SST-Forced and Internal Variability of a Winter Wave Train over the Tropical Indo–Western Pacific and East Asia

Abstract

Previous studies have indicated that a high-level wave train from the tropical Indo–Western Pacific to East Asia (IWP-EA, expressed as geopotential height at 200 hPa) is triggered by dipolar convective activity anomalies over the IWP during the boreal winter. The current study highlights the relative importance of sea surface temperature forcing versus atmospheric internal variability on the IWP-EA pattern, based on an Atmospheric Model Intercomparison Project (AMIP) experiment with 30 integrations. It was found that the SST-forcing component can reproduce the observed IWP–EA pattern and the related rainfall dipole well, for both the spatial features and temporal evolutions. The internal variability of the rainfall dipole is strong in the southern and eastern Indian Ocean and region north of Australia, while the internal variability of height generally increases with latitude. The signal-to-noise ratios are just over 1 over the northernmost lobe of the IWP-EA (Japan and the region to its east), while ratios over the other centers reach values greater than 3. An inter-member EOF analysis of the rainfall dipole indicates that the variability associated with the first two modes can explain more than 70% of the total spread over most regions with large spread for both rainfall dipole and IWP-EA, including the region over Japan. Thus, some parts of internal variability of rainfall dipole and IWP-EA are connected.

1. Introduction

The El Niño–Southern Oscillation (ENSO), which is the strongest interannual climate phenomenon currently known in the global climate system, has a very apparent influence on global climate. The variation of tropical convection associated with ENSO is a key medium for ENSO and global climate through the excitation of stationary Rossby wave trains [1,2]. The so-called Pacific–North American pattern in the Northern Hemisphere [2] and the Pacific–South American pattern in the Southern Hemisphere [3,4] are two well-known wave trains triggered by convective anomalies in the equatorial central Pacific. Both wave trains can affect local climate by inducing extratropical circulation anomalies [4,5,6].

Another apparent wave train during the winter of ENSO originates from the tropical Indo–Western Pacific (IWP) to East Asia (hereafter referred to as the IWP-EA pattern), which is associated with dipolar convection anomalies over the IWP [7]. The wave train, with positive geopotential height anomalies at 200 hPa (hereafter Z200) over the IWP, negative Z200 anomalies centered over eastern China, and positive Z200 anomalies from Lake Baikal to Japan during El Niño, is the dominant mode of winter Z200 variability over East Asia. This pattern may share some similarities to the impact of ENSO on East Asian trough or East Asian upper-tropospheric westerly jet [8,9,10], and thus could lead to anomalies of surface temperature and precipitation over East Asia [11].

In addition to ENSO, the tropical Indian Ocean Dipole (IOD), which is characterized by an anomalous zonal gradient in sea surface temperature (SST) across the tropical Indian Ocean [12,13], also influences the IWP-EA pattern [7]. This influence results from the modulation of the IOD on the Walker circulation and the convective activities over the IWP. Although the IOD peaks in autumn and decays in winter, its impact on the wintertime IWP-EA pattern is not negligible. For the two extreme El Niño events of 1997/98 and 2015/16, the IWP-EA pattern was only apparent during the winter of 1997/98, because an IOD occurred in the autumn of 1997 but not in 2015 [14]. A positive IOD can increase (decrease) the rainfall anomalies over the tropical western (eastern) Indian Ocean, which are important in inducing the IWP-EA pattern.

The tropical SST anomaly (SSTA) is a main forcing of the atmospheric variability. It was found that the changes of monsoon circulation or monsoon precipitation during the second half of the 20th century could be reproduced in observational SST-forced atmospheric models [15,16]. However, atmospheric model control integrations with constant lower boundary conditions still display fluctuations on a wide range of scales, even longer than decades [17]. Based on Atmospheric Model Intercomparison Project (AMIP) experiments, for example, Chen and Zhou [18] found that internal variability accounted for one-third of the total variability of the summer heat waves over the Yangtze River Valley during 1979–2008, while the other two-thirds of the variability was due to SST forcing. Additionally, by analyzing AMIP experiments, it was found that the internal atmospheric dynamics can explain 80% of the extreme precipitation variation during winter over the West coast of the United States and that SST forcing accounts for only 20% of the variation [19].

The current study investigates the influence of atmospheric internal variability on the IWP–EA pattern, compared with the tropical SST-induced convective forcing in our previous study [7], using a 30-member AMIP experiment. Section 2 of this paper describes the experiment design. Section 3 provides results on the IWP-EA pattern in each model ensemble, showing the SST-forced mean response and the internal variability. It also demonstrates the connection between the internal components of tropical rainfall and the IWP-EA pattern. Section 4 discussed the possible source of internal variability and Section 5 concludes the paper with a summary of the main findings.

2. Model Experiments

AMIP-type simulation is a useful tool to understand interannual and long-term atmospheric variability [20,21,22,23,24], and has also been widely used to investigate the predictability of drought around the world [25]. These studies usually used outputs from multi-model with one realization ensemble, while a single model with multiple realization ensembles is used here to focus on internal variability instead of model bias. The current AMIP experiment was run with the ECHAM5 model from the NOAA Earth System Research Laboratory/Physical Sciences Division (ESRL/PSD). ECHAM5 was developed at the Max Planck Institute for Meteorology. Version 5.4, with a horizontal resolution of 0.75° × 0.75° and 31 layers in the vertical direction, was used in this study [26]. The simulation consisted of 30 ensemble members, forced with the same observed monthly SST/sea ice and radiative forcing but initiated from different atmospheric conditions. Any single ensemble included both internal and SST-forced contributions. With these 30 simulations, the internally generated variability in an individual run could be substantially reduced to reveal the model’s response to observed SST. Then, the difference between each model run and the forced response represents the contribution of atmospheric internal variability.

The time period was from 1979 to 2017. The respective seasonal cycles and linear trends were removed to obtain monthly anomalies. The analyses here focus only on boreal winter (December–February) when the IWP–EA pattern appears [7].

3. Results

As the IWP-EA pattern is triggered by convective anomalies over the IWP, we performed singular value decomposition (SVD) analyses between rainfall anomalies and Z200 to obtain the IWP-EA pattern. The observed results, based on rainfall from the Climate Prediction Center Merged Analysis of Precipitation [27] and Z200 from the NCEP-DOE Reanalysis 2 [28], are shown as a reference. The IWP-EA pattern and the associated rainfall anomaly in the observation and the ECHAM5 simulations are shown in Figure 1 and Figure 2. In the observations, the first SVD mode displays a dipolar pattern for rainfall anomalies in the IWP and alternate positive and negative Z200 anomalies from the IWP to East Asia (Figure 1a). The patterns resemble the anomalous rainfall dipole and the IWP-EA in the previous study, respectively. The model ensemble mean (EM) patterns exhibit a result similar to that of the observations (Figure 1b). The rainfall shows opposite sign in the tropical Indian Ocean and the western Pacific, and the Z200 anomalies are positive over the IWP, negative over eastern China and the ocean east of China, and positive over Japan and the region to its east.

One difference between the EM and the observation is that the wet west‒dry east pattern in the Indian Ocean in the observation is replaced by a dry west‒wet east pattern in the EM (Figure 1), which is a common feature in individual members (Figure 2). From the difference map between model mean and observation (see Figure 1c and Figure S1), the rainfall anomalies are stronger over the tropical eastern Indian Ocean and western Pacific. This is one reason why the centers of IWP–EA pattern might shift eastward in the model. This is a bias of ECHAM5 compared with the AMIP experiment from other models (not shown). A similar bias pattern was noticed in a scientific documentation of ECHAM5 through compositing several warm ENSO events (Figure 16 in Hagemann et al. [29]). Another difference is that the Z200 over northwestern Asia has an opposite sign in the EM and in the observation. Positive Z200 covers only the region east of Japan in the EM, whereas it extends to ~80° E in the observation.

The complete set of SVD patterns from each model simulation is shown in Figure 2. Each member basically reproduces the IWP-EA pattern and the rainfall dipole but with different strength and location. For example, in member 12, positive Z200 anomalies over the region east of Japan are less than 15 gpm, but they are much stronger in members 1 and 5. In member 2, negative Z200 anomalies cover a larger region and positive Z200 anomalies are shifted northeastward. Especially, the IWP–EA pattern appears as the second SVD mode in member 20, in contrast to it being the first mode in the other members. In member 20, the rainfall anomalies in the tropical Indian Ocean and the Z200 anomalies east of Japan are very weak. This mode accounts for only 30.6% of the covariance, which is the lowest value among the 30 members. This may indicate that the internal variability is strong in this member.

To measure the degree of resemblance between the observed and the simulated IWP–EA patterns, the centered pattern correlation and root mean square (RMS) difference were calculated for rainfall and Z200, respectively (Figure 3). For rainfall, the pattern correlation ranges from 0.42 to 0.74, and the RMS differences range from 0.7 to 1.3 mm/day, with a close linear relationship between the two (i.e., lower RMS differences correspond to higher pattern correlation and vice versa). In other words, the model performs well or poorly on pattern and amplitude concurrently. The situation of Z200 is similar to that of rainfall, but there is closer linear relationship between its RMS differences and pattern correlation. We also compared these quantities between rainfall and Z200. Both pattern correlation and RMS difference display a close linear relationship between rainfall and Z200. Thus, it has been demonstrated that the heating associated with the rainfall dipole induces the IWP-EA pattern.

The expansion coefficients (EC) of SVD1 (SVD2 for member 20) are shown in Figure 4. Although there is some noise in each member, extreme values of both rainfall and Z200 occur preferentially in El Niño years, such as 1982/83, 1997/98, and 2015/16. The mean EC of the 30 members is highly correlated with the observed EC (greater than 0.8) for both rainfall and Z200. The result indicates that the observed IWP–EA pattern and the rainfall dipole are forced mainly by SST variability.

We calculated the spread (defined here as the standard deviation across the 30 members) of the IWP-EA/rainfall dipole (Figure 5a). The spread represents the atmospheric internal variability. Large spread of rainfall occurs in the southern and eastern Indian Ocean and north of Australia. The spread of Z200 mainly increases as latitude increases. The spread is large over northeast China and Japan, where the northernmost lobe of the IWP-EA pattern is located. This implies a strong impact of internal variability on this lobe. The signal-to-noise ratio (SNR, ensemble mean divided by spread) between the SST-forced and the internal variability is shown in Figure 5b. The regions where the ratio is larger than 1 correspond to centers of the rainfall dipole and lobes of the IWP-EA pattern in the EM. This indicates that the ensemble mean patterns are robust. The ratio of Z200 east of Japan is relatively smaller than that of the other two lobes of the IWP-EA because of the strong internal variability there.

The variety of the IWP-EA and rainfall dipole in the individual model runs results from the superposition of internal atmospheric variability and the SST-forced response. To make this point clear, we divided the total anomalies (original results calculated from individual member) into contributions from the SST-forced response (mean of all ensemble runs) and the internal variability (difference of total anomalies in individual ensemble and the forced response), which is displayed in Figure 6. The internal components are different from member to member but with some common features. For example, rainfall anomalies are basically stronger in southern Indian Ocean and north of Australia, and Z200 anomalies are relatively larger in mid-latitude. These features are consistent with the spread patterns in Figure 5a. In some runs, the internal components exhibit similar patterns but with opposite signs, such as runs 21 and 25. The two runs have the minimal and maximal Z200 RMS difference respectively (Figure 3b), and also hold extremes for the other quantities shown in Figure 3. The internal Z200 shows a north–south dipole with zero lines near 30° N in both runs. The rainfall anomalies are negative at the intersection of the Indian Ocean and Pacific in run 21, whereas they are positive in run 25. For the internal components, the Z200 anomalies may also be associated with the rainfall anomalies.

To test the above hypothesis, we first used empirical orthogonal function (EOF) analysis to identify the most prominent patterns of rainfall internal variability. Here, we computed the leading EOF of the set of 30 rainfall dipole patterns over the IWP (as shown in Figure 2). Unlike conventional applications of EOF analysis in the time domain, we applied it in the “ensemble member” domain to find the dominant pattern of internal rainfall dipole patterns. Then, we regressed the Z200 onto the corresponding principal components (PC) to obtain the associated Z200 pattern. The leading two EOFs account for 23.5% and 19.2% of the variance in rainfall dipole across the 30 ensemble runs, respectively (Figure 7). The EOF1 of rainfall exhibits negative anomalies in the south-central Indian Ocean and positive anomalies north of Australia. The associated Z200 shows negative anomalies over a zonal belt between 10° and 40° N, west of 120° E, and in Japan and the region to its east. Weak positive Z200 anomalies occur over northwest Asia and the Tropics. The EOF2 of rainfall displays a tripole pattern over the IWP, while the corresponding Z200 shows northwest–southeast oriented anomalies with alternate signs. The regions with large values in the two EOFs and the associated Z200 fields typically correspond to the regions with large spread across the ensemble runs (Figure 5a), for example, the large rainfall spread in the southern and eastern Indian Ocean and north of Australia, and the large Z200 spread extending to the east from Japan. We also performed EOF on Z200 patterns and regressed rainfall on their PCs (Figure S2). The results of the first two modes are similar to those in Figure 7, especially for EOF1. This implies that some parts of internal variability of rainfall and Z200 are correlated. It is worth noting that the percentage variances explained by the first two modes are close and they are not separated from each other based on North’s rule (Figure 7c). However, the first two EOFs can be separated from the other modes. In the following, our analysis focuses on the two modes together rather than any single one.

To examine the extent to which the two EOFs can explain the total internal variability, we reconstructed the rainfall and Z200 based on the two EOFs:

$$x_i={{\displaystyle \sum }}_{j=1}^2{a}_{j}P{C}_j\left(i\right), i=1,2,\cdots ,30,$$

where PC is from the rainfall EOF, a is the linear regression coefficient (Figure 7), x is the reconstructed rainfall or Z200, and i represents the 30 ensemble runs. Figure 8 shows the standard deviations (STDs) of reconstructed rainfall and Z200 across the 30 runs and a comparison to the total spread, as shown in Figure 5a. The rainfall STDs are large in the southern and eastern Indian Ocean and north of Australia, which is consistent with the total spread map. In these regions, the two EOFs can explain more than 70% of the total internal variability (Figure 8b). For the Z200, the reconstructed STD is large over South Asia, Southeast Asia, and the Japan Sea, which account for more than 70% of the total spread. But it is apparent that the reconstruction has poor performance over northwestern Asia, where mid-latitude disturbances should dominate the internal variability.

By adding the reconstructed variability to the SST-forced pattern, the estimated total variability is obtained. Figure 9 displays such estimations for runs 21 and 25, with the minimum and maximum Z200 RMS difference, as discussed for Figure 6. The estimated total patterns not only clearly show the rainfall dipole and the IWP–EA wave train, but also reproduce some differences between the two runs, such as the eastward shift of two extratropical lobes of the IWP-EA in run 25. The pattern correlations between the estimated anomalies and total anomalies are high (with minimum 0.88), implying well-reconstructed patterns in the two runs.

4. Discussion

One possible source of the internal variability is intraseasonal variability. Intraseasonal data are generated from the monthly output following the methodology of Vimont et al. [31]. First, the ensemble mean is removed from each member. For individual members, each winter’s mean (DJF mean) is subtracted from the corresponding monthly data (December, January, and February). Then we obtain unforced intraseasonal data for the winter. Finally, standard deviation is calculated to represent intraseasonal variability (Figure 10). The variability of rainfall and Z200 exhibits a similar pattern across 30 realizations. For rainfall, strong variability is found from Madagascar to Java in the Indian Ocean, north of Australia, and in the western Pacific south of the equator. The Z200 variability mainly increases wide latitude, with relatively small (big) values over central China (the Japan Sea). The intraseasonal variability of both rainfall and Z200 in individual ensembles resemble the pattern spread across 30 members (Figure 5a). The pattern correlation between intraseasonal variability and pattern spread is higher than 0.8 (0.9) for rainfall (Z200) in all 30 runs. We speculate that the intraseasonal variability should play a vital role in constituting the internal variability.

5. Conclusions

This study explored the relative roles of SST forcing and atmospheric internal variability in determining the teleconnection from the IWP to East Asia (IWP-EA pattern), which is induced by a rainfall anomalous dipole in the IWP during winter. We made use of a 30-member AMIP run with ECHAM5. Each member was subjected to the identical observed monthly SST forcing from 1979 to 2017, but each began from a different atmospheric initial condition. Thus, the impact of SST forcing on the IWP-EA pattern/rainfall dipole was estimated from the ensemble mean of 30 runs, and the residual from single run after removing the ensemble mean indicated the impact of internal atmospheric variability.

The IWP-EA pattern/rainfall dipole in the SST forcing component closely resembled the observed counterpart, both spatially and temporally. About 68.9% of the total variance of the observed IWP-EA teleconnection can be explained by SST forcing (r = 0.83 in Figure 4b). This implies that the observed variability is mainly due to SST forcing. The internal atmospheric variability alters the strength and location of the IWP-EA pattern/rainfall dipole, but does not eliminate them. Additionally, there was a close relationship between the model simulations of rainfall dipole and the IWP-EA pattern. That is, ensemble runs that produced a better rainfall dipole also produced a better IWP-EA pattern. Thus, there is again a coincidence suggesting that the IWP-EA pattern is induced by the rainfall dipole.

The spread of Z200 across the ensemble members basically increased with higher latitude, while the spread of rainfall displayed local maximums in the southern and eastern Indian Ocean and the region north of Australia. The SNRs between the forced and internal variability were high in the region where the centers of the rainfall dipole/IWP-EA pattern are located. Among these centers, the SNR of Z200 was smallest over the region east of Japan (just larger than 1), indicating strong internal variability there. An inter-member EOF analysis on rainfall indicated that the first two modes could explain more than 70% of the total spread in the southern and eastern Indian Ocean and the region north of Australia, where a large rainfall spread is located. The Z200 variability related to the two modes contributed more than 70% of total spread over South Asia, Southeast Asia, and the Japan Sea. The latter was covered by the northernmost lobe of the IWP-EA pattern.