Biases and Improvements of the ENSO- East Asian Winter Monsoon Teleconnection in CMIP5 and CMIP6 Models

The inuence of El Niño–Southern Oscillation (ENSO) on the East Asian winter monsoon (EAWM) is investigated based on the outputs of phase 6 of the Coupled Model Intercomparison Project (CMIP6) models and compared to that in phase 5 (CMIP5). Results show that the CMIP6 models generally reproduce the ENSO-EAWM teleconnection more realistically than the CMIP5 models, although they still somewhat underestimate the ENSO-EAWM teleconnection than observed. Based on the inter-model spread of ENSO-EAWM teleconnection simulated in the CMIP5/CMIP6 models, we reveal that the commonly underestimated ENSO-EAWM teleconnection among the models can be traced back to the excessive cold tongue bias in the equatorial western Pacic. A model with a stronger climatological cold tongue favors generating a more westward extension of the ENSO-related SST anomaly pattern, which in turn forces an anomalous cyclonic circulation over the Northwest Pacic (NWP). It offsets the anticyclonic anomalies in the NWP triggered by the warm ENSO-related SST anomalies in the tropical Indian Ocean and the central-eastern Pacic and weakens the ENSO-EAWM teleconnection. Compared with the CMIP5 models, CMIP6 models better simulate SST mean state and the resultant ENSO-EAWM teleconnection. The present results suggest that substantial efforts should be made to reduce the bias in the mean-state SST for further improving the simulation and projection of the East Asian-western Pacic winter climate.


Introduction
The East Asian winter monsoon (EAWM), one of the most active monsoon systems during boreal winter, exerts a large in uence on the weather and climate in East Asia (Chen et Yang and Huang 2021). An anomalous anticyclone (cyclone) generates over the Northwest Paci c (NWP) during El Niño (La Niña) mature winter, and its associated meridional anomalies weaken (enhance) the EAWM ow and lead to warmer (cooler) and wetter (dryer) climate in East Asia and Southeast Asian countries (Zhang et al. 1996; Wang et al. 2000; Wang et al. 2013). Therefore, the anomalous anticyclone (cyclone) is a key bridge linking ENSO to the East Asia-western Paci c winter climate (Zhang et al. 1996).
Climate models are effective tools for understanding and projecting EAWM variability. The performance of models in reproducing ENSO-EAWM teleconnection behaviors is important for determining the reliability of the model's projections of future East Asia-western Paci c winter climate change. Previous studies suggested, using the climate models participating in phase 3 of the Coupled Model Intercomparison Project (CMIP3) and phase 5 (CMIP5), that the representation of the ENSO-EAWM relationship in CMIP3 and CMIP5 models depends on the amplitude and the longitudinal extension of ENSO-related sea surface temperature (SST) pattern (Gong et al. 2014;Gong et al. 2015). Moreover, the performance of CMIP5 models in representing ENSO's in uences on the East Asian-western Paci c winter climate is better than that of CMIP3 models (Gong et al. 2015). Recently, the outputs from the latest climate system models for CMIP6 have been released. CMIP6 models have been improved in comparison to CMIP5 models in terms of the dynamic core and the model physics (Jiang et  Based on the outputs from 20 CMIP5 and 20 CMIP6 models, the present study investigates the biases of CMIP6 models in reproducing the observed ENSO-EAWM teleconnection and compares the results to those in CMIP5 models. We show that the simulation of the ENSO-EAWM teleconnection is signi cantly improved in CMIP6 and reveal a solid linkage of the simulations between the ENSO-EAWM teleconnection and the climatological cold tongue strength in the equatorial Paci c. Our results highlight the importance of simulating a realistic cold tongue strength in reproducing ENSO-EAWM teleconnection.
The rest of the paper is organized as follows. Section 2 describes the reanalysis datasets, models and methods used in this study. Section 3 shows the biases in simulating the ENSO-EAWM teleconnection in models. Section 4 investigates the origins of simulated ENSO-EAWM teleconnection biases based on 40 CMIP5/6 models. Section 5 compares the performances of the two generations models. Section 6 provides a summary and discussion.

Models and observations
The monthly mean outputs of the historical run from 20 CMIP5 models (Taylor et al. 2012) and 20 CMIP6 models (Eyring et al. 2016) are used in this study. The names of these models are listed in Table 1. We only analyzed the rst realization ('r1i1p1' for CMIP5 and 'r1i1p1f1' for CMIP6) of each model. The monthly mean SST, atmospheric variables and precipitation from 1961 to 2000 are used. For comparing with observations, we also use the observational monthly mean SST data from the Extended Reconstructed SST, version 3 (ERSST. v3) (Smith et al. 2008), which has a horizontal resolution of 2°×2°( available at https://www.esrl.noaa.gov/psd/data/gridded/data.noaa.ersst.html). The monthly mean wind and precipitation are from the National Centers for Environment Prediction-National Center for Atmospheric Research (NCEP/NCAR) reanalysis dataset with a horizontal resolution of 2.5°×2.5°, covering the period from 1961 to 2010 (Kalnay et al. 1996) (available at https://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis.html). All the model outputs and observational datasets are horizontally interpolated onto the same 2.5°×2.5° grid before analysis.

Methods
The boreal winter Niño-3.4 index [de ned as the December-February (DJF) SST anomalies averaged over the region (5°S-5°N, 120°-170°W)] represents the ENSO variability. The unstandardized Niño-3.4 index is regressed onto the interannual anomalies of SST, circulation, and precipitation to display the ENSOrelated variability. In this study regressed anomalies are calculated in each model rst and then are averaged across the models with equal weights, which is denoted as the MME. The inter-model consensus is considered as high if more than 70% of models agree on the sign of the MME (

The Enso-eawm Teleconnection Biases In Cmip5 And Cmip6 Models
ENSO is a crucial factor to impact the interannual variability of the EAWM (Huang et al. 2012). The primary climate effect of ENSO is represented in terms of precipitation and atmospheric circulation anomalies (Bellenger et al. 2014). To evaluate the model performance in simulating the ENSO-EAWM relationship in CMIP5 and CMIP6 model, Figs. 1a-c show the precipitation anomalies and 850-hPa wind anomalies associated with ENSO in observations, the CMIP5 MME and the CMIP6 MME. Although the two generations of models can roughly reproduce the ENSO-related anomalous low-level anticyclone located over the NWP which bridges the teleconnection between ENSO and EAWM (Figs. 1a-c), there still exist some biases, such as a weak rain belt stretching from southern China to the south of Japan, weak anticyclonic circulation anomalies and resultant rainfall anomalies over the NWP and stronger rainfall anomalies over the equatorial western Paci c (EWP) in CMIP5 MME and CMIP6 MME relative to those in observations (Figs. 1d-e).
Since the ENSO affects the EAWM mainly through the southerly anomalies of the west side of anomalous anticyclone (e.g., Chen et al. 2000; Wang and Chen 2010b; Gong et al. 2015), the low-level ENSO-related meridional wind anomaly bias in the region of the EAWM index de ned in Ji et al. (1997) is employed to quantitatively illustrate the in uence of ENSO on the EAWM in models. Here, the area-averaged 1000-hPa ENSO-related meridional wind anomaly bias in the region (10°-30°N, 115°-130°E) in individual models are shown in Fig. 2. Although the simulated ENSO-related meridional wind anomaly bias varies from model to model, most models simulated an overly weak meridional wind anomaly (with a negative bias) ( Fig. 2a-b). In contrast, a small number of models can roughly reproduce the observed meridional wind anomaly associated with ENSO, such as ACCESS1-3 and GFDL-ESM2G in CMIP5 (Fig. 2a), CAMS-CSM1-0, CESM2, CNRM-CM6-1, FGOALS-f3-L, FGOALS-g3, and GFDL-ESM4 in CMIP6 (Fig. 2b).

Origins Of The Models Bias In Enso-eawm Teleconnection
Based on the inter-model diversity, we investigate the origins of ENSO-related EAWM variability biases in CMIP5 and CMIP6 models. The inter-model MV-EOF analysis of the combined three atmospheric variables is performed to extract the leading spread of the simulated ENSO-EAWM teleconnection in the East Asia-NWP region (0°-40°N, 100°-150°E). Figure 3 shows the rst inter-model MV-EOF mode of the ENSO-related low-level atmospheric circulation and precipitation anomalies and the corresponding normalized PC1. The rst mode explains 27% of the total variance of the inter-model spread in simulating ENSO-EAWM teleconnection. The rst inter-model MV-EOF exhibits positive rainfall anomalies over the NWP and negative rainfall anomalies over southern China and over the equatorial central Paci c. In association with the precipitation anomaly pattern, there generates a cyclonic circulation anomaly over the NWP and an anticyclonic circulation anomaly over southern China (Fig. 3a). The rst inter-model MV-EOF pattern is very similar to the biases simulated in the CMIP5 MME and the CMIP6 MME (Figs. 1d-e), which suggests that the biases in CMIP5 MME and CMIP6 MME dominate the inter-model spread in the  (Fig. 4b), and a pair of low-level cyclonic patterns straddle the equator over the NWP and Australia as a Gill-type Rossby wave response (Matsuno 1966;Gill 1980) (Fig. 4c). The lowlevel cyclonic circulation over the NWP reduces the original anticyclonic circulation anomalies over the NWP, leading to an underestimated ENSO-EAWM relationship in most models. The inter-model MV-EOF results suggest that the underestimated ENSO-EAWM teleconnection in CMIP5/6 models may be attributed to the excessive westward extension of ENSO-related SST variability. The regressed ENSO-related precipitation and low-level circulation anomaly patterns onto CT index (Fig. 5a) closely resemble the leading inter-model mode in simulating the ENSO-related precipitation and low-level circulation anomaly patterns shown in Fig. 3a. It means that a model with a stronger CT tends to simulate a weaker ENSO-EAWM relationship, namely a weaker rain belt over southern China and the south of Japan, stronger cyclonic circulation anomalies, and resultant more rainfall anomalies over the NWP (Fig. 5a). The regressed ENSO-related SST pattern onto CT index (Fig. 5b) is very similar to the leading inter-model mode in the simulated ENSO-related SST pattern (Fig. 4a). Jiang et al. (2021) revealed the CT strength is the leading source of the inter-model spread of simulated ENSO-related SST variability in the EWP. Moreover, the standardized PC1 is highly correlated with the CT strength, with a high inter-model correlation of 0.64, exceeding the 99% con dence level based on the Student's t test (Fig. 5c). These results indicate that the simulated CT strength is the leading source of the inter-model spread of ENSO-EAWM teleconnection. Figure 6 reorganizes the linkage from the simulated CT strength, the ENSO-related SST anomalies in the EWP and the ENSO-EAWM relationship in the individual CMIP5 and CMIP6 models. The CT strength is closely related to the simulated ENSO-related SST anomalies in the EWP with an inter-model correlation coe cient of 0.77 (Fig. 6a), which is because that a stronger CT leads to a larger zonal SST gradient in the EWP, then generates stronger zonal advection feedback in the EWP and nally a larger SST variability in the EWP (Jiang et al. 2021). The warmer EWP SST anomalies could force a stronger cyclonic circulation over the NWP through triggering a Gill-type atmospheric response, then make a weaker relationship between ENSO and EAWM ( Fig. 6b; correlation coe cient − 0.54). Finally, a model with a strong CT tends to simulate a weaker ENSO-EAWM relationship ( Fig. 6c; correlation coe cient − 0.37). The above processes exceed the 95% con dence level based on the Student's t test.

Improvements In Simulating The Enso-eawm Teleconnection In Cmip6 Models
The MME result shows that the biases of the ENSO-EAWM teleconnection in CMIP6 are relatively smaller than those in CMIP5 (Figs. 1d-f). Moreover, although both the CMIP5 MME and CMIP6 MME underestimate the ENSO-EAWM relationship, the magnitude of ENSO-related meridional wind bias in CMIP6 MME is almost one-third of that in CMIP5 MME (Fig. 2). Figure 6 also exhibits the improvement of CMIP6 MME in each physical node from the CT strength to the relationship between ENSO and EAWM relative to the CMIP5 MME. Thus, there is an overall improvement in the simulated ENSO-EAWM relationship in CMIP6 models.
Inspired by these results based on the inter-model diversity, we suppose that the improvement in simulating ENSO-EAWM teleconnection in CMIP6 models may be associated with the improved ENSOrelated SST variability in the EWP, and ultimately with the improved CT strength. Therefore, Fig. 7 provides the ENSO-related SST anomaly patterns in observations, the CMIP5 MME, the CMIP6 MME and their differences, respectively. Although the positive ENSO-related SST anomaly bias in the EWP still exists in the CMIP6 MME (Figs. 7c and 7e), the simulated ENSO-related SST pattern in CMIP6 MME are closer to observations than that in CMIP5 MME (Figs. 7a − c), and the magnitude of this bias in the CMIP6 MME has been signi cantly reduced relative to the CMIP5 MME (Figs. 7d − f).
Similarly, the improvement of the simulated CT pattern can be clearly seen in the CMIP6 MME relative to the CMIP5 MME from Fig. 8. A common cold SST bias generates in the equatorial western-central Paci c, i.e. the excessive CT bias, with a center at around 180° both in the CMIP5 and CMIP6 MME (Figs. 8d and   8e). However, the magnitude of the cold SST bias in the equatorial western-central Paci c is signi cantly improved in the CMIP6 MME (Fig. 8f), which con rms the role of the simulated overly strong CT bias in the underestimated ENSO-related EAWM variability biases in CGCMs. Namely, the improvement of CMIP6 MME in simulating ENSO-EAWM teleconnection can trace back to the improvement of simulated CT strength. All these results highlight the importance of simulated CT strength for the ENSO-EAWM teleconnection.

Summary And Discussions
In this study, the ENSO-EAWM teleconnection is evaluated based on 20 CMIP5 models and 20 CMIP6 models. Overall, these state-of-the-art models underestimate the ENSO-EAWM teleconnection with weak anticyclone anomalies over the NWP and rainfall anomalies over southern China and the south of Japan, potentially limiting the model's skill in predicting the winter climate in the East Asia-western Paci c.
However, comparing with the CMIP5 models, the simulation skill of the ENSO-EAWM teleconnection has been commonly improved in the CMIP6 MME, with more realistic anticyclone anomalies over the NWP and rainfall anomalies in southern China and the south of Japan. Based on the inter-model diversity of simulated ENSO-related EAWM variability, we identify a solid linkage between the ENSO-EAWM teleconnection and the CT strength in the EWP in CMIP5 and CMIP6 models. Speci cally, a model with an overly strong CT tends to simulate an excessive westward extension of ENSO-related SST anomalies in the EWP through modulating zonal advection feedback (Jiang et al. 2021). Meanwhile, a model with a stronger ENSO-related SST variability in the EWP tends to underestimate the ENSO-EAWM relationship. The warm SST anomalies in the EWP can directly force a Rossby wave response in the NWP, weakening the anomalous anticyclonic circulation and associated rainfall anomalies there triggered by the tropical central-eastern SST anomalies. Therefore, as expected, the improvement of ENSO-EAWM teleconnection in the CMIP6 MME can be traced back to the improvement of climatological CT strength.

Declarations
Con icts of interest/Competing interests There are no con icts of interest to declare.
Availability of data and material The data in this study are available for open access.

Code availability Not applicable
Authors' contributions All authors contributed to the concept and design of the research. WJ did the analysis and prepared the draft. HG, PH, LW, GH and LH acquired the funding and supervised the research. All authors contributed to the revising and editing of the paper.
Ethics approval Not applicable Consent to participate Not applicable Consent for publication All authors agree the submission and publication of the paper.  ENSO-related 1000-hPa meridional wind anomaly bias in the East Asia (10°-30°N, 115°-130°E) (the region is selected as the EAWM index de ned by Ji et al. 1997 and highlighted by a green box in Fig. 1a) in (a) CMIP5 models and in (b) CMIP6 models.      As in Fig. 1, but for SST mean state.