HTML
--> --> -->Coupled ocean-atmosphere models are major tools to simulate and predict ENSO, since ENSO arises from the air-sea interactions in the tropical Pacific Ocean (Bjerknes, 1969). In the past several decades, various coupled models with different degrees of complexity have been developed for ENSO studies, including intermediate coupled models (ICMs), hybrid coupled models (HCMs), and comprehensive coupled general circulation models (CGCMs). In recent years, CGCMs have progressed remarkably (Gualdi et al., 2003; Delworth et al., 2006; Guilyardi et al., 2009; Hurrell et al., 2013), and most current CGCMs are capable of simulating a good resemblance of the observed ENSO and associated variability. However, many CGCMs still suffer from the problem of climate drift, with considerable errors in reproducing the climatology of the equatorial Pacific, such as a too-cold cold tongue and warm SST biases in the eastern boundaries of the Pacific and Atlantic basins (Latif et al., 2001; Davey et al., 2002; Luo et al., 2005; Gupta et al., 2013; Wang et al., 2014; Richter, 2015; Zhu and Zhang, 2017, Zhu and Zhang, 2018). Among the aforementioned models, ICMs are efficient in computing, and show certain advantages for ENSO simulation and prediction. For example, compared to a complex CGCM, they can lead to ways in which coupling mechanisms are more easily understood. Furthermore, ICMs are free from the climate drift problem because they are normally designed to be anomaly models. In addition, owing to their low computational cost compared to CGCMs, ICMs allow a large number of experiments to be performed, feasibly and affordably. Famously, the Zebiak-Cane model was the first dynamical coupled model to successfully predict ENSO events (Cane et al., 1986). It is an intermediate anomaly model that describes the evolution of anomalies with respect to a prescribed seasonally varying background state. Since 1986, the Zebiak-Cane model has been used to conduct realistic ENSO simulations, and is widely used in dynamics and predictability studies of ENSO (Zebiak and Cane, 1987; Chen et al., 1995, Chen et al., 2004; Mu et al., 2007a, b; Duan et al., 2014; Hou et al., 2017).
(Zhang et al., 2003) developed an improved ICM on the basis of an intermediate ocean model (IOM) designed by (Keenlyside and Kleeman, 2002). This ICM has been used at the Institute of Oceanology, Chinese Academy of Sciences (IOCAS), and is referred to as IOCAS ICM. It is an anomaly model, including an IOM with an embedded nonlinear SST anomaly model describing the thermodynamics for the surface mixed layer, and a statistical atmosphere model. Being constructed by a singular value decomposition (SVD) analysis technique, the atmosphere model determines the wind stress interannual anomaly from the ocean model SST anomaly. The SVD analysis is conducted to derive the relationship between the historical wind stress anomalies and SST anomalies, in which the wind stress data are the ensemble mean of 24-member ECHAM4.5 simulations and hence filter out unrelated atmospheric noise. Another striking feature of the ICM is an explicit parameterization of the subsurface entrainment temperature (Te) in terms of the thermocline variability based on SVD analysis (Zhang and Gao, 2016a, b). IOCAS ICM not only realistically reproduces the interannual variability associated with ENSO, but also shows high prediction skill over the central equatorial Pacific. Therefore, it has been widely used for ENSO modeling and predictability studies (Zhang et al., 2005a, b, 2008, 2013; Zheng et al., 2006; Gao et al., 2017; Tao et al., 2017, Tao et al., 2018). In particular, it has been routinely used for ENSO real-time predictions since 2003, and the model results are available at the website of the International Research Institute for Climate and Society, Columbia University (http://iri.columbia.edu/climate/ENSO/currentinfo/update.html).
However, ENSO events in IOCAS ICM are much more regular in terms of their spatiotemporal evolution compared to observations. This is mostly due to the lack of modulation from the stochastic wind forcing, which is excluded by the statistical atmosphere model of the ICM since it only represents the interannual wind response to SST anomalies (Zhang2003,Zhang2008). Actually, an increasing number of studies are highlighting the influence of high-frequency synoptic-scale atmospheric variations on the complexity and predictability of ENSO. For example, (Zhang et al., 2008) indicated that atmospheric stochastic forcing can cause significant irregularity of ENSO and modulate the ENSO period and amplitude randomly. Westerly wind bursts (WWBs), a form of tropical weather noise usually occurring in the western and central Pacific, have been argued to play an important role in modulating the strength, asymmetry, irregularity or timing of ENSO events (McPhaden, 1999; McPhaden and Yu, 1999; Fedorov, 2002; Kessler, 2002; Lengaigne et al., 2004). In particular, (Fedorov et al., 2015) demonstrated that WWBs, when acting on different ocean initial states, can result in different flavors of El Ni?o——namely, central Pacific (CP) and eastern Pacific (EP) El Ni?o——thereby contributing to ENSO diversity (Lopez and Kirtman, 2013; Lian et al., 2014; Hu et al., 2014; Chen et al., 2015; Hayashi and Watanab, 2017). (Lopez and Kirtman, 2014) indicated WWB activities can lead to a rapid spring decline in the signal-to-noise ratio of a coupled system that induces a spring predictability barrier of ENSO prediction (Webster and Yang, 1992; Mu et al., 2007a, b; Duan et al., 2009). For the coupled internal variability, (Wittenberg, 2009) demonstrated that there was a long-term modulation of ENSO within a 2000-year simulation of GFDL CM2.1. (Vega-Westhoff and Sriver, 2017) further showed that the internal variations within CESM dominate the SST changes in the eastern Pacific compared to the forcing of climate change. It can be seen that a good representation of atmospheric internal variability and the coupled internal variability in the model is important for simulating ENSO characteristics and improving ENSO predictions. In addition, although ICMs do not suffer from climate drift, they cannot directly simulate and predict climate anomalies in the extratropics associated with ENSO-related SST forcing, whereas this can be achieved by adopting a complete AGCM. Therefore, we set out to improve the atmospheric component of IOCAS ICM for ENSO simulation and prediction, despite the model already being competitive in terms of ENSO prediction compared to many CGCMs.
In this study, we construct a new coupled model for ENSO studies by coupling an AGCM (namely, ECHAM5) with the ocean component of IOCAS ICM (i.e., the IOM). Typically and previously, an HCM connects an OGCM to a simplified atmosphere model (e.g., Zhang, 2015). Here, we couple a simple ocean model with a complex AGCM. Despite this difference, the general hybrid concept (i.e., including one complex component and one simple component) is similar, and thus the coupled model presented here is still referred to as an HCM. Such an HCM is computationally efficient to run compared to CGCMs, and yet physically realistic enough to depict the atmosphere, allowing the investigation of ENSO modulations from the inherent variability of the atmosphere and global atmospheric response to ENSO, which are missing in any type of ICM. The main purpose of this paper is to document and validate the HCM simulations in terms of ENSO variability and global atmospheric responses to ENSO.
To facilitate comparison, the framework of IOCAS ICM is first introduced in section 2, followed by descriptions of the atmospheric component model and the coupling procedure of the HCM, as well as the datasets used to validate the model. Section 3 presents the simulated interannual variability of SST in the tropical Pacific associated with ENSO within the HCM, and a comparison between the HCM and IOCAS ICM is also discussed in this section. In section 4, the global atmospheric response to ENSO is examined. Finally, a summary and discussion are given in section 5.
2.1. IOCAS ICM
IOCAS ICM includes an IOM, an SST anomaly model with an empirical parameterization of Te, and a statistical atmosphere model in the tropical Pacific Ocean.The dynamical IOM was originally designed by (Keenlyside, 2001) and (Keenlyside and Kleeman, 2002), and consists of a linear and a nonlinear component. The linear component is extended from the (McCreary, 1981) baroclinic modal model with a horizontally varying background stratification. The first 10 baroclinic modes are resolved in the vertical layers, plus two surface layers governed by Ekman dynamics to simulate the combined effect of high-order baroclinic modes. The nonlinear component, a simplified model derived from the residual nonlinear momentum equations, is incorporated within the two surface layers to provide corrections to the linear component where the nonlinearity cannot be ignored.
An SST anomaly model is embedded into the ocean dynamical framework to describe the evolution of interannual temperature anomalies in the surface mixed layer. The governing equation includes ocean horizontal advection and entrainment by both specified mean currents and model simulated anomalous currents. (Zhang et al., 2003) demonstrated that the performance of SST simulations in the equatorial Pacific is significantly affected by the parameterization of Te. It has been shown that the interannual variability of the sea level and Te are closely correlated in the tropical Pacific (Zhang et al., 2004). Thus, an empirical Te submodel is constructed from the historical data of the sea level and Te anomalies by using the SVD method (Zhang et al., 2005a). The surface heat flux anomaly is negatively proportional to the local SST anomaly, with a thermal damping coefficient of (100 d)-1.
A statistical atmospheric model is also constructed based on the SVD analysis. The SVD is used to determine the relationship between the historical wind stress and SST anomaly, and thus the wind stress and SST anomaly fields are specifically related. As such, according to the constructed wind stress model, the interannual wind response can be calculated given an SST anomaly.
IOCAS ICM spans the tropical Pacific and Atlantic oceans (only the Pacific basin is considered in this work). Its domain covers (33.5°S-33.5°N, 124°-30°E), with a realistic representation of the continents. The model has a zonal grid with 2° spacing and a meridional grid stretching from 0.5° within 10° of the equator to 3° at the meridional northern and southern boundaries. Vertically, the ocean is assumed to be flat-bottomed with a depth of 5500 m. The linear component has 33 levels and 8 levels are in the upper 125 m. The two surface layers, in which nonlinear effects and high-order baroclinic modes are simulated, span the upper 125 m and are divided by a surface mixed layer whose depth is determined by a stability criterion from the annual mean temperature and salinity data in (Levitus, 1982). The dynamical ocean model and the SSTA model have the same grids. The model time step is 4800 s. The model's climatological fields include the SST of (Reynolds and Smith, 1995), model currents generated using the Florida State University wind stress (Stricherz et al., 1995), and thermocline depth constructed from (Levitus, 1982). The climatological fields are updated once monthly. More detailed descriptions of IOCAS ICM are given by (Zhang and Gao, 2016b). A 50-year control run of the ICM is used for comparison with the HCM constructed in this study.
2
2.2. AGCM
The atmosphere model used in this study is ECHAM5, which is a global spectral model based on the primitive equations. Prognostic variables consist of vorticity, divergence, temperature, surface pressure, cloud water and water vapor. The horizontal spectral resolution of ECHAM5 used in this work is T63 (1.875°× 1.875°), with 19 vertical hybrid levels up to a pressure level of 10 hPa. The model employs a semi-implicit leapfrog time-stepping scheme with a weak time filter to inhibit the spurious computational modes. The model time step is 1200 s for dynamics and physics, and the radiation is calculated at 2-h intervals. A more detailed model description of ECHAM5 is given by (Roeckner et al., 2003).Before being coupled to the ocean component, two experiments are carried out with ECHAM5 alone, in which the SST boundary data are from AMIP II. First, ECHAM5 is run for 50 years, forced by the climatological monthly mean of SST, from which the climatology of ECHAM5 can be obtained and is used to calculate the anomalous coupling flux in the HCM. Second, ECHAM5 is forced by monthly historical SST from 1956 to 2000 in the tropical Pacific Ocean and the SST beyond the tropical Pacific is set to be the climatological monthly mean. Our analysis focuses on the period between 1961 and 2000. This experiment, together with observations, is used to evaluate the global atmospheric simulations of the HCM.
2
2.3. Hybrid coupled ocean-atmosphere model
In this work, an HCM is constructed by coupling the ocean component of IOCAS ICM (i.e., the IOM) to ECHAM5. The atmosphere and ocean models are only coupled in the tropical Pacific Ocean. Beyond the active coupling regions, the underlying SST of the atmosphere is specified as the climatological monthly mean. To maintain the continuity of the boundary forcing, sponge layers are introduced at the northern and southern boundaries of the tropical Pacific, acting to relaxing SSTs to the climatological monthly mean. The coupling frequency between the atmosphere and ocean model is once per day. Given an SST, the atmosphere produces a total wind stress field and a net surface heat flux field, which is the sum of the solar radiation, longwave radiation, and sensible and latent heat fluxes. By subtracting the atmospheric climatology taken from the uncoupled simulation mentioned in section 2.2, the wind stress and heat flux anomalies are calculated and passed to force the ocean model. The ocean sends daily mean SST anomalies, superimposed on the climatological mean, back to the atmosphere. In order to maintain the radiative-convective equilibrium, the total SST is limited to be no greater than 30°C (Jin et al., 2003). Before being transferred to force the ocean model, the magnitude of wind stress anomalies is adjusted by multiplying a scalar parameter (ατ). This parameter is called the relative coupling coefficient and represents the strength of the interannual wind forcing on the ocean. Several tuning experiments are performed with different values of ατ to examine the coupled behavior in the HCM. It is found that taking ατ=0.8 can produce a reasonable interannual variability in the tropical Pacific. Because the model grids are different in the atmosphere and ocean, the exchanged variables are interpolated during the transfer process. The initial atmospheric field is a restart file from the previous 50-year integration. The initial condition of the oceanic component is the steady state after a long-term integration of the ocean model. The coupled model is integrated for 200 years and the last 100 years are used in the following analyses.2
2.4. Datasets
Observational and reanalysis datasets used for evaluating the model simulation include the following: SST from ERSST.v4 from 1951 to 2000 (Huang et al., 2014; Liu et al., 2014); SSH and zonal wind stress from GODAS from 1981 to 2017 (Behringer and Xue, 2004); SLP, 2-m temperature, and 500-hPa geopotential height from the NCEP-NCAR reanalysis from 1951 to 2000 (Kalnay et al., 1996); and precipitation from the GPCP, version 2.2, combined precipitation dataset from 1979 to 2012 (Huffman et al., 2009). These data are referred to as "observations" in the following. Interannual anomalies of all variables are defined as the deviations from their corresponding mean seasonal cycle.A time series of monthly SST anomalies averaged in the Ni?o3.4 region (5°S-5°N, 120°-170°W) from the HCM, IOCAS ICM and observations are shown in Fig. 1. It shows that the HCM is able to realistically simulate the interannual variability of SST anomalies in the tropical Pacific. As in the observations, the simulated time series of Ni?o3.4 in the HCM exhibits pronounced irregularity. For example, there are extreme El Ni?o events that peak in 1983 and 1998 in the observations. Similar warm events appear during the model years of 20-40 and 65-75, with anomalies of up to 3°C. The model also shows a suppressed period with anomalies of less than 1°C for model years 1-20 and 40-50, similar to the observations during the years 1975-80. However, ENSO events in the ICM are quite regular (Fig. 1c), probably due to the absence of modulation from the stochastic forcing in the atmosphere (Zhang et al., 2008). In addition, as shown in Fig. 1, the simulated ENSO events in the ICM are significantly stronger than those in the HCM and observations. The standard deviation of Ni?o3.4 index is often used to measure the amplitude of ENSO, and the standard deviation derived from the observations, ICM and HCM are 0.85°C, 1.33°C and 0.95°C, respectively. Compared with observations, the amplitude of ENSO is substantially overestimated in the ICM, whereas the ENSO amplitude in the HCM is comparable to observation.
Figure1. Time series of SST anomalies (units: °C) averaged over the Ni?o3.4 region from (a) observation, (b, d) the HCM and (c) IOCAS ICM. The results of the HCM and ICM are based on the model periods of 1-100 and 1-50 respectively, while the observations are during 1951-2000.It is well known that ENSO events are inclined to lock the peak times to boreal winter (Tziperman et al., 1998; An and Wang, 2001). In the ICM and HCM, the standard deviation of SST anomalies in the Ni?o3.4 region is minimum in late spring and maximum in winter (Fig. 2a). That is, the feature of phase-locking is effectively captured by both the ICM and HCM. However, the magnitude of the standard deviation is excessively large in the ICM, while it is more reasonable in the HCM.
Figure2. (a) Standard deviation of SST anomalies (units: °C) over the Ni?o3.4 region as a function of calendar month. (b) Power spectra of Ni?o3.4 index in the models and observation.The power spectra of the Ni?o3.4 SST anomaly time series are illustrated in Fig. 2b. The dominant observed period of the Ni?o3.4 index is four years, with a broad spectrum between two and five years. In the ICM, the SST variation is characterized by a sharp peak at a period of three years, but the dominant frequency presents a much too strong power and a too narrow band width compared to observations. The ICM run generates another peak at a period of 18 months, while it is non-significant in observations. In contrast, the HCM shows a broader peak between two and five years, although the largest spectral peak in the simulation is at three years. In addition, the power of the dominant frequency in the HCM is also closer to observation. In other words, the HCM simulation compares better with the observed spectrum than the ICM in terms of dominant period and spectral range. The significant difference in spectra between the HCM and ICM is mainly due to the modulation of atmospheric stochastic forcing, which is implicitly present in the atmospheric component of HCM but missing in the ICM.
In Fig. 3, the standard deviation of SST in the tropical Pacific is shown to further quantify the interannual SST variability. In the observations, the strongest interannual SST variability is located in the eastern equatorial Pacific and near the Peruvian coast. The HCM reproduces the observed maximum variance center near the Peruvian coast but overestimates the interannual variability of SST in the central Pacific. In contrast, the significant SST variability in the ICM tends to occur in the central equatorial Pacific, extending a little too far west than observed. Also, the interannual variance of SST in the ICM is much stronger than observed. Therefore, compared to the ICM, the HCM can capture the amplitude and structure of interannual variability in the tropical Pacific more effectively. By the way, both the HCM and ICM simulations cannot capture the so-called EP and CP El Ni?o events (Ashok et al., 2007; Kao and Yu, 2009; Kug et al., 2009; Timmermann et al., 2018).
Figure3. Standard deviation of interannual variability of tropical Pacific SST (units: °C) in (a) observation, (b) the HCM, and (c) IOCAS ICM.The evolution of SST, zonal wind stress and sea level anomalies on the equator from the models and observations is shown in Fig. 4. As in observations, both the ICM and HCM exhibit a coherent interannual oscillation between corresponding atmospheric and oceanic anomaly fields. The SSTAs are characterized by a standing pattern with the largest anomalies centered in the central-eastern tropical Pacific (Figs. 4a-c), while the maximum wind stress anomalies are located near the date line (Figs. 4d-f). During the development of ENSO events, the zonal wind stress anomalies feature an eastward propagation from the western equatorial Pacific into the central Pacific, consistent with that found in nature. The sea level anomalies, closely related to the variation of the thermocline, show an obvious eastward phase propagation along the equator, initially emerging in the western equatorial Pacific and then propagating eastward as they amplify (Figs. 4g-i). The simulated anomaly fields in the ICM are much smoother than those in the HCM, while the HCM variability is quite irregular, as illustrated in the Ni?o3.4 time series (Fig. 1), and thus reproduces a more realistic interannual variability as in observations.
Figure4. Evolution of SST (top panels; units: °C), zonal wind stress (middle panels; units: dyn cm-2), and sea level (bottom panels; units: cm) anomalies averaged between 2°S and 2°N in the (a, d, g) observation, (b, e, h) HCM and (c, f, i) IOCAS ICM. The results of the models are based on the model-year period of 11-25 and the observations during 1986-2000, respectively.To evaluate the model's performance, the correlations are examined between the SST anomalies averaged over the Ni?o3.4 region and anomalies of SLP, 200-hPa zonal wind, 2-m temperature, precipitation, and 500-hPa geopotential height, for boreal winter (December-January, February; DJF) and summer (June-July-August, JJA), when the global atmosphere exhibits the clearest response to the ENSO-related SST forcing.
Results for the SLP and 200-hPa zonal wind are shown in Figs. 5 and 6 for DJF and JJA, respectively, for the model and observations. For DJF, the overall correlation patterns from the model and observations show a large degree of similarity, although the HCM yields somewhat weaker correlations outside the tropical Pacific than those found in the observations. For SLP, the Southern Oscillation signature dominates, with anomalies of opposite sign over the western and eastern tropical Pacific (Figs. 5a and c). Consistent with the SLP, a weakening of the upper-tropospheric westerlies in the eastern tropical Pacific is simulated, with a meridional wave-like structure straddling the equator (Figs. 5b and d). The model correlations for JJA also represent good approximations to those observed in the Pacific Ocean. However, there are obvious differences outside the Pacific. For example, the HCM is unable to reproduce the positive correlations of SLP in the tropical Atlantic (Fig. 6c). The correlations of 200-hPa zonal wind anomalies in the tropical Atlantic and Indian Ocean also seem to be much weaker than those inferred from observations (Fig. 6d). This can be ascribed to the fact that SST fields in these ocean basins are prescribed to the climatology without interannual variations.
Figure5. Correlations of (a, c, e) SLP (units: hPa) and (b, d, f) 200-hPa zonal wind (U200; untis: m s-1) with Ni?o3.4 index for DJF from observation (top panels), the HCM (middle panels) and ECHAM5 forced by observed SST (bottom panels). The black dotted areas indicate that the correlations are statistically significant at the 5% level, as determined by a t-test.
Figure6. As in Fig. 5 but for JJA.Figures 7a and c compare the observed and modeled correlations between the 2-m temperature and Ni?o3.4 index for DJF. In observations, significant positive correlations are in the central and eastern tropical Pacific. One can also find similar positive correlations in the tropical Indian Ocean and along the Pacific coast in North and South America. In contrast, significant negative correlations are seen over the western parts of the Pacific spreading eastward from the western equatorial Pacific in both hemispheres and showing a characteristic horseshoe pattern. The model correlations represent good approximations to those observed in the Pacific Ocean basin but the meridional band width of high correlations is smaller than in nature. Outside the Pacific, correlations are too weak compared to those observed over certain regions, such as the Indian Ocean and South Pacific. Figure 7 also shows the correlation patterns of precipitation. Strong positive correlations are observed in the central equatorial Pacific, with a narrower meridional width than those of the surface temperature (Figs. 7b and d). Significant negative correlations are generally found in the subtropical Pacific, Maritime Continent region, and northern Brazil. The model's responses closely match those observed, especially in the central equatorial Pacific and subtropical Pacific, where the correlations between the precipitation and ENSO are positive and negative, respectively. The response over the Maritime Continent is somewhat underestimated by the model, while stronger negative correlations are found in the north band of the subtropical Pacific. Positive precipitation correlation across the southern United States and negative correlation over Brazil are reproduced well by the model, whereas the response over Africa and the Indian Ocean is not well matched to what is observed.
Figure7. Correlations of (a, c, e) 2-m temperature (TMP; units: °C) and (b, d, f) precipitation (PRECIP; units: mm d-1) with Ni?o3.4 index for DJF from observation (top panels), the HCM (middle panels) and ECHAM5 forced by observed SST (bottom panels). The black dotted areas indicate that the correlations are statistically significant at the 5% level, as determined by a t-test.Comparing the correlation patterns between JJA and DJF (Figs. 7 and 8), it clearly shows that, for the 2-m temperature, there is a significant warming signal throughout the year in the central-eastern tropical Pacific. Different from the temperature, the integrated global responses of precipitation in JJA are much weaker than those in DJF of the HCM, as well as in the observations. In JJA, the model simulations do well in reproducing the positive correlations of precipitation in the central tropical Pacific, but the response over the Maritime Continent is underestimated and overly strong negative correlations compared to those observed are found in the south and north bands of the subtropical Pacific (Fig. 8d).
Figure8. As in Fig. 7 but for JJA.To evaluate the performance of the HCM in reproducing the observed teleconnections between ENSO and mid-tropospheric atmospheric circulation, the correlation fields between Ni?o3.4 and 500-hPa geopotential height are presented in Fig. 9 for DJF and JJA. Positive (negative) correlations imply high (low) heights corresponding to El Ni?o (La Ni?a) events. In the tropical and subtropical latitudes, positive correlations are high at all longitudes, indicating a tropospheric warming during El Ni?o events. A wave train pattern spans the Pacific-American region, which initially emerges in the western equatorial Pacific and then propagates into the extratropics. It consists of three centers of pressure anomalies, located over the northeastern Pacific, Canada, and southeastern USA, respectively. The strongest circulation impact is felt in DJF, when the correlation amplitude is nearly double compared to that in JJA. The model captures the high positive correlations throughout the tropics but with a smaller meridional extent than in nature.
As mentioned in section 2.2, we conducted a stand-alone AGCM experiment, in which ECHAM5 was forced with observed interannual SST in the tropical Pacific Ocean. The results are shown in Figs. 5-9 in order to compare with the HCM properties. They show that ENSO-related global atmospheric anomalies in the surface climate and circulation simulated by the HCM are in a better agreement with those in ECHAM5 than observations, both with respect to the amplitude and spatial structure of correlations. For example, during JJA in Fig. 6c, the observed positive correlations of SLP in the tropical Atlantic are absent in the HCM, which is also the case for ECHAM5 (Fig. 6e). In the Indian and Atlantic oceans, both models show weaker correlation signals of 200-hPa zonal wind anomalies than observed. For the temperature and precipitation, the weak responses outside the tropical Pacific in the HCM are consistent with those in ECHAM5, while this is not the case for the observation (Figs. 7 and 8). The lack of agreement outside of the tropical Pacific between the HCM and observations can be due to the prescribed SST climatology without interannual variations beyond the tropical Pacific basin. This is evidenced by the good agreement between the HCM and ECHAM5, which is forced with observed interannual SST in the tropical Pacific.
Figure9. Correlations between 500-hPa geopotential height (H500) and Ni?o3.4 index for (a, c, e) DJF and (b, d, f) JJA from observation (top panels), the HCM (middle panels) and ECHAM5 forced by observed SST (bottom panels). The black dotted areas indicate that the correlations are statistically significant at the 5% level, as determined by a t-test.The HCM is able to reproduce realistic and irregular occurrence of ENSO events with substantial active and inactive periods. Both the observed and simulated Ni?o3.4 SST anomaly shows a broad spectrum of oscillation periods between two and five years, although the dominant period is four years in the observations and three years in the model. The HCM successfully locates the maximum center of ENSO variance near the Peruvian coast, but somewhat overestimates the SST variability in the central tropical Pacific. The model ENSO indicates reasonable annual phase-locking with minimum variance in late spring and maximum variance in boreal winter. Time-longitude distributions of observed and the HCM-simulated SST, zonal wind stress, and sea level anomalies illustrate that these fields exhibit a coherent interannual variation. In the simulation, a standing pattern of SST anomalies exists, with the largest anomalies in the central-eastern tropical Pacific, while the largest variation of zonal wind stress anomalies is near the date line. Eastward propagation is apparent for zonal wind stress and sea level anomalies, similar to that in nature. In contrast, the ENSO cycle in IOCAS ICM is far too regular, dominated by a three-year oscillation; and the maximum interannual variance of SST is too far west and with a much stronger amplitude.
It is also found that the HCM is capable of characterizing realistic atmospheric variations in response to ENSO-related SSTA forcing for DJF and JJA, respectively. Particularly, similar to the observations, a high correlation is seen between the tropical Pacific SST anomalies and atmospheric response in the winter. This is evidenced by examining the correlation fields between Ni?o3.4 index and anomalies of SLP, 200-hPa zonal wind, 2-m temperature, precipitation, and 500-hPa geopotential height. Within the tropics, the signature of Southern Oscillation is well reproduced, characterized by opposite anomalies of SLP over the western and eastern tropical Pacific, with a weakening of the upper-tropospheric westerlies. The model also reasonably captures the response of 2-m temperature and atmospheric circulation in the middle troposphere, including the tropospheric warming throughout the tropics and the remote impacts in the Pacific-American sector. The model-simulated correlation patterns of precipitation in DJF closely match the observations; however, the patterns in JJA deviate from the observations over some regions, such as the weaker signals in the Maritime Continent and stronger signals in the north and south bands of the subtropical Pacific. Compared with observations, the HCM is in better agreement with the stand-alone ECHAM5 forced with observed interannual SST in the tropical Pacific, both with respect to the amplitude and spatial structure of the atmospheric response. Therefore, the lack of agreement outside the tropical Pacific Ocean between the HCM and observations is mainly due to the fact that SST fields in these regions are prescribed to the climatology without interannual variations.
Compared to IOCAS ICM, the HCM shows clear advantages for ENSO-related studies. For example, the simulated ENSO characteristics show substantial improvement in the HCM compared to those in IOCAS ICM, including a more reasonable irregularity, amplitude, periodicity and life cycle of ENSO. By adopting an AGCM, the HCM can realistically describe the atmospheric variability, and thus can represent the nonlinear processes of the coupled system better than the ICM. Also, the HCM can characterize the global atmospheric responses to ENSO, which are not possible in the tropics-only ICM. In addition, by employing an anomaly coupling strategy, this HCM not only is free from the problem of climate drift, but also can effectively reduce the computational cost compared to fully coupled general circulation models. Nevertheless, like other HCMs, this HCM also has its disadvantages. For example, the atmosphere and ocean are only actively coupled in the tropical Pacific Ocean in the HCM, which may significantly underestimate the atmospheric response outside the tropical Pacific, particularly at regional or local scales. The tropical oceans have long been recognized as a key player in global climate, in which large-scale convection over the warm water provides an important portion of the driving energy for the general circulation of the atmosphere. Therefore, as a step forward, the HCM constructed in this work can be further expanded to include the tropical Indian and Atlantic oceans, with the expectation of improving the simulation of global climate. In fact, the relevant work is currently underway and will hopefully be reported in the future.
This paper presents our efforts in developing a new HCM and a preliminary analysis of the HCM-based simulation for ENSO variability and the associated global atmospheric response. There are still questions that remain to be answered, and further detailed analyses are clearly needed. For example, compared to IOCAS ICM, the HCM produces a more realistic irregularity of the ENSO cycle. But what model physics are responsible for the ENSO regularity? As mentioned in the introduction, an increasing number of studies are highlighting the potential influence of stochastic atmospheric noise (e.g., WWBs) on the complexity and predictability of ENSO events (Zhang et al., 2008; Lopez and Kirtman, 2013, Lopez and Kirtman, 2014; Hu et al., 2014; Lian et al., 2014; Chen et al., 2015). With the HCM constructed in this study, the impact of stochastic atmospheric forcing on ENSO needs to be further discussed. In addition, the diagnoses of the simulation suggest that the HCM may be suitable for the prediction of interannual climate anomalies associated with ENSO. But is the HCM more skillful in forecasting ENSO and the related climate variability compared with those in IOCAS ICM? These questions should be addressed by conducting hindcast experiments and more detailed diagnoses in the future.
