Realistic ocean initial condition for stimulating the successful prediction of extreme cold events in the 2020/2021 winter

In the first half of winter 2020/2021, several unprecedented cold events occurred in most parts of China and caused record-breaking low temperatures in many cities. However, seasonal predictions related to the onset and evolution of extreme cold events are still challenging. In this study, we first evaluated the short-term climate prediction skills in winter of the CAS-ESM-c global coupled model and revealed the key dynamic processes of the onset and development of 2020/2021 extreme cold events based on the ensemble forecasts starting on October 1st, 2020. Under the background provided by the synergistic effect of the warm Arctic and the cold tropical Pacific (La Niña), the model captured the abnormal meridional atmospheric pattern well, including the intensification of the Ural Blocking High and the negative phase of the Arctic Oscillation, forecasted the negative geopotential height anomalies over the Eastern Asian region, and finally, successfully predicted the outbreak of cold waves in December 2020 two months in advance. Moreover, the dominant differences in the initial ocean fields between the best and the worst forecast members were further compared to isolate the triggering factor for predicting cold events by CAS-ESM-c. The initial warm sea surface temperature over the Barents Sea can gradually form a meridional pattern between the warm Arctic and the cold Asian continent, which could lead to a reduction in the large-scale meridional temperature gradient at mid-high latitudes and weaken the atmospheric baroclinicity with more conducive to the southward outbreak of cold waves, highlighting that realistic Arctic ocean conditions in autumn can be reasonably assimilated into coupled models to stimulate the successful prediction of extreme cold events in the 2020/2021 winter.


Introduction
As catastrophic events, extreme cold waves are illustrated in previous studies to have serious effects on agriculture, transportation, electricity, industry and human health (e.g., Cohen et al. 2014;Ding et al. 2020), and global warming is increasing the frequency of serious extreme cold events in Eastern Asia (e.g., Ding et al. 2008, Luo et al. 2020. From the end of 2020 to the beginning of 2021, China was hit by multiple cold waves, causing widespread cooling and windy weather, with 58 meteorological observation stations recording new record low temperatures (Zheng et al. 2022a). The National Meteorological Center issued an orange alert on 28 December, the first such alert in China for recent 5 years (Zhang et al. 2021a).
Under global warming, Arctic regions have become one of the most rapidly warming regions in the world, which is called Arctic amplification (AA; Cohen et al. 2014;Huang et al. 2017), leading to a massive reduction in Arctic sea ice. Many studies have shown that the increased frequency of extreme cold events seems to be associated with the loss of Arctic sea ice (Wu et al. 2011a, b;Liu et al. 2012). However, there are also other studies that argue that the influence of Arctic sea ice on cold winters in the mid-latitudes is small (e.g., McCusker et al. 2016;Blackport et al. 2019). In addition to the sea ice loss in Arctic Ocean, cold sea surface temperature (SST) over the tropical Pacific also plays an important role in modifying the frequency of Arctic air intrusions (Zheng et al. 2022a, b). On a global scale, the synergistic effects of warm Arctic and cold tropical Pacific, with the temperature gradient between the equator and Arctic largely being reduced, further provides favorable background conditions for an exceptionally cold winter. This process intensifies the meridional sloping between the Ural High and the East Asian Trough, leading to a stronger ridge over the Ural Mountains, an enhanced East Asian Trough over Japan, and a more northward subtropical westerly jet (Liu and Ding 1992;Huang and Chen 2002;Yang et al. 2002;Zhang et al. 2008), further influencing cold conditions in Eastern Asia.
Although the economic and social impacts of the significant winter and summer extremes in Asia are equally important, prediction of winter extremes events, as demonstrated to be challengeable in current climate models and observations, has received less attention than that of summer extremes (Lee et al. 2010;Lee and Wang 2012;Lee et al. 2013). Stockdale et al. (2015) suggested that the stratosphere is an important carrier of model predictability during the early winter, though seasonal forecast models typically show low predictability of Arctic Oscillation (AO). Dunstone et al. (2016) used a large ensemble member to achieve high firstwinter North Atlantic Oscillation (NAO) skill and emphasized that the high skill can be achieved only when the noise is reduced by taking the mean of a large ensemble. Previous studies (e.g., Feng et al. 2021;Jung et al. 2014) mostly focused on the sub-seasonal forecast and Luo et al. (2018) found that the multi-model ensemble prediction has notoriously low skill over China, which confirmed that China is a region of great challenge for winter temperature prediction beyond one month. Wang et al. (2020) illustrated that China has established a climate prediction system with a complete structure and functions, which mainly includes climate prediction for different time scales. Wu et al. (2017a) have confirmed that the BCC (Beijing Climate Center) second-generation model has an ability to forecast winter temperature and summer precipitation in China. Li et al. (2021) successfully predicted above-average rainfall over the Yangtze River basin on the seasonal prediction. However, the development of climate prediction research is still a work in progress, and the intrinsic limits of short-term predictability is still unclear (Domeisen et al. 2018).
As indicated by Zheng et al. (2022b), most climate models failed to predict the onset of the massive 2020/2021 cold wave event. For example, regarding the monthly mean characteristics of the below-normal temperatures in most parts of China in December 2020, many advanced seasonal dynamic models showed poor forecasting skill, and the fundamental reason was that the dynamic models showed low capacity for predicting the middle-high-latitude circulations and were unable to foresee the meridional circulation that resulted in an intensified Ural High and a deepened East Asian Trough in December 2020. Notably, although most climate models predicted above-normal temperatures in most parts of China before one month, the CAS-ESM-c successfully predicted the emergence of cold waves that occurred in December 2020 starting on October 1st, 2020, 2 months in advance.
In this study, we focused on the key dynamic drivers of the 2020/2021 extreme cold event based on the ensemble forecasts from CAS-ESM-c and the triggering initial conditions that supported successful prediction. The paper is organized as follows. In Sect. 2, the forecast system, data sources, and retrospective experiment designs are described. The capabilities of CAS-ESM-c model for predicting future winter conditions are evaluated in Sect. 3.1, and the ensemble mean forecasts of the 2020/2021 cold event by CAS-ESM-c are displayed in Sect. 3.2. In Sect. 4, we discuss the fundamental processes of these cold events captured by the model, evaluate the dominant differences in the initial SST fields between the best and the worst forecast members to isolate the triggering factor for predicting cold events using CAS-ESM-c. A conclusion and discussion are presented in Sect. 5.

Forecast system
The fully coupled climate system model (CAS-ESM-c) was developed at the Institute of Atmospheric Physics (IAP), Chinese Academy of Science (CAS) (Sun et al. 2012 (Dickinson et al. 2006). A coupler (cpl6) developed at the U.S. National Center for Atmospheric Research was used to couple each of the model components together. In this paper, as described by Lin et al. (2019a) and Du et al. (2020), we assimilated observed SST, altimetry and EN4 profile data (Good et al. 2013) into the oceanic component of CAS-ESM-c based on the coupled framework to generate the coupled initial conditions. The assimilation interval was seven days, and the predictions were initialized using seven-member conditions. Further details on this ocean data assimilation system can be obtained from Du et al. (2020Du et al. ( , 2021 and Dong et al. (2021). The basic performance of this model has also been systematically evaluated (Dong et al. 2014;Dong and Xue 2016;Su et al. 2015).

Datasets
A high-resolution gridded temperature dataset for China was produced by the National Climate Centre of the China Meteorological Administration (CN05; Xu et al. 2009;Wu and Gao 2013;Wu et al. 2017b). The monthly and daily reanalysis data used in this study are from the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) (Kalnay et al. 1996) and include the low-level zonal and meridional wind components (U850 and V850), 500-hPa geopotential height (Z500), and 850-hPa geopotential height (Z850). The observed AO index and the global surface temperature dataset (Vose et al. 2012) are from the National Oceanic and Atmospheric Administration (NOAA). The hindcasts were evaluated by comparison with these gridded observational and reanalysis data.

Hindcast and forecast experiments
As described by Lin et al. (2019b), we designed the ensemble prediction scheme as following: (1) a reasonable initial ocean condition is generated through assimilating the oceanic observations into CAS-ESM-c by the ensemble optimal interpolation (EnOI) assimilation approach based on the air-sea coupled framework, in which the initial atmospheric condition is forced by the updated SST.
(2) To generate the ensemble initial conditions, only the initial oceanic uncertainties are considered by adopting the LAF (Lagged Average Forecast) method, and keep the initial atmospheric uncertainties as zero. The seven initial ensemble members are then designed as only perturbing the oceanic initial conditions, and they have the same atmospheric initial fields to avoid the initial incongruity between multiple atmospheric samples.
The hindcast experiments were performed for the 9-year period from 2011 to 2019 using CAS-ESM-c. For each year, the initial conditions were provided by an oceanic data assimilation system from the end of September to the beginning of October based on CAS-ESM-c. However, a more accurate initial field of the atmosphere might help to make better predictions, which would also require a more reasonable and complex coupled assimilation system (Zhang et al. 2020). The predictions were initialized using the sevenmember conditions from October 1st, and 6-month long ensemble hindcasts were carried out to evaluate the performance of forecast system. In 2020, the system was used for the operational prediction experiment, and the experimental designs for the real predictions were kept the same as those for the hindcasts, which were initialized using the seven-member conditions starting on October 1, 2020, and a 6-month-long real prediction was performed.

Capabilities of hindcast prediction in winter and ensemble forecasts for 2020/2021 extreme cold events
In this section, we evaluated the prediction capabilities of CAS-ESM-c for winter conditions by not only comparing the hindcasted and observed cold wave indices, but also choosing physical variables such as the 500-hPa geopotential height anomalies, 850-hPa wind anomalies and surface air temperature in China to examine the prediction capabilities in spatial. After that, the ensemble forecasts for the 2020/2021 extreme cold waves are displayed to successfully predict the cooling pattern two months in advance.

The prediction skills for winter conditions evaluated by hindcast
The hindcast skills of CAS-ESM-c were evaluated by using the pentad mean data in December from 2011 to 2019. We chose the pentad data for the significance evaluation to increase the number of samples, and the similar examination was performed by adopting the monthly data as shown in Figure S1 (the supplementary information). The temporal correlation coefficient (TCC) (Kharin and Zwiers 2001) was used to measure the prediction accuracy. Figure 1 shows the TCCs calculated from the pentad mean data between hindcasts and observations in December of each year from 2011 to 2019. The ensemble mean has good prediction skills for the 500-hPa geopotential height two months in advance, especially in the mid-high latitudes, which indicates that the model can capture the key atmospheric processes for stimulating cold waves, as well as the phase change of AO and the intensification of Ural Blocking High (UBH). The forecast skills of zonal and meridional winds can achieve good correlation only in the Siberian region, where the critical area of the cold waves occurred, suggesting that the model has the ability to capture the fundamental pathways of cold waves. The surface air temperature (SAT) was also predicted in the northwestern part of China well (Fig. 1d). Figure S1 shows that there were relatively few areas that statistical significance is above the 95% confidence level due to the small sample size, especially for the 500-hPa geopotential height anomalies. In our experiments, we only assimilated oceanic observations at the initial time, and the atmospheric condition was consequently updated by the coupled model through air-sea interaction, because the atmospheric chaos in the first two months (October and November) were significant and can lead the results unstable. After that the forecast errors were saturated, and the prediction skill was much higher in December than before as shown in the Figure S2 (see the supplementary information file).

Figure 2 displays the variations in key indices of cold waves for observations and ensemble mean hindcasts in
December from 2011 to 2019. All three indices, including the AO index constructed by projecting the 1000-hPa geopotential height anomalies poleward of 20° N onto the loading pattern of AO, the East Asian winter monsoon (EAWM) index calculated from the area mean (35° N ~ 40° N, 110° E ~ 130° E) of standardized 500-hPa geopotential height, and the Siberian High (SH) index calculated from the area mean (40° N ~ 60° N, 80° E ~ 120° E) of sea level pressure, were well predicted by the system in tendency and intensity. In this paper, we chose the above three indices as the key indices of cold waves because the negative phase of AO index represents the cold air from the Arctic outbreak southwards, and it influences the climate over Eurasia especially during the boreal winter (He et al. 2017). Theoretically, the southward invasion of cold air would result in the negative geopotential height anomalies which means the negative EAWM index (Wang and Chen 2010), and the higher SH index represents a stronger cold air (Wu et al. 2002). For the seasonal forecast, it is difficult to accurately predict the magnitude of the indices at present, so we also adopted the ratio of the same symbol (RSS) as a way to evaluate the prediction skill. The correlations for the AO index, EAWM index and SH index are 0.299, 0.266, and 0.443, and the RSSs are 62%, 58%, and 65%, respectively. Above evaluations on the spatial fields and the key indices of cold waves indicate that the model has good capability for predicting the cold waves and can meet the needs of actual operations.

The prediction of cold waves in 2020/2021 winter by CAS-ESM-c
In the first half part of 2020/2021 winter, China suffered record-breaking cold events (Fig. 3a). From the observed monthly temperature anomalies (i.e., the climatology for all variables is calculated from 1981-2010 in this work) in December, the average temperature during this time was approximately 1-2 °C below normal in most parts of China, except for the Tibetan Plateau. In parts of northern and southern China, the average temperature anomalies were as low as − 4 °C, as described by Zheng et al. (2022a). Figure 3b shows the ensemble mean forecasts of monthly mean temperature anomalies in December 2020. For these cold events, the model successfully predicted the cooling pattern in the central and eastern regions of China two months in advance. Although the negative temperature anomaly center was shifted from the observations, the total drop in temperature over the central and eastern regions was correctly predicted, which offers an important indicator for disaster prevention and mitigation. Figure 4 displays the observed and predicted variations of the three key indices of cold events (i.e., the AO index, the EAWM index and the SH index) in December. The negative AO index and EAWM index are associated with the southward invasion of cold air from the Arctic Ocean, and a higher SH index indicates stronger cold air accumulated in the Siberian region. For the AO index, the model acceptably predicted the overall trend in December, but the predicted AO intensity, was lower than the observed one. For the EAWM index, the model showed good prediction proficiency in terms of trend and intensity. However, for the SH index, the prediction and observation have opposite results, which might because the stimulated atmospheric state of the model was slightly different from the actual state, according to the model systematic bias. It is notable that the overall prediction abilities of the model for the Siberian High index and the SAT anomalies over northwestern China were good, consistent with the findings by Jung et al. (2014). However, in the 2020/2021 cold events, it seemed that the outbreak of the cold waves were not through the northwest path, but rather through the north path that hit the most parts of China in the model, causing the cooling mainly in the east of China.

The key processes and dominant initial conditions of 2020/2021 cold waves successfully captured by CAS-ESM-c
In this section, we explored the key dynamic processes of the 2020/2021 extreme cold events through the forecasts of CAS-ESM-c, and then we highlighted the dominant differences in the initial fields between the best and the worst forecast members, which identified the triggering factors for predicting the cold events in winter by CAS-ESM-c.

The key processes of 2020/2021 cold waves
For these cold events, the ensemble-mean forecast successfully predicted the large-scale cooling in central and eastern China (Fig. 3b) two months in advance. To explore the key dynamic processes captured by the prediction, we separately discussed the seven-member predictions by the CAS-ESM-c and selected the best and worst members for comparative analysis. Since these cold events mainly affected the central and eastern China (the region east of 100° E) as evident by observations and it's also the region with the highest prediction skill of the system. It is reasonable to focus the comparison analysis on this region of China, and we selected the best and worst members based on the predictions of the SAT east of 100° E among the seven members. As shown in Fig. 5, considering the spatial distribution of the temperature drop during the cold events, the correlation coefficient and the root mean square error (RMSE) between different forecast members and the observation, we selected member 4 ( Fig. 5f, Corr = 0.362, RMSE = 1.69 °C) as the best member and member 7 (Fig. 5i, Corr = − 0.001, RMSE = 3.14 °C) as the worst member.
Based on the selected best and worst members and the ensemble mean, we examined the differences between these three predictions to find the key dynamic processes for the cold wave prediction. Figure 6 shows the anomalies of the 500-hPa geopotential height, 850-hPa geopotential height, and 850-hPa winds in December from the observations, as well as the predictions of the best member, the ensemble mean, and the worst member. According to the observations, there was an obvious negative phase of AO and the intensification of UBH at both 500 hPa and 850 hPa geopotential heights in December. For the 850-hPa wind field, there was an abnormal anticyclone at 40° N, which continuously transported cold air to central and eastern China. Studies have shown that the continuous 2020/2021 cold wave events were closely related to the continued northerly position of the UBH and the phase change of AO (Bueh et al. 2022;Yao et al. 2022;Dai et al. 2022), which changed from a positive phase to a negative phase. For the ensemble mean, although there was a low-pressure system at the 500-hPa geopotential height, its strength was weak and its position was more west and north than the observations. In addition, the intensification of UBH was not adequately predicted, which may be the fundamental reason for the weaker intensities of the predicted cold waves compared with the observations. For the 500-hPa and 850-hPa geopotential heights, only the best member can successfully predict the intensification of UBH and the negative phase of AO, and the northerly winds near 40° N, leading to the invasion of cold waves. The predictions of the worst member at 500 hPa level were almost contrary to observations, and southerly winds were dominant at the mid-high latitudes.

The dominant initial conditions triggering successful prediction
In order to explore what caused the differences between the best and the worst forecast members during the twomonths integration, the dominant differences in the initial fields between the best and the worst forecast members were further compared to isolate the triggering factor for predicting cold events by CAS-ESM-c. Studies have shown that several consecutive cold waves broke out at the end of 2020 due to the synergistic effect of La Niña and warm Arctic (Zhang et al. 2021b;Zheng et al. 2022a). The SST in the Arctic Ocean was unusually warm, while the La Niña status extended westward in the tropical Pacific, causing the configuration of cold SST anomalies in the tropics and warm SST anomalies in the Arctic Ocean, which was also represented in the initial conditions by the ensemble mean (Fig. 7a). Figure 7b shows the dominate differences between the best and the worst members at the initial time, which were in the Arctic Ocean especially over the Barents Sea. The primary reason for this is most likely that the maximum uncertainties in both of the observations and model simulations are concentrated over the Arctic Ocean, which further leads to the dominant differences in the analysis field occurring in the Arctic at the initial time (Bunzel et al. 2016). Under the realistic status of anomalous SST configuration reproduced by the ensemble mean, the best member still has obvious warm SST anomalies compared with the worst member in the Arctic. The warm anomalies occurred in the Arctic at the initial time continued to expand during the subsequent integrations (Fig. 7d, f).
In November, the best member formed a more obvious "warm Arctic-Cold Siberia" pattern than the worst member. The stronger westerlies in November (Fig. 7f) indicative of a positive AO, supported by the negative 500-hPa geopotential height anomalies at higher latitudes and positive anomalies at lower latitudes (Fig. 9d). As mentioned above, the sharp differences between the warming Arctic and the cooling Fig. 5 The monthly mean SAT anomalies (relative to 1981-2010) in December 2020 in China (only east of 100° E) of the observation, ensemble mean and forecasts by seven members. The root mean squared error (RMSE, unit is o C) and the pattern correlation (r) in each member were calculated from itself and observations. The blue box represents the selected best member, and the red box represents the selected worst member Siberia could inevitably lead to a reduction in the largescale meridional temperature gradient at mid-high latitudes, which weakens the atmospheric baroclinicity, and then influences the upper-level jet stream and Rossby wave activities. The intensified meridional movement of Rossby wave would lead to more cold air from the Arctic southwards (Outten et al. 2012, Luo et al. 2016Tao et al. 2019), resulting in the cold events in Eastern China being greatly strengthened and expanded (Zhang et al. 2021a). Under the influence of the different initial conditions in the Arctic Ocean, the best member and the worst member gradually stimulated the different atmospheric status. Figure 8 shows the differences between the best member and the worst member of the 500-hPa geopotential height anomalies between October 1-6 pentads stimulated by the SST differences in the initial conditions over the Arctic Ocean. According to the experiments design, the coupled initial conditions were provided by an oceanic assimilation experiment by the end of September based on CAS-ESM-c and the initial differences are all on the ocean fields. There was almost no difference in the first and second pentads. After that, the atmosphere fields gradually changed to a different stimulated state in which the UBH was intensified in the best member compared to the worst member, and AO had changed from a positive phase to a negative phase in the best member, providing a favorable state for the outbreak of cold waves. Figure 9 displays the observed 500-hPa geopotential height anomalies in October and November, and The monthly mean reanalysis data, the best member, the ensemble mean, and the worst member for 500-hPa geopotential height anomalies (relative to 1981-2010), 850-hPa geopotential height anomalies, and 850-hPa zonal and meridional wind anomalies in December 2020. (a~d) Represent the monthly mean observation, the best member, the ensemble mean, and the worst member of 500-hPa geopotential height anomalies. (e~h) Represent the same as (a~d) but for 850-hPa geopotential height anomalies, and the gray color denote the missing value in the 850 hPa because the Tibet topography. (i~l) Represent the same as (a~d) but for 850-hPa zonal and meridional wind anomalies (m/s) the differences between the best and worst members. The relatively warm SST in the Arctic Ocean at the initial time (autumn) (Fig. 7b) stimulated an intensification of UBH and the negative phase of AO (Fig. 9b), eventually leading to a large-scale southward invasion of the cold waves in December. In Figure S3 (see the supplementary information file), the anomalous SST (relative to 1981-2010) differences in the initial Arctic Ocean among all seven members are compared. All the seven members show a consistent characteristic of warm SST anomalies in the key Arctic region, the selected best member has the highest initial SST and the selected worst member has the lowest initial SST, showing they have the largest difference in their initial Arctic SST anomalies. This is also the emphasis of this work that the initial warm Arctic ocean (as shown in Fig. 7a, 7b) play an important role in triggering the cold wave invasion in Eastern Asia (e.g., Johnson et al. 2018;Ma et al. 2018).

Conclusion and discussion
Most parts of China experienced several unprecedented cold events in the first half of 2020/2021 winter. The CAS-ESM-c global coupled model predicted the below-normal temperature in most parts of China two months in advance. In this study, we evaluated the seasonal prediction capabilities in winter by the CAS-ESM-c model through using the pentad mean data in December from 2011 to 2019, and the results indicates that this system can forecast the onset and development of cold events well. To better understand the predictable signals of the 2020/2021 cold events, we identified key dynamic processes for these exceptional cold events, which may assist future winter forecasting and disaster mitigation. The dominant differences in the initial fields between the best and the worst forecast members were further compared Fig. 7 The initial conditions of the SST, the monthly mean SAT and the 850-hPa zonal and meridional wind anomalies (relative to 1981-2010). (a) Represents the ensemble mean SST anomalies at the initial time (October 1st, 2020). (b) Represents the differences in SST anomalies between the best member and the worst member at the initial time. (c, e) Represent the reanalysis data of the monthly mean surface temperature and 850-hPa zonal and meridional wind anomalies in October and November 2020. (d, f) Represent the same as (c, e) but for the differences between the best member and the worst member to isolate the triggering factor for allowing the prediction of cold events by CAS-ESM-c.
For the ensemble mean forecast of CAS-ESM-c starting on October 1 st , negative SAT anomalies appear over most of China but with a weaker intensity than that observed. The successful forecast of the below-normal temperature in most parts of China lies in the predictable signals from the abnormal atmospheric geopotential height at middlehigh latitudes. The model also shows a good forecast of the anomalous movements of the UBH and the phase shift of AO, exhibiting a negative geopotential height anomaly near Eastern Asia. The dominant difference in the initial conditions between the best member and the worst members occurred around the Arctic Ocean, where SST of the best member was warmer than that of the worst member under the important background provided by the synergistic effect of warm Arctic Ocean and cold tropics. A meridional pattern between the warm Arctic and the cold Asian continent was then formed in the following two months, leading to a reduction in the large-scale meridional temperature gradient at mid-high latitudes and the weakened atmospheric baroclinicity, further inducing more conducive to the southward outbreak of cold waves (Overland et al. 2011).
In this paper, only the NCEP reanalysis data are used for verification, and other reanalysis data, such as ERA5 or JRA55, can be considered in our future evaluations, which will make the verification conclusion more robust. Through this work, we highlighted that realistic conditions of Arctic Ocean in autumn, especially under the case of the La Niña conditions, can be reasonably assimilated into coupled models to stimulate the successful prediction of extreme cold events in 2020/2021 winter, similar to previous study presented by Jung et al. (2020), in which better Arctic condition prediction would improve winter climate forecasts over midlatitudes through the Arctic-midlatitude teleconnection. However, this work was just a case study, and whether the warm Arctic-cold Siberia or the Arctictropical ocean temperature difference will influence the midlatitude cold event is still an ongoing discussion because of the limited case samples. Also, with a significant global Fig. 8 The evolution of differences between the best member and the worst member at the 500-hPa geopotential height anomalies (relative to 1981-2010) in October 2020, stimulated by the SST differences in the initial conditions over the Arctic Ocean. (a~f) Represent the 1 to 6 pentad in October 2020 warming background, seasonal prediction and physical processes related to the onset and evolution of extreme cold events are still a work in progress (Mu et al. 2020). Generally, the theory, method and technology of climate prediction are still in the exploratory stage, and climate prediction is a typical research meteorological business. The development of future climate prediction research is both an international hotspot and a scientific difficulty (Hsu et al. 2016(Hsu et al. , 2017Wang et al. 2020).

Data availablity
The monthly and daily reanalysis data used in this work are available in the National Centers for Environmental Prediction (NCEP)/ National Center for Atmospheric Research (NCAR) repository, http:// www. psl. noaa. gov/ data/ gridd ed/ data. ncep. reana lysis. deriv ed. html.
The high-resolution gridded temperature dataset for China used in this work are available from the researcher (gaoxuejie@mail.iap.ac.cn) on reasonable request.
The monthly and daily forecast results by CAS-ESM-c are available from the researcher (linrenping@mail.iap.ac.cn) on reasonable request. Fig. 9 (a, c) Represent the observed monthly mean 500-hPa geopotential height anomalies (relative to 1981-2010) in October and November 2020. (b, d) Represent the monthly mean differences between the best member and the worst member at the 500-hPa geopotential height anomalies in October and November 2020