively, and \(N\) is the number of hindcasts. This is a common empirical bias correction method used in operational seasonal predictions (Barnston et al. 2017), through which the model drifts are removed as the part of the predicted climatology. Applying another bias correction method, the cross validated bias correction (Smith et al. 2013) gets the similar results.

We used the anomaly correlation coefficient (ACC) and root mean squared error (RMSE) to evaluate the predictive skill. The Niño 3.4 index is defined as the area-averaged SSTAs in the equatorial central and eastern Pacific (5°S–5°N, 170°W–120°W). The IOBM index is defined as the area-averaged SSTAs in the TIO (20°S–20°N, 40°E–100°E). The IOBM-related variable anomalies are obtained through regressing onto the normalized observed IOBM index for the observation, initialization runs and hindcast runs, so that accuracies of initial conditions and skill of prediction for the IOBM can be assessed simply through comparing spatial patterns of related variables with corresponding observations or reanalysis.

To understand physical processes responsible for the IOBM in hindcast runs, mixed-layer heat budget (MLHB) is diagnosed for both the reanalysis and Hindcast runs. The mixed layer ocean temperature equation is written as
$$\frac{{\partial T^{\prime } }}{\partial t} = D_{O}^{\prime } + Q_{net}^{\prime } + Res$$
(2)
where \(Q_{net}^{\prime }\) and \(Res\) are the net surface heat flux and the residual term, respectively, and \(D_{O}^{\prime }\) the ocean temperature transport effect due to 3-D advections. \(D_{O}^{\prime }\) consists of zonal, meridional and vertical advection anomalies.
The net surface heat flux converted to temperature units is equal to
$$Q_{net}^{\prime } = \frac{{Q_{LH}^{\prime } + Q_{SH}^{\prime } + Q_{LW}^{\prime } + Q_{SW}^{\prime } }}{{h \cdot \rho \cdot C_{p} }}$$
(3)

The net heat flux consists surface latent heat flux (\(Q_{LH}^{\prime }\)), sensible heat flux (\(Q_{SH}^{\prime }\)), longwave radiation flux (\(Q_{LW}^{\prime }\)) and shortwave radiation (\(Q_{SW}^{\prime }\)) flux. \(\rho\) and \(C_{P}\) are the density of seawater and the specific heat at constant pressure. \(h\) is the mixed layer depth.

3 IOBM simulated by the historical runs

Before assessing the predictive skill, we first evaluate IOBM simulated by the historical runs of FGOALS-s2. For the observation, the IOBM is the leading EOF mode of 9-year high-pass filtered SSTAs in the TIO, which accounts for 31.6% of the total variance (Fig. 1a). The second EOF mode is the Indian Ocean dipole mode (IOD) (Saji et al. 1999; Webster et al. 1999), which accounts for 20.0% of the total variance (Fig. 1b). For the historical runs, the spatial pattern of the EOF1 shows the basin-wide warming, with intensity stronger than that in the observation (Fig. 1c). The simulated EOF2 shows the zonal dipole structure, but its eastern pole extends westward relative to that in the observation (Fig. 1d). This bias exists in most CMIP3 and CMIP5 models, which is related to a stronger thermocline-SST feedback caused by the steeper climatological west–east slope of the equatorial thermocline in models (Cai and Cowan 2013). In addition, there is a false center of cold SSTAs in the off-equatorial southern Indian Ocean. For the historical runs, the first and second modes accounts for 19.8% and 18.3% of total variances and thus cannot be separated from each other significantly based on the North role (North et al. 1982).
Fig. 1

Left panels: the spatial patterns of the first EOF mode of monthly SSTAs in the tropical Indian Ocean (area encircled by dashed lines) derived from a observation and c FGOALS-s2 historical simulations for the period of 1976–2005 (shading, units: °C). Variance contributions are noted on the upper right corners. e Seasonal cycles of standard deviations of the normalized first principal components (PC1) time series. Right panels: b, d, f as in the a, c, e, but for the EOF2

Seasonal dependence is an important feature of both the IOBM and IOD. In the observation, the amplitudes of the PC1 time series in winter and spring are much larger than those in summer and fall and reach the peak in the February (Fig. 1e). Different from the PC1, the amplitudes of the PC2 reach the peak in fall (Fig. 1f). In the FGOALS-s2, the seasonal evolutions of the amplitudes of the PC1 and PC2 time series are generally consistent with those in the observation. The major bias is that their peaks both lag to the corresponding observational references (Fig. 1e, f).

The IOBM is driven by the ENSO remote forcing. In the observation, the lagged correlation of the PC1 with the D(0)JF(1)-mean Niño 3.4 index indicates that the IOBM forms in El Niño mature winter and persists to El Niño decaying summer (Fig. 2a) (Hereafter, numerals “0” and “1” denote the developing and decay years of El Niño, respectively). FGOALS-s2 reproduces the observed ENSO cycle and the delayed response of the IOBM (Fig. 2b). It is worth noting that the IOBM starts from January (0) in the FGOLAS-s2, lagging to that in the observation by one month, consistent with the lag shown in Fig. 1e.
Fig. 2

Lead-lag correlations of the monthly Niño 3.4 index (black line) and PC1 time series (blue line) with the D(0)JF(1)-mean Niño 3.4 index for a observation and b FGOALS-s2 historical simulations

The FGOALS-s2 can reproduce the major features of the variability modes in the tropical Indian Ocean, which is the foundation to conduct the seasonal prediction for the IOBM. Meanwhile, we will show below that the model biases, especially the one-month lag of the IOBM formation, have effects on IOBM predictions.

4 Predictive skill for the IOBM

For simplicity of calculation, we define an IOBM index as the area-averaged SSTAs over the TIO (20°S–20°N, 40°E–100°E), which agrees well with the PC1 for both the observation and the historical run, with their correlations greater than 0.87. We use two metrics, ACC and RMSE, as functions of both target season and lead time, to measure the predictive skill for the IOBM index by the seasonal prediction experiments (Fig. 3). In the observation, the IOBM generally maintains from ENSO mature winter to the following decaying summer (Fig. 2a). Hence, we focus on three target seasons, DJF, MAM and JJA.
Fig. 3

ac Anomaly correlation coefficients (ACC) as functions of target seasons (a DJF, b MAM, c JJA) and lead times (horizontal axis) for the Hindcast-A (red) and the Hindcast-F (blue), respectively. df as in ac, but for root mean squared error (RMSE). The red (blue) lines represent the ensemble means of the Hindcast-A (Hindcast-F). The vertical bars represent the member spreads. The dashed lines represent the persistence predictions

The skill scores of the Hindcast-A are higher than the Hindcast-F for all target seasons at all lead times. The Hindcast-A also has higher skill than the persistence prediction except at short lead times in terms of both ACC and RMSE. If the threshold of skilful prediction is set as the ACC of 0.5, the predictive skill for the IOBM index in DJF, MAM and JJA reaches 7, 10 and 10 months in the Hindcast-A, respectively. In contrast, Hindcast-F only has skill at around a 4-month lead for all target seasons. The ensemble-mean predictions often have higher skill than the individual members, especially on the RMSE.

The hindcasts initialized from November cover the entire life cycle of the IOBM from ENSO mature winter to decaying summer, and generally have higher predictive skill than hindcasts initiated from other months (Fig. 3). Hence, we focus on the predictions initialized from November in the following analysis. The IOBM persisting from D(0)JF(1) to JJA(1) is captured by the Hindcast-A, with its intensity being comparable with that in the observation. Meanwhile, the ENSO evolution in the Hindcast-A is highly consistent with that in the observation. The SSTAs in the equatorial central-eastern Pacific transform from positive to weak negative in JJA(1) (Fig. 4a–f). In contrast, the IOBM and ENSO-related SSTAs dissipate more quickly after D(0)JF(1) in the Hindcast-F (Fig. 4g–i).
Fig. 4

Seasonal mean SST (shading, units: °C) and 850 hPa wind (vector, units: m/s) anomalies regressed onto normalized observed JFMA(1)-mean IOBM index for ac observation, df Hindcast-A initialized from November, gi Hindcast-F initialized from November. White dots represent values reaching the 5% significance levels

5 Factors causing the differences in the predictive skill

In this section, we investigate what causes the differences in the predictive skill for the IOBM between the Hindcast-A and Hindcast-F from following three aspects, initial conditions, remote forcing of ENSO and local processes in the tropical Indian Ocean.

5.1 Initial conditions

Considering that the seasonal prediction is seen as an initial value problem (e.g., Doblas-Reyes et al. 2013), we first investigate the differences in the initial conditions produced by the AI and FI. In the observation, the TIO SSTAs in October show a dipole pattern, indicating that the IOBM is always preceded by the IOD (Fig. 5a; Hong et al. 2010; Wu et al. 2012). Both the AI and the FI simulate the IOD-like SSTA pattern (Fig. 5b, c). However, they both have false cold SSTA center in the central southern TIO split from the cold pole located to west of Sumatra, similar with that in the historical run (Fig. 1d). The bias suggests that though the observation records have been assimilated, the tropical variability is modulated by internal air-sea interactions in the model to a large extent. Generally, this bias in the AI is more severe than that in the FI. Quantitively, the pattern correlation for the TIO SSTAs between the FI (AI) and the observational reference is 0.64 (0.22). The FI offers a more accurate initial condition than the AI. However, we have shown that the predictive skill of the Hindcast-F is lower than that of the Hindcast-A. The contradiction suggests that the accuracy of initial condition cannot be the root cause of the differences in the predictive skill.
Fig. 5

SSTAs in October (0) regressed onto the normalized observed JFMA(1)-mean IOBM index for a observation; b Anomaly initialization; c Full-field initialization. The white dots denote values passing the 5% significance levels. The numbers on the upper-right corner of b and c are the pattern correlation coefficients with the observation shown in a for the area within the dashed boxes (25°S–25°N, 40°E–115°E)

5.2 Remote forcing of ENSO

Because the IOBM is primarily driven by the remote forcing of ENSO through the atmospheric bridge (e.g. Alexander et al. 2002; Li et al. 2003), we investigate the differences in the ENSO-driven anomalous Walker circulation between the Hindcast-A and the Hindcast-F. In the observation, the eastern TIO is covered by the descending branch of the anomalous Walker circulation with its center located in the western Pacific in D(0)JF(1). The negative precipitation anomalies over the eastern TIO drive low-level equatorial easterly anomalies (Fig. 6a). In the MAM(1), the ENSO-driven anomalous Walker circulation still maintains, but TIO is largely dominated by an local asymmetric mode with precipitation anomalies asymmetric about the equator and southward cross-equatorial low-level wind anomalies over the TIO (Fig. 6b), due to the local wind-evaporation-SST positive feedback (Du et al. 2009). In JJA(1), the descending branch of the anomalous Walker circulation moves to the northwestern Pacific, while the TIO is dominated by anomalous ascending motions (Fig. 6c).
Fig. 6

Seasonal mean precipitation (shading, units: mm/day), 850 hPa wind (vector, units: m/s) and 200 hPa potential velocity (contour, units: m2/s, interval value: 1 × 106) anomalies regressed onto the normalized observed D(0)JF(1)-mean Niño 3.4 index for (a–c) observation, (d–f) Hindcast-A initialized from November, (g–i) Hindcast-F initialized from November

Both the types of the hindcasts capture the evolution of precipitation and large-scale circulation anomalies associated with the ENSO during D(0)JF(1), but the suppressed precipitation over the tropical southern TIO extends farther westward than that in the observation (Fig. 6d, g). During MAM(1) and JJA(1), the Hindcast-A provides more skillful predictions of the ENSO turnabout process (Fig. 4d–i), and the associated “atmospheric bridge” (Fig. 6d–i). In contrast, for the Hindcast-F, the large-scale descending branch of the anomalous Walker circulation moves eastward too fast due to unrealistic fast decay of ENSO (Fig. 4g–i). Mixed-layer temperature budget indicates that the fast decay of El Niño in Hindcast-F is primarily caused by the stronger negative anomalous thermal vertical advection by the mean currents (\(- \overline{\text{w}} \frac{{\partial {\text{T}}^{\prime } }}{{\partial {\text{z}}}})\), suggesting that the faster decay of thermocline anomalies may be the major reason of the lower predictive skill for the decaying rate of El Niño in Hindcast-F (figure not shown).

5.3 Local processes in the Indian Ocean

In this subsection, we try to understand how the differences in the predictive skill form and grow, through investigating the local processes in the tropical Indian Ocean.

In both the initial conditions [Oct(0)] and first month of the hindcast integrations [Nov(0)], the basin-averaged TIO SSTAs from both the Hindcast-A and Hindcast-F are much colder than those in the observation (Fig. 7). In the observation, the life cycle of the IOBM can be simply separated into three stages, developing [ND(0)JF(1)], maintaining [MAM(1)] and decaying [JJA(1)] stage. The major difference in the temporal evolutions between the two types of hindcasts forms in the developing stage. The growth rate of the IOBM intensity in the Hindcast-A is much larger than that in the Hindcast-F and the observation, while the IOBM decaying rates in the both hindcasts are close to that in the observation. As a result, the IOBM intensity in the Hindcast-A is close to that in the observation after its peak, while that in the Hindcast-F is much weaker during its entire life cycle (Fig. 7).
Fig. 7

Temporal evolutions of area-averaged SSTAs in the tropical Indian Ocean (20°S–20°N, 40°E–100°E) regressed onto the normalized observed JFMA(1)-mean IOBM index for observations (black line), Hindcast-A (red line) and Hindcast-F (blue line), respectively. Thin and thick lines represent individual ensemble members and ensemble means. Red and blue dots represent initial states of the Hindcast-A and Hindcast-F derived from the AI and FI, respectively

During the developing stage, difference in time tendency of the TIO SSTAs between the Hindcast-A and the Hindcast-F is mainly found in the southern TIO, with the former approximately two times larger than the latter (Fig. 8). To investigate what physical processes causing the difference, MLHB analyses are performed for the two types of the hindcasts in the area of 20°S–0°, 40°E–110°E (box in Fig. 8). The stronger warming in the southern TIO in the Hindcast-A is primarily contributed by surface heat flux (\(Q_{net}^{\prime }\)) (Fig. 9a). Previous studies noted that the ocean wave dynamics plays an essential role in the warming tendencies in the southwestern TIO, where thermocline is shallow (Xie et al. 2002). However, the difference in the \(- \overline{\text{w}} \frac{{\partial {\text{T}}^{\prime } }}{{\partial {\text{z}}}}\) (dominant term in the ocean dynamics) between the two types of the hindcasts is far smaller than the differences in the surface fluxes (Fig. 9a), and thus play a secondary role in their different predictive skill. For both types of the Hindcasts, the anomalous anticyclone over the southern TIO (Fig. 6d, g) excites westward propagating downwelling Rossby waves during the IOBM developing stage [ND(0)JF(1)]. The warm subsurface temperature anomalies in the Hindcast-A are shallower than those in the Hindcast-F, while the mean upward motions in the Hindcast-A are stronger than those in the Hindcast-F (figure not shown). As a result, their magnitudes of anomalous warm vertical advections by mean upward motions (\(- \overline{\text{w}} \frac{{\partial {\text{T}}^{\prime } }}{{\partial {\text{z}}}}\)) are very close.
Fig. 8

Time tendency of mixed layer temperature anomalies during IOBM developing stage [ND(0)JF(1)] derived from regressions onto the normalized observed JFMA(1)-mean IOBM index for a Hindcast-A and b Hindcast-F (Units: °C/month). The dashed boxes represent the area of the southern TIO (20°S–0°, 40°E–110°E)

Fig. 9

a Mixed-layer heat budget analysis for the southern TIO (dashed boxes in Fig. 8) during the IOBM developing stage [ND(0)JF(1)] derived from regressions onto the normalized observed JFMA(1)-mean IOBM index for the Hindcast-A (red) and Hindcast-F (blue) (Units: °C/month). \(dT/dt\), \(Sum\), \(D_{O}\), \(Q_{net}\) represent the time tendency of mixed-layer temperature anomalies, sum of \(D_{O}\) and \(Q_{net}\), sum of the anomalous dynamic terms and the net surface flux anomalies in Eq. (1), respectively. b Components of \(Q_{net}\). \(Q_{LH}\), \(Q_{SW}\), \(Q_{SH}\), \(Q_{LW}\) represent the anomalous surface latent heat flux, shortwave radiation flux, sensible heat flux and longwave radiation flux, respectively. The vertical segments denote spreads of the ensemble members

The difference in the \(Q_{net}^{\prime }\) term is dominated by latent heat flux (\(Q_{LH}^{\prime }\)) and surface shortwave radiation flux (\(Q_{SW}^{\prime }\)), with comparable relative contributions (Fig. 9b). The \(Q_{LH}^{\prime }\) term is closely associated with surface wind speed anomalies, which are determined by both climatological and anomalous surface wind. During the IOBM developing stage, the southern TIO is covered by an anomalous anticyclone for both the Hindcast-A and the Hindcast-F, which is a Rossby wave response to the negative precipitation anomalies over the southeastern TIO driven by descending branch of the ENSO-related anomalous Walker circulation (Fig. 6d, g, Wang et al. 2003). Surface wind anomalies over the southern TIO associated with the anomalous anticyclone are very close in both the pattern and intensity between the two types of the hindcasts (Fig. 10c, d). It is conceivable that the difference in the background mean wind plays a key role. The major feature of the climatological wind over the southern TIO is northeasterly (southeasterly) wind converging toward the ITCZ from the north (south) (Fig. 10a, b). For the Hindcast-F, the anomalous southeasterly highly overlaps with the climatological southeasterly, which greatly enhances surface wind speed and thus leads to negative downward latent heat flux anomalies (Fig. 10f). In contrast, for the hindcast-A, the climatological southeasterly retreats southward for the southward shift of the ITCZ (Fig. 10a). Hence, the corresponding enhanced surface wind speed and negative downward latent heat flux anomalies are shifted southward and much weaker in the southern TIO than the counterparts in the Hindcast-F. Meanwhile, the climatological northerly to the coast of Somalia in the Hindcast-A is much stronger than that in the Hindcast-F, associated with its southward shift of the mean ITCZ. As a result, though the southerly anomalies exist in the both types of the hindcasts, the reduced wind speed and associated positive downward latent heat flux anomalies are stronger in the Hindcast-A (Fig. 10e).
Fig. 10

Left panels: a The climatological ND(0)JF(1)-mean precipitation (shading, units: mm/day) and 1000-hPa climatological winds (vectors, units: m/s) for the Hindcast-A. c ND(0)JF(1)-mean precipitation (shading, units: mm/day) and 1000 hPa winds (contour, units: m/s) anomalies regressed onto the normalized observed JFMA(1)-mean IOBM index for the Hindcast-A. e As in c, but for the surface latent heat flux (shading, units: W/m2) and the 1000 hPa wind speed (contour, units: m/s, interval value: 0.3) anomalies. Right panels: As in the left panels, but for the hindcast-F

The \(Q_{SW}^{ '}\) term is dominated by local total cloud fraction anomalies, with their spatial distributions highly consistent for both types of the hindcasts respectively (Fig. 11a, b). For the Hindcast-A, the spatial distributions of the \(Q_{SW}^{ '}\) and associated total cloud fraction anomalies over the TIO are highly similar with the local precipitation anomalies (Figs. 10c, 11a). The suppressed convection over eastern TIO reduces deep convective cloud and associated cirrostratus and cirrocumulus (high cloud) and thus increases incoming solar radiation, consistent with previous studies (Klein et al. 1999). In contrast, the positive \(Q_{SW}^{\prime }\) and the negative total cloud fraction anomalies in the Hindcast-F are restricted to the coast of Sumatra, with area much smaller than that in the Hindcast-A (Fig. 11b). It is worth noting that the spatial patterns of the precipitation anomalies over the TIO from the two types of the hindcasts are generally similar (Fig. 10c, d), while the local total cloud fraction anomalies are quite different (Fig. 11a, b), suggesting that other physical processes modulating the cloud cover in the Hindcast-F.
Fig. 11

Left panels: Regression pattern of anomalous surface shortwave radiation (shading, units: W/m2) and total cloud fraction (contour, units: %, interval value: 2) derived from a Hindcast-A and b Hindcast-F. c The differences between a and b. Right panels: as in the left panels, but for the SSTAs (shading, units: °C) and low cloud fraction (contour, units: %, interval value: 1)

Further analysis indicates that the difference in the total cloud fraction anomalies over the southern TIO are mainly contributed by the difference in low cloud (Fig. 11d–f). In the Hindcast-F, the climatological SST in the southern TIO is lower than that in the Hindcast-A by 0.5–1 °C, and thus has more stable marine boundary layer and more amount of marine stratiform cloud than the latter by about 4–8% (Fig. 12). As a result, the colder SSTAs in the southern TIO remaining from the initial condition of the Hindcast-F cause larger positive low cloud fraction response (Fig. 11e), which greatly counteract the negative high cloud fraction anomalies associated with the local negative precipitation anomalies, and thus reduce the surface shortwave radiative flux and associated warming tendency of the SSTAs in the southern TIO.
Fig. 12

Left panels: The climatological ND(0)JF(1)-mean SST (shading, units: °C) for a Hindcast-A and b Hindcast-F. c The difference between a and b. Right panels: As in the left panels, but for the climatological ND(0)JF(1)-mean low cloud fraction (shading, units: %)

6 Discussion

The IOBM intensity after the boreal winter predicted by the Hindcast-A is close to that in the observation (Fig. 7). It results from an offset of two biases, the cold bias of TIO SSTAs in the initial conditions (Fig. 5) and the overestimation of the warming tendency in the southern TIO during the IOBM developing stage (Fig. 8). The initial cold bias is speculated to be associated with lagged formation of the IOBM, an inherent discrepancy of the model as shown in the historical runs (Figs. 1e, 2b). The warming tendency in the southern TIO is associated with the shortwave radiative flux and latent heat flux anomalies, consistent with that in the observation, but with stronger intensities. As shown in Sect. 5.3, both the surface flux anomalies are modulated by background mean states.

For the Hindcast-A, only the observational anomalies are assimilated to produce the initial conditions, suggesting that its climatology should be close to model preferred mean states, but far away from observational climatology. In fact, the climatological surface wind and precipitation in the Hindcast-A are less like the observations than those in the Hindcast-F (Figs. 10a, b, 13). These results imply that when a model has discrepancies in the simulation of the IOBM, it would be a good choice to keep the model preferred mean states, which is essential for the model to maintain its inherent key physical processes responsible for the IOBM formation.
Fig. 13

The climatological ND(0)JF(1)-mean precipitation (shading, units: mm/day) and 1000 hPa winds (vector, units: m/s) in the observation

7 Conclusion

Using the seasonal hindcast experiments conducted by the IAP-DecPreS, a climate prediction system based on the initialized coupled general circulation model, we investigate the differences in the seasonal predictions for the IOBM between full-field and anomaly initialized hindcasts (Hindcast-F and Hindcast-A). The main conclusions are summarized as follows:
  1. 1.

    Based on two metrics, anomaly correlation coefficient and root mean squared error, the Hindcast-A has higher predictive skill than the Hindcast-F for nearly all target seasons and lead time. If a threshold of skillful prediction is defined as ACC > 0.5, the lead time of skillful prediction by the Hindcast-A reaches 7, 10, and 10 months for target seasons of winter, spring and summer, respectively (IOBM generally maintains from ENSO mature winter to decaying summer). In contrast, the lead time by the Hindcast-F is only 4 months for all the target seasons. For the hindcasts initialized from November, 1-month lead of the IOBM, both the types of hindcasts can predict Indian Ocean basin-wide warming during ENSO mature winter, but only the Hindcast-A reproduces the persistence of the IOBM to the following summer

     
  2. 2.

    The differences in the predicative skill between the Hindcast-F and Hindcast-A cannot be attributed to their differences in the accuracy of initial conditions. The full-field initialization runs (FI) instead even produce more accurate initial conditions for hindcasts initialized from November than the anomaly initialization runs (AI). Spatial distribution of zonal dipole pattern of SSTAs preceding to the IOBM simulated by the FI is more consistent with that in the observation than the AI.

     
  3. 3.

    The differences in the predictive skill for the IOBM between the two types of the hindcasts are closely associated with their performances in predicting ENSO persistence. ENSO in the Hindcast-F decays much faster than that in the observation and the Hindcast-A. Considering the importance of the ENSO remote forcing in the maintenance of the IOBM, the Hindcast-F underestimates the forcing signal of ENSO for the IOBM.

     
  4. 4.

    The differences in the predictive skill are also associated with their skill in capturing the local air-sea interaction processes responsible for the formation of the IOBM during ENSO mature winter. In the initial states, the basin-averaged SSTAs in both the Hindcast-F and the Hindcast-A are colder than that in the observation. However, because the warming tendency of SST in the southern TIO from the Hindcast-A (Hindcast-F) is far larger (smaller) than that in the observation, the Hindcast-A catches up the observation in the IOBM intensity after the winter. The difference in the warming tendency is caused by that the Hindcast-A (Hindcast-F) overestimates (underestimates) the positive downward latent heat flux and shortwave radiative flux anomalies due to the modulations of the climatological surface wind, and climatological SST and associated marine stratiform cloud forcing, respectively.

     

In summary, for the climate prediction system constructed based on the FGOLAS-s2 model and EnOI-IAU initialization scheme, the anomaly initialization approach is superior to the full-field approach in the IOBM prediction. Because this study is based on a single model, the conclusion should be checked in further multi-model intercomparisons. An important implication is that the seasonal climate prediction for the IOBM is not a simple initial value problem, that is, the predictive skill is not solely determined by the accuracy of initial conditions. Instead, it is closely associated with predictive skill for ENSO decaying rate and local surface heat flux anomalies modulated by the model climatology.

Notes

Acknowledgments

This work is supported by National Key Research and Development Program of China (Grant No. 2017YFA0604201), National Natural Science Foundation of China under Grant Nos. 41775091, 41661144009, 41675089 and International Partnership Program of Chinese Academy of Sciences, Grant No. 134111KYSB20160031.

References

  1. Adler RF, Huffman GJ, Chang A, Ferraro R, Xie P, Janowiak J et al (2003) The Version 2 global precipitation climatology project (GPCP) monthly precipitation analysis (1979–present). J Hydrometeorol 4(6):1147–1167.  https://doi.org/10.1175/1525-7541(2003)004%3c1147:TVGPCP%3e2.0.CO;2 CrossRefGoogle Scholar
  2. Alexander MA, Blade I, Newman M, Lanzante JR, Lau NC, Scott JD (2002) The atmospheric bridge: the influence of ENSO teleconnections on air-sea interaction over the global oceans. J Clim 15(16):2205–2231.  https://doi.org/10.1175/1520-0442(2002)015%3c2205:TABTIO%3e2.0.CO;2 CrossRefGoogle Scholar
  3. Balmaseda MA, Anderson D (2009) Impact of initialization strategies and observations on seasonal forecast skill. Geophys Res Lett 36(1):737–739.  https://doi.org/10.1029/2008GL035561 CrossRefGoogle Scholar
  4. Bao Q, Lin P, Zhou T, Liu L, Liu J, Bao Q et al (2013) The flexible global ocean-atmosphere-land system model, spectral version 2: FGOALS-s2. Adv Atmos Sci 30(3):561–576.  https://doi.org/10.1007/s00376-012-2113-9 CrossRefGoogle Scholar
  5. Barnston AG, Tippett MK, Ranganathan M, L’Heureux ML (2017) Deterministic skill of ENSO predictions from the North American Multimodel Ensemble. Clim Dyn 1:1–20.  https://doi.org/10.1007/s00382-017-3603-3 Google Scholar
  6. Briegleb BP, Bitz CM, Hunke EC, Lipscomb WH, Holland MM, Schramm JL et al (2004) Scientific description of the sea ice component in the Community Climate System Model, version three. NCAR Technical Note, NCAAR/TN-463 + STR, p 70Google Scholar
  7. Cai W, Cowan T (2013) Why is the amplitude of the Indian Ocean dipole overly large in CMIP3 and CMIP5 climate models? Geophys Res Lett 40(6):1200–1205.  https://doi.org/10.1002/grl.50208 CrossRefGoogle Scholar
  8. Chiang JCH, Lintner BR (2005) Mechanisms of remote tropical surface warming during El Niño. J Clim 18(20):4130–4149.  https://doi.org/10.1175/jcli3529.1 CrossRefGoogle Scholar
  9. Chiang JCH, Sobel AH (2002) Tropical tropospheric temperature variations caused by ENSO and their influence on the remote tropical climate. J Clim 15(18):2616–2631.  https://doi.org/10.1175/1520-0442(2002)015%3c2616:tttvcb%3e2.0.co;2 CrossRefGoogle Scholar
  10. Chowdary JS, Xie SP, Lee JY, Kosaka Y, Wang B (2010) Predictability of summer northwest Pacific climate in 11 coupled model hindcasts: local and remote forcing. J Geophys Res Atmos 115:D22121.  https://doi.org/10.1029/2010JD014595 CrossRefGoogle Scholar
  11. Chowdary JS, Xie SP, Luo JJ, Hafner J, Behera S, Masumoto Y et al (2011) Predictability of Northwest Pacific climate during summer and the role of the tropical Indian Ocean. Clim Dyn 36(3–4):607–621.  https://doi.org/10.1007/s00382-009-0686-5 CrossRefGoogle Scholar
  12. Chowdary JS, Harsha HS, Gnanaseelan C, Srinivas G, Parekh A, Pillai PA, Naidu CV (2017) Indian summer monsoon rainfall variability in response to differences in the decay phase of El Niño. Clim Dyn 48(7–8):2707–2727.  https://doi.org/10.1007/s00382-016-3233-1 CrossRefGoogle Scholar
  13. CMIP–WGCM–WGSIP Decadal Climate Prediction Panel (2011) Data and bias correction for decadal climate predictions. WCRP Rep, ICPO Publ. Series 150, pp 3. http://www.wcrp-climate.org/decadal/references/DCPP_Bias_Correction.pdf. Accessed 1 Feb 2019
  14. Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi S et al (2011) The ERA-Interim reanalysis: configuration and performance of the data assimilation system. Q J R Meteorol Soc 137:553–597.  https://doi.org/10.1002/qj.828 CrossRefGoogle Scholar
  15. Doblas-Reyes FJ, García-Serrano Javier, Lienert F et al (2013) Seasonal climate predictability and forecasting: status and prospects. Wiley Interdiscip Rev Clim Change 4(4):245–268.  https://doi.org/10.1002/wcc.217 CrossRefGoogle Scholar
  16. Du Y, Xie SP, Gang H, Hu KM (2009) Role of air-sea interaction in the long persistence of El Niño-induced north Indian Ocean warming. J Clim 22(8):2023–2038.  https://doi.org/10.1175/2008JCLI2590.1 CrossRefGoogle Scholar
  17. Du Y, Xie SP, Yang YL et al (2013) Indian Ocean variability in the CMIP5 multimodel ensemble: the Basin Mode. J Clim 26(18):7240–7266.  https://doi.org/10.1175/JCLI-D-12-00678.1 CrossRefGoogle Scholar
  18. Good SA, Martin MJ, Rayner NA (2013) EN4: quality controlled ocean temperature and salinity profiles and monthly objective analyses with uncertainty estimates. J Geophys Res Oceans 118(12):6704–6716.  https://doi.org/10.1002/2013JC009067 CrossRefGoogle Scholar
  19. Hong CC, Li T, Lin H et al (2010) Asymmetry of the Indian Ocean Basinwide SST anomalies: roles of ENSO and IOD. J Clim 23(13):3563–3576.  https://doi.org/10.1175/2010JCLI3320.1 CrossRefGoogle Scholar
  20. Klein SA, Soden BJ, Lau N (1999) Remote sea surface temperature variations during ENSO: evidence for a tropical atmospheric bridge. J Clim 12:917–932.  https://doi.org/10.1175/1520-0442(1999)012,0917:RSSTVD.2.0.CO;2 CrossRefGoogle Scholar
  21. Lau NC, Nath MJ (1996) The role of the “Atmospheric Bridge” in linking tropical Pacific ENSO events to extratropical SST anomalies. J Clim 9:2036–2057CrossRefGoogle Scholar
  22. Lau NC, Nath MJ (2003) Atmosphere-ocean variations in the Indo-Pacific sector during ENSO episodes. J Clim 16(1):3–20.  https://doi.org/10.1175/1520-0442(2003)016%3c0003:aoviti%3e2.0.co;2 CrossRefGoogle Scholar
  23. Li T, Zhang Y, Lu E, Wang D (2002) Relative role of dynamic and thermodynamic processes in the development of the Indian Ocean dipole: an OGCM diagnosis. Geophys Res Lett 29(23):25-1CrossRefGoogle Scholar
  24. Li T, Wang B, Chang CP, Zhang Y (2003) A theory for the Indian Ocean dipole–zonal mode. J Atmos Sci 60(17):2119–2135CrossRefGoogle Scholar
  25. Liu HL, Lin PF, Yu YQ, Zhang XH (2012) The baseline evaluation of LASG/IAP Climate System Ocean Model (LICOM) version 2.0. Acta Meteorol Sin 26:318–329CrossRefGoogle Scholar
  26. Magnusson L, Alonso-Balmaseda M, Corti S, Molteni F, Stockdale T (2013) Evaluation of forecast strategies for seasonal and decadal forecasts in presence of systematic model errors. Clim Dyn 41(9):2393–2409CrossRefGoogle Scholar
  27. Masumoto Y, Meyers G (1998) Forced Rossby waves in the southern tropical Indian Ocean. J Geophys Res Oceans 103(C12):27589–27602.  https://doi.org/10.1029/98jc02546 CrossRefGoogle Scholar
  28. North GR, Bell TL, Cahalan RF, Moeng FJ (1982) Sampling errors in the estimation of empirical orthogonal functions. Mon Weather Rev 110(7):699–706CrossRefGoogle Scholar
  29. Oleson KW, Dai Y, Bonan G, Bosilovich M, Dickinson R, Dirmeyer P et al (2004) Technical description of the Community Land Model (CLM) (NCAR Tech. Note, NCAR/TN-4611STR, 173 p.). Boulder, CO: National Center for Atmospheric ResearchGoogle Scholar
  30. Pierce DW, Barnett TP, Tokmakian R, Semtner A, Maltrud M, Lysne J, Craig A (2004) The ACPI project, element 1: initializing a coupled climate model from observed conditions. Clim Change 62(1):13–28CrossRefGoogle Scholar
  31. Rayner NA, Parker DE, Horton EB, Folland CK, Alexander LV, Rowell DP et al (2003) Global analyses of SST, sea ice and night marine air temperature since the late nineteenth century. J Geophys Res Atmos 108:4407.  https://doi.org/10.1029/2002JD002670 CrossRefGoogle Scholar
  32. Saji NH, Goswami BN, Vinayachandran PN, Yamagata T (1999) A dipole mode in the tropical Indian Ocean. Nature 401:360–363Google Scholar
  33. Saji NH, Xie SP, Yamagata T (2006) Tropical Indian Ocean variability in the IPCC twentieth-century climate simulations. J Clim 19(17):4397–4417.  https://doi.org/10.1175/jcli3847.1 CrossRefGoogle Scholar
  34. Schneider EK, Huang B, Zhu Z, DeWitt DG, Kinter JL, Kirtman B, Shukla J (1999) Ocean data assimilation, initialization, and predictions of ENSO with a coupled GCM. Mon Weather Rev 127:1187–1207.  https://doi.org/10.1175/1520-0493(1999)127%3c1187:ODAIAP%3e2.0.CO;2 CrossRefGoogle Scholar
  35. Schott FA, Xie SP, McCreary JP (2009) Indian Ocean circulation and climate variability. Rev Geophys 47:RG1002.  https://doi.org/10.1029/2007rg000245 CrossRefGoogle Scholar
  36. Smith DM, Cusack S, Colman AW et al (2007) Improved surface temperature prediction for the coming decade from a global climate model. Science 317(5839):796–799CrossRefGoogle Scholar
  37. Smith DM, Eade R, Pohlmann H (2013) A comparison of full-field and anomaly initialization for seasonal to decadal climate prediction. Clim Dyn 41(11–12):3325–3338.  https://doi.org/10.1007/BF02967433 CrossRefGoogle Scholar
  38. Srinivas G, Chowdary JS, Gnanaseelan C, Prasad KVSR, Karmakar A, Parekh A (2018) Association between mean and interannual equatorial Indian Ocean subsurface temperature bias in a coupled model. Clim Dyn 50(5–6):1659–1673CrossRefGoogle Scholar
  39. Tao W, Huang G, Hu K, Qu X, Wen G, Gong Y (2014) Different influences of two types of El Niños on the Indian Ocean SST variations. Theoret Appl Climatol 117(3–4):475–484.  https://doi.org/10.1007/s00704-013-1022-x CrossRefGoogle Scholar
  40. Tao W, Huang G, Hu K, Gong H, Wen G, Liu L (2016) A study of biases in simulation of the indian ocean basin mode and its capacitor effect in CMIP3/CMIP5 models. Clim Dyn 46(1–2):205–226CrossRefGoogle Scholar
  41. Wang B, Wu R, Li T (2003) Atmosphere–warm ocean interaction and its impacts on Asian-Australian monsoon variation. J Clim 16:1195–1211CrossRefGoogle Scholar
  42. Webster PJ, Moore AM, Loschnigg JP, Leben RR (1999) Coupled ocean–atmosphere dynamics in the Indian Ocean during 1997–98. Nature 401:356–360.  https://doi.org/10.1038/43848 CrossRefGoogle Scholar
  43. Wu B, Zhou TJ, Li T (2009) Seasonally evolving dominant interannual variability modes of east asian climate. J Clim 22(11):2992–3005.  https://doi.org/10.1175/2008JCLI2710.1 CrossRefGoogle Scholar
  44. Wu B, Zhou T, Li T (2012) Two distinct modes of tropical Indian Ocean precipitation in boreal winter and their impacts on equatorial western Pacific. J Clim 25(3):921–938CrossRefGoogle Scholar
  45. Wu B, Zhou TJ, Zheng F (2018) EnOI-IAU initialization scheme designed for decadal climate prediction system IAP-DecPreS. J Adv Model Earth Syst 10(2):342–356.  https://doi.org/10.1002/2017MS001132 CrossRefGoogle Scholar
  46. Xie SP, Annamalai H, Schott FA, McCreary JP (2002) Structure and mechanisms of south Indian Ocean climate variability. J Clim 15(8):864–878.  https://doi.org/10.1175/1520-0442(2002)015%3c0864:SAMOSI%3e2.0.CO;2 CrossRefGoogle Scholar
  47. Xie SP, Hafner J, Tokinaga H, Du Y, Sampe T, Hu KM et al (2009) Indian Ocean capacitor effect on Indo–Western Pacific climate during the summer following El Niño. J Clim 22(3):730–747.  https://doi.org/10.1175/2008JCLI2544.1 CrossRefGoogle Scholar
  48. Yang J, Liu Q, Xie SP, Liu Z, Wu L (2007) Impact of the Indian Ocean SST basin mode on the Asian summer monsoon. Geophys Res Lett 34(2):155–164.  https://doi.org/10.1029/2006GL028571 CrossRefGoogle Scholar
  49. Yu W, Xiang B, Liu L, Liu N (2005) Understanding the origins of interannual thermocline variations in the tropical Indian Ocean. Geophys Res Lett 32:L24706.  https://doi.org/10.1029/2005GL024327 CrossRefGoogle Scholar
  50. Zhou T, Yu Y, Liu Y, Wang B (2013) Flexible global ocean–atmosphere–land system model: a modeling tool for the climate change research community. Springer, Berlin HeidelbergGoogle Scholar
  51. Zhou T, Chen X, Dong L, Wu B, Man W, Zhang L et al (2014) Chinese contribution to CMIP5: an overview of five Chinese Models’ performances. J Meteorol Res 28(4):481–509CrossRefGoogle Scholar
  52. Zhou T, Wang B, Yu Y, Liu Y, Zheng W, Li L et al (2018) The FGOALS climate system model as a modeling tool for supporting climate sciences: an overview. Earth Planet Phys 2(4):276–291CrossRefGoogle Scholar

Copyright information

© The Author(s) 2019

Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Authors and Affiliations

  1. 1.LASG, Institute of Atmospheric Physics, Chinese Academy of ScienceBeijingChina
  2. 2.University of Chinese Academy of SciencesBeijingChina

Personalised recommendations