Contribution of the Tibetan Plateau Winter Snow Cover to Seasonal Prediction of the East Asian Summer Monsoon

ABSTRACT How to improve the prediction skill of the East Asian summer monsoon (EASM) is a challenging but essential issue. This study examines the impact of the winter Tibetan Plateau (TP) snow cover (TPSC) on the subsequent EASM during the past two decades. Based on the high-resolution MODIS/Terra snow cover data, a new snow cover critical area (76°−83°E, 28°−35°N) is identified in the southwestern TP for the EASM seasonal prediction. Results show that the increase of the TPSC within this critical area during prior winter significantly increases summer precipitation over the Yangtze River Basin (YRB). The TPSC anomaly induces anomalous cooling in the overlying atmospheric column, leading to an anomalous cyclonic circulation in the upper troposphere. Such anomalous cyclonic circulation may further contribute to the local snow cover increase, and through such a snow-albedo feedback process, the excessive TPSC anomaly is strengthened and persists through the following summer. Coexisting with the positive anomalous TPSC, the South Asian High, the western Pacific Subtropical High, and the Subtropical Westerly Jet shift southward. A deep cyclonic circulation is induced in northeastern China by the excessive TPSC anomaly, which is reproduced in the linear baroclinic model simulation. Northerly flow is crucial for accumulating water vapour and favours more rainfall over the YRB. A physical empirical prediction model is established to quantify the TPSC contribution to the seasonal prediction of the EASM. Empirical hindcast output shows the prediction skill of the EASM is significantly improved with the additional predictor of the winter TPSC. In particular, the TPSC has greatly improved the prediction of the extreme EASM in 2020. The above results indicate that the prior winter TPSC anomaly in this critical area can provide another predictability source for the EASM, besides El Niño-Southern Oscillation and the North Atlantic Oscillation.


Introduction
As the main component of the Asian monsoon, the variability of the East Asian summer monsoon (EASM) dramatically affects the summer monsoon rainfall and drought/flood events in East Asia (Ding, 1992;Ding & Chan, 2005;Kitoh, 2004;Li et al., 2017). Accurate forecasting of the EASM can effectively enhance the defense capabilities of countries in this region to deal with various disasters caused by abnormal monsoons (Tang & Duan, 2021;Wang et al., 2005;. Therefore, improving the prediction ability of the EASM remains an important research area. Determining the sources of the EASM predictability is vital to the seasonal prediction of the EASM and associated uncertainties . El Niño-Southern Oscillation (ENSO) is widely regarded as the primary source of predictability of the East Asian subtropical summer monsoon (Gong & Ho, 2002;Wang et al., 2008). Gong and Ho (2002) found that the sea surface temperatures of the eastern tropical Pacific and tropical Indian Ocean affect the interdecadal variation of the East Asian subtropical summer monsoon by changing the intensity of the subtropical northwestern Pacific high. Wang et al. (2000) proposed the Pacific-East Asian teleconnection theory and pointed out that the anticyclone in the Northwest Pacific is a bridge connecting ENSO and the EASM. Wang et al. (2008) further revealed that ENSO increased after the 1970s, accelerating the air-sea coupling system between ENSO and the EASM. In addition to low latitudes, the sources of EASM predictability also come from high latitudes. Previous studies demonstrated that the North Atlantic Oscillation (NAO) is also a significant predictor of the EASM (Gong & Ho, 2003;Gu et al., 2009;Sung et al., 2006). Wu et al. (2009) investigated the relationship between the spring (April-May, AM) NAO and the EASM since the late 1970s. They pointed out that the NAOinduced SST anomaly tripole pattern over the North Atlantic Ocean plays a crucial role in linking the spring NAO and the EASM.
With an average altitude of more than 4000 m, the Tibetan Plateau (TP) is the highest and the most complicated plateau in the world, which is known as "the roof of the world" and "the third pole" (Kang et al., 2010). The high albedo, high emissivity, and low thermal conductivity of snow play a decisive role in controlling regional climate and energy balance (Brown & Mote, 2009;You et al., 2020). As an essential factor characterizing the thermal condition, the TP snow cover (TPSC) can not only control the energy exchange and water vapour transport between the underlying surface and the atmosphere through snow albedo, snow hydrological effect, and other processes but also has the ability to "remember" atmospheric signals. Many scholars have researched the snow thermal and dynamical impacts (Duan et al., 2011;Li et al., 2005;Lin & Wu, 2011;Liu & Chen, 2011;Qian et al., 2003;Wang et al., 2018;Wu et al., 2012;Wu et al., 2016). Qian et al. (2004) considered that the snow anomaly over the plateau in winter and spring, especially in the early winter, can generally modulate the location and intensity of the South Asian high (SAH) and have an impact on the Asian monsoon precipitation. Zhao et al. (2010) analyzed the influence of the snow cover anomaly over the TP on the sea-land thermal difference between Asia and the Pacific, East Asian monsoon, and precipitation under the background of global warming. Xiao and Duan (2016) revealed that the preceding winter snow cover extent anomalies over the Himalayas can prolong the signature to the following summer and impact the summer interannual rainfall variability along the Yangtze River Basin (YRB) through eastward-propagating synoptic disturbances and their attendant water vapour transport. Jin et al. (2018) demonstrated that the TP-sensible heat air pump-related potential vorticity forcing favours stimulating a large-scale cyclonic system surrounding the TP, providing favourable moisture conditions for the YRB summer rainfall.
However, the observation datasets of TPSC are lacking due to the unique geographical environment and inaccessibility. For the anomaly of TPSC, different authors have obtained different results by using snow data from different periods and sources, and the physical mechanism of TPSC affecting the EASM anomaly is still controversial (Liu et al., 2014;Wu et al., 2007;Wu et al., 2012;Wu et al., 2015;Yanai & Wu, 2006;Zhang et al., 2004). Although there are more than 100 meteorological stations on the TP with the support of the China Meteorological Administration, most of them are in the east and middle of the TP. In snowy mountain areas, especially in the west of the TP, there are few or no stations (Xiao et al., 2007;Yang et al., 2019). Due to its high resolution in both space and time, the Moderate Resolution Imaging Spectroradiometer (MODIS) data has a great potential for monitoring snow cover change in regions with complex terrain Pu & Xu, 2009), and has already been used extensively to examine the distribution and seasonal changes of snow cover over the TP (Yang et al., 2015). Yang et al. (2015) discovered the overall accuracy of the MODIS Terra and Aqua Snow Cover Daily L3 Global 500-m Grid products (MOD10A1 and MYD10A1) is higher than 91% against stations observations. Pu et al. (2007) found that the accuracy of the MODIS/Terra Snow Cover 8-Day L3 Global 500-m Grid product (MOD10A2) neared 90% when compared with meteorological station observations in the TP. The MODIS/Terra Snow Cover 8-Day L3 Global 0.05 Degree CMG product (MOD10C2) was used to investigate the spatial and temporal distribution of snow cover over the Tibetan Plateau because this product virtually eliminates cloud obscuration (Pu & Xu, 2009). We also compare the MOD10A2 data with the in-situ data to show the reliability and uncertainty of the satellite data for the period March 2000 to December 2021 (Fig. S1). The overall accuracy of the MODIS snow detection rate is 86.37%, while the rates of omission and commission errors are 3.71% and 9.92%, respectively (Tables S1 and S2). Generally, the MODIS snow cover data shows a good accuracy over the TP with the ability to identify non-snow pixels and reduce cloud distractions. Therefore, it is reasonable to use the MOD10C2, derived from the high-resolution 500 m observations, to do the seasonal forecast.
Then, using high-resolution MODIS snow cover data, can we find that the previous TPSC provides a forecasting signal for the EASM? If so, how about the corresponding physical mechanism? To settle the above issues, this paper is organized as follows. The datasets and methods are introduced in section 2. The observed relationship between the TPSC and the EASM is presented in section 3. Section 4 investigates the possible physical mechanism of how winter TPSC affects the EASM variations. Section 5 performed seasonal forecast experiments for the EASM. In the final section, the main conclusions are summarized, and some outstanding issues are presented.
Niño 3.4 index (defined within the area of 5°N−5°S, 120°E −170°W) is obtained from the Climate Prediction Center (CPC). The NAO index (NAOI) is defined by Li and Wang (2003), which is the optimal representation of the spatio-temporal variability associated with the NAO. It is designated as the normalized monthly sea level pressure zonally-averaged over the North Atlantic sector (80°W-30°E) at 35°N minus that at 65°N. The index to describe the EASM variability (EASMI) is designated as the regional average 850-hPa zonal wind anomaly in 22.5°-32.5°N, 110°-140°E minus that in 5°-15°N, 90°-130°E, which is best correlated with the leading principal component of the EASM and reversed to the original definition (Wang & Fan, 1999).
b Model Numerical experiments carried out in this work are based on the linear baroclinic model (LBM), which is developed on the dynamical core of the Atmospheric General Circulation Model (AGCM) and designed by the Center for Climate System Research, University of Tokyo, and the National Institute for Environmental Studies, Japan (Watanabe & Kimoto, 2000). The numerical model can reproduce the linear atmospheric response via time integration method by imposing climatological mean flow and the idealized forcing associated with observational anomalies in the critical area. Here, we employed the dry version with 20 vertical sigma levels and a horizontal T42 resolution. Figure 1 displays the distribution of the June-July-August (JJA) precipitation anomalies regressed to the EASMI. The red boxes indicate the regions where the EASMI is defined, also the areas where the subtropical front and the monsoon trough are located. The distribution of precipitation shows a meridional tripolar distribution, with wet anomalies along with the Yangtze River Basin and over the Maritime Continent, and dry anomalies in the northern South China Sea (SCS) and the Philippine Sea. The rainfall distribution is closely connected with the red boxes, which reflects the inverse phase relationship between the subtropical front and the monsoon trough.

EASM associated with winter TPSC
To determine the region and season of the TPSC that may be related to the variation of the EASM, the temporal correlation coefficients between the EASMI and the TPSC are calculated from the previous winter to the simultaneous summer, as presented in Fig. 2. In the summer period, several areas of anomalous TPSC regressed against the EASMI are scattered in the central and eastern TP regions (Fig. 2a). In the previous Fig. 1 Anomalous JJA mean precipitation rate (shading, unit: mm/day) regressed to the EASM index (EASMI) for 2001-2020. The dotted areas represent the significant anomalies at the 95% confidence level. The EASMI is an inverse Wang-Fan index (Wang & Fan, 1999) and is defined as the 850-hPa area-averaged zonal wind difference between the two red boxes (the north box minus the south box).
spring, the significant positive TPSC is observed over the southwestern and northeastern TP (Fig. 2b). The most significant TPSC is observed in the prior winter when a large area of the middle and eastern part of the TP is covered by positive TPSC (Fig. 2c). It is worth noting that the positive anomalous TPSC in the southwest of TP in spring has appeared in the same region in the early winter and continues to the following summer, which means that this region is likely to be the critical area. The TPSC anomalies to composite difference of the EASM index (e.g. strong minus weak monsoon) also shed some light on the robustness of the relationship (Fig. S2). We classify a year as a positive or negative year when the EASMI is greater than 0.5 standard deviation or less than −0.5 standard deviation, respectively. There is still a robust relationship between the EASMI and the TPSC from winter to summer in the critical area. During the half-year time period, considerable snow cover typically remains in this critical area (Fig. S3). The mean TPSC in winter is 49% with standard deviation of 29% for the yearly values, whereas summer TPSC remains significant with a mean of 25% and standard deviation of 27% for the yearly values.
These percent values are calculated based on the part of the purple box that is north of the black line.
To quantitatively measure the variation of the TPSC, the snow cover index (TPSI) is constructed by averaging the snow cover anomaly averaged within the TP domain (76°-83°E, 28°-35°N) in the purple box (excluding the white region south of the black line and considering only the part of the box that is north of the black line) according to Fig.  2. Figure 3 shows the temporal evolution of the normalized EASMI and its corresponding normalized TPSI in the previous winter for the past 20 years (2001-2020). The two time series show substantial year-to-year variabilities, and their changes have good synchronization. The correlation coefficient between the time series is about 0.59, beyond the 99% confidence level based on the Student's t-test. It suggests that the EASMI is closely correlated with the variation in the winter TPSC over the southwestern TP.
The lead-lag correlation between the EASMI and the TPSI is then calculated from the previous December-January-February (D(−1)JF(0)) to the following D(0)JF(+1) and is presented in Fig. 4, where "−1" represents the previous year, "0" is the simultaneous year, and "+1" means the next year. The two time series are correlated at a confidence level above 95% from the previous DJF until the previous May-June-July (MJJ(0)) when the changes in the TP snow lead the EASM, while the correlation is nearly 90% confidence level during the simultaneous JJA(0). Figure 4 reveals that the variation in the TPSC of the southwestern TP is closely  correlated with the EASM and might be used as a predictor for the EASM with a lead time up to nearly half a year.
In contrast, the TPSC over the central and northeastern TP does not present the same kind of persistence (Fig. S4). Once we select a larger area, including most of the positive TPSC in the previous winter (Fig. S4a), the lead-lag correlation between the EASMI and the TPSI does not last until the previous spring (Fig. S4b), which is consistent with the determination by Xiao and Duan (2016) that the preceding snow cover over the central and eastern TP exerts little influence over either the in-situ summer atmospheric heat source or the EASM due to its limited persistence. In addition, the EASMI appears to be anti-correlated with the TPSI in the following winter (Fig. 4), which is more evident in Fig. S3b. Indeed, correlation does not equate to causality, and the physical mechanisms involved need to be further investigated. Perhaps it is also a result that both the EASM and the following TPSC are affected by the third-party factors, such as ENSO. In addition, correlations are not necessarily transferable, i.e. moderate correlations (or anti-correlations) between A vs B and B vs C do not necessarily imply that A is significantly correlated with C. The association of the EASMI with the TPSI in the preceding and following winters does not result in a tendency for the EASMI to be anti-correlated with itself in successive years.
To better understand the TPSC-EASM linkage, the regression of the JJA precipitation anomalies on the preceding winter TPSI is depicted in Fig. 5. The results are pretty similar to Fig. 1. The rainfall distribution shows a north-south dipole precipitation pattern of dry anomalies in the northern South China Sea and the Philippine Sea and wet anomalies along the Yangtze River valley to southern Japan. Compared with Fig. 1, the positive anomalous area over the Maritime Continent is much smaller. The northern SCS and the Philippine Sea still show negative precipitation anomalies with roughly the same area. The positive anomalies along the Yangtze River Basin become narrower and longer, and the tail extends eastward. Figure 6 presents the regression of the JJA low-level (850-hPa) wind on the preceding winter TPSI and marks the critical areas defining the EASMI. An apparent inverse phase of the EASM related wind anomalies can be observed between the north and south regions. The increased TPSC corresponds to the anomalous westerly wind of the subtropical front and the anomalous easterly wind of the monsoon trough, in other words, the increased EASMI. The above analysis indicates that the positive relationship between the preceding winter TPSC and the ensuing EASMI inter-annual variability is statistically robust. Then, how can the TPSC impact the downstream EASM? Fig. 6 The regression map of the JJA 850-hPa wind anomalies (vector, unit: m/s) upon the preceding winter TPSI for 2001-2020. The orange and dark vectors indicate the significant anomalies at the 90% and 95% levels, respectively. Fig. 4 The lead-lag correlation coefficients between the EASMI and the TPSI from the previous D(−1)JF(0) to the following D(0)JF(+1). "−1" represents the previous year, "0" is the simultaneous year, and "+1" means the next year. The dashed lines represent the 90% and 95% confidence levels according to Student's t-test.

Physical mechanisms a Local Effects of TPSC Changes
To shed light on the possible cross-season snow-monsoon linkage, the time evolution of the snow anomalies over the southwestern TP from winter to the subsequent summer is examined. Figure 7 shows the lag correlation coefficients between the winter TPSI and the TPSI from DJF to JJA. The lag correlation coefficient of the TPSI decreased significantly from winter to spring. By the spring (March-April-May, MAM) period, the correlation coefficient with the previous DJF period has reduced to 0.52, which is beyond the 95% confidence level based on the Student's t-test. From spring to summer, the decline rate of the lag correlation coefficient slowed down. By JJA, the correlation coefficient is 0.47, which still passes the 95% confidence level. It shows that the anomalous snow over the southwestern TP has good persistence during the study period. As with the leadlag correlation between the EASMI and the TPSI (Fig. S4b), the persistence of the lag correlation coefficient of the TPSI over the larger area failed to last into the spring (Fig. S4c).
To investigate the local effect of the TPSC anomalies on the atmosphere, Fig. 8 displays the time evolution of the TPSI-related sensible heat (SH) anomalies from DJF to the subsequent JJA. Corresponding to more snow cover in the southwestern Plateau, there is a decrease in the surface SH flux, which signifies the albedo effect of snow cover. Comparing Fig. 2c and 8a, it can be found that the significant positive TPSC in the critical area exactly corresponds to the significant negative SH, and the same is true in Fig. 2a and 8d.
The distribution of the surface net thermal radiation (SNTR) anomalies is shown in Fig. 9. The radiation analysis shows that more snow cover at the surface causes decreased the SNTR over the southwestern TP. During DJF and FMA, the negative range of the SNTR far exceeded the critical area (Fig. 9a,b), which may be related to the local positive TPSC anomalies (Fig. 2c). Afterward, as the positive TPSC anomalies are no longer present in the central TP (Fig. 2a,  b), the negative SNTR anomalies in these areas also disappear and even turn positive during AMJ and JJA (Fig. 9c,d).
As the presence of snow cover suppresses the energy exchange between the land surface and the overlying atmosphere, pronounced negative surface air temperature (SAT) anomalies are observed over the TP from winter to the following summer (Fig. 10). Unlike the thermal radiation in Fig. 9, the SAT is affected by the TPSC in a much wider range, and the critical area is almost entirely covered by significant positive SAT anomalies, especially in summer (comparing Figs 9d and 10d). The positive TPSC anomalies that last from winter to summer exerts a strong and continuous cooling effect on the SAT. The difference in the location between Fig. 7 The lag correlation coefficients between the winter TPSI and the TPSI from DJF to JJA. The dashed lines represent the 90% and 95% confidence levels according to Student's t-test. the TPSC and the SAT decreasing regions may be due to the contribution of land-atmosphere interactions to the SAT variations. TPSC-induced negative temperature anomalies prevail over the whole troposphere, which can be viewed from the regression maps of the longitude-pressure cross sections of air temperature drawn in Fig. 11. Under the cooling effect of snow cover, there are significant cold anomalies from surface to 200-hPa above the TP. Of course, the intensity and scope of the negative anomalies would weaken over time, with the strongest in winter (Fig. 11a) and the weakest in summer (Fig. 11d). Figure 12 displays the transformation of the TPSI-related large-scale atmospheric circulation anomalies in the upper troposphere from winter to summer for each epoch. It can be seen that most areas of the TP are always controlled by the conspicuous negative geopotential height anomalies, which is consistent with the previous cold temperature anomalies. An evident cyclonic circulation anomaly prevails mainly over the southwestern TP, associated with the positive Correlation coefficients significant at the 95% confidence level are denoted by the black dots. snow cover anomalies there. This cyclonic system, consistent with the anomalous positive TPSC and anomalous cooling over the southwestern TP, displays the atmospheric response to the underlying snow forcing. Meanwhile, the anomalous cyclonic circulation may favour the anomalous convective activities in the southwestern TP, which may contribute to positive snow cover anomalies there. Figure 13 shows the time evolution of the snowfall anomalies from DJF to the subsequent JJA. Corresponding to the anomalous cyclonic circulation, there are significant positive snowfall anomalies in the critical area, which also provides evidence for the persistence of the TPSC during the study period.
Since the snow cover and surface variables are from different data sources, the temporal correlation coefficients between the EASMI and TPSC of ERA5-Land reanalysis are calculated from the previous winter to the simultaneous summer (Fig. S5). Compared with Fig. 2, the correlation between the EASM and the MODIS data is similar to the EASM and the ERA5-Land reanalysis, especially in the critical area.  The relationships of the TPSC in ERA5-Land with other surface variables are also shown (Figs S6-S9), and the distribution patterns are similar to those displayed above (Figs 8-10 and 13). In addition, for the possible positive bias in the surface variables in ERA5-Land (Lin et al., 2021;Orsolini et al., 2019), we verify our results against another reanalysis, where the snow bias is diminished, like MERRA-2 (Gelaro et al., 2017). Corresponding to positive TPSC anomalies in the southwestern Plateau, there are persistent negative anomalies of the SH and the SAT and positive anomalies of the snowfall from DJF to JJA in the critical area (Figs S10-S12), indicating that this bias is not playing an important role and the selection of critical areas is undoubtedly correct. In general, the above suggests that once the anomalous winter TPSC has occurred over southwestern TP, the snow-albedo effect will be stimulated to maintain the positive TPSC anomalies from winter to summer. The main process is proposed as follows: the positive TPSC anomalies can exert cooling effects on the overlying air by modulating the energy budget. The cooling effect tends to stimulate the negative potential height anomalies and cyclone system, which in turn, leads to a stronger TPSC. It is this positive feedback process that further strengthens the TPSC until summertime.
b Remote Effects of TPSC Changes This section will discuss the circulation changes in the surrounding areas related to the snow cover anomalies over the TP. Two anomalous cyclones are observed over the southwestern TP and northeast China to Japan island, respectively (Fig.  12d). The enhanced westerlies occur between 20°and 35°N, and the intensified easterlies are found between 40°and 50°N . The climatological subtropical westerly jet (SWJ) is overlaid as contours in Fig. 14a. The positive zonal wind anomalies are located on the south of the SWJ, while the negative anomalies are on the north, indicating that the position of the SWJ is southerly, coexisting with the positive anomalous snow cover. The YRB is located under the anomalous westerly winds in the southwestern part of the cyclone circulation. The air currents diverge at high altitudes, which is conducive to ascending movement, in turn causing excessive precipitation.
The climatological South Asian High (SAH) is overlaid as contours in Fig. 14b. The SAH is a solid and stable semi-permanent high system in the upper troposphere and lower stratosphere during boreal summer over Eurasia. The positive geopotential height anomalies are located in the low latitude region, and a low-pressure anomaly is observed over the southwestern TP. There is a positive anomalies belt in lowlatitude and a negative in mid-latitude (Fig. 14c). The western Pacific subtropical high (WPSH) position is southerly, and its intensity is stronger. At mid-tropospheric 500-hPa (Fig. 15a), the centre of the abnormal cyclone circulation from northeastern China to Japan shifts eastward by about 8 degrees and deepens southward. Anomalous southerly winds prevail in the YRB, which contributes to the rainy circulation in the region.
As for 850-hPa (Fig. 15b), corresponding to the positive anomalous TPSC, the anomalous cyclone circulation still occupies the vicinity of northeastern China to Japan, and the anomalous northerly airflow on the southwest side flows south to the YRB. Simultaneously, the anomalous southerly airflow is located north of the SCS to the YRB, bringing plenty of water vapour. The low-level wind field converges over the YRB, and an anomalous ascending movement zone Dotted areas indicate the significant anomalies at the 95% levels.

Contribution of the Tibetan Plateau Winter Snow / 33
occupies the YRB to southern Japan (Fig. 15c), highly contributing to the precipitation during this period. The anomalous westerlies between 20°and 30°N indicate an anomalous positive EASMI. Considering the reversed EASMI used in this study, the significant positive TPSC actually weakens the EASM and increases precipitation in the YRB, which is consistent with the traditional Chinese meaning of a strong EASM corresponding to a deficient Meiyu.
To diagnose the effect of the positive anomalous TPSC on the downstream atmospheric circulation, we imposed a cooling experiment by adding southwestern TP cooling anomalies with the summer basic state in the LBM (Fig. 16). The cooling has a cosine-squared profile in an elliptical region, with a centre at 80°E, 32°N. The radiuses of the region in latitudinal and longitudinal directions are 8°N and 8°E (Fig. 16a). In the vertical, the cooling has a sinusoidal profile with a maximum at sigma of 0.75 (about 400 hPa) to  mimic the TPSC-induced diabatic cooling effect. The maximum cooling is 1 K/day (Fig. 16b). The model is integrated for 40 days, and the variables at last 10 days as the stabilizing state are then averaged for further analysis. Figure 17 provides the model simulated atmospheric circulations according to the experiment. Corresponding to the southwestern TP cooling forcing, two evident upper tropospheric local cyclones overlaid the low-pressure systems over the southwestern TP and northeast China (Fig. 17a) are similar with the southwestern TPSC induced circulation in Fig. 12d. Moreover, the cyclone system over the northeast China is also reoccurred at 500-hPa and 850-hPa (Fig. 17b,  c), which is consistent with the observed circulation in Fig.  15a and b. Furthermore, there are obvious northwestly winds in the southwest of the cyclone, which is also conducive to the convergence of the wind field and the occurrence of precipitation in the YRB.
Because the experiment with the LBM does not include the feedback of moisture processes, the possible role of snow hydrological effect of the TPSC is not reflected in the results (Fig. 17). The detailed characteristics, e.g. the observed westerly in Fig. 15a

Seasonal prediction
Previous studies have pointed out that ENSO is the critical predictor of the EASM variation (Wang et al., 2008;Xiao Fig. 16 (a) The spatial pattern of the cooling forcing (shading, unit: K/day) at the sigma level of 0.75 and (b) the vertical profile of the cooling forcing (blue curve, unit: K/day) around the horizontal maximum cooling centre. . Wang et al. (2009)  In this study, the anomalous winter TPSC is demonstrated to be an additional predictor for the EASM. To test how well the winter TPSC contributes to the prediction skill of the EASM and restricted by the length of the TPSI, we apply the cross-validation method using the NAOI, ENSO decay , ENSO develop and the TPSI to hindcast the EASMI. Based on the strategy of Blockeel and Struyf (2003), to prevent wasting data or overfitting, test years accounting for 20% −30% of the data should be proposed, and the remainders are used as the training set to establish the forecasting regression model. In this study, 25% of the entire hindcast period (20 years) equals five years. Thus, the leave-five-out strategy is chosen for the forecasting experiments. The relevant procedures for calculating the hindcast values are as follows: the period 2021-2020 is divided evenly into four segments of five years each, and the hindcast values for each component are calculated from the model obtained through the other three segments.
The hindcast result is presented in Fig. 18 (red curve), and the correlation coefficient with the observational EASMI is 0.66 (reaching the 99% confidence level). Significantly, the hindcast result for 2020 has been greatly improved after adding the TPSI as the prediction factor. Given the YRB was hit by exceptional rainstorms in 2020, these results imply that the winter TPSC is an invaluable predictor and can enhance the seasonal forecast skill of the EASMI variation. Moreover, the 20-year cross-validated empirical model estimates based on the NAOI, ENSO decay , and ENSO develop are also established (not shown), with a correlation coefficient of 0.30 (below the 90% confidence level).

Conclusion and discussion
The influence of the TP on the EASM prediction has long been noticed, but the influential factors and the physical mechanisms have not been determined. The TPSC may have a significant impact on the inter-annual variability of the monsoon because it can change the surface albedo and regulate soil moisture, which in turn will act as a persistent or long memory signal to affect the subsequent monsoon circulation (Jin et al., 2018;Zhang et al., 2021). Ding et al. (2015) pointed out that under the influence of global climate change, the driving forces and characteristics of the East Asian monsoon system have changed. In the present article, we have examined the potential influence of the previous winter TPSC on the subsequent EASM inter-annual variability since the beginning of the twenty-first century and revealed the relevant physical mechanism.
Based on the high-resolution MODIS/Terra snow cover data, we define a new snow cover critical region within the TP domain (76°−83°E, 28°−35°N). Our analyses show a cross-season relationship between the winter TPSC over the southwestern TP and the EASM, and the winter TPSC might be used as a predictor for the EASM with a lead time up to nearly half a year. Further analyses imply that the anomalous winter TPSC can stimulate the positive feedback process of the snow-albedo effect. The anomalous TPSC weakens the local surface SH flux and leads to the decrease of the SNTR, indicating that the snow cover blocks the heat exchange between the land surface and the upper atmosphere. Negative SAT anomalies are observed, and the cold anomalies propagate to the upper troposphere. A cyclonic circulation anomaly prevails over the southwestern TP, which favours the anomalous convective activities, accompanied by the positive snowfall anomalies that help maintain the snow cover there.
Coexisting with the positive anomalous snow cover, the SAH, the WPSH, and the SWJ shift southward. In northeastern China, a deep column of cold air is induced by the significant positive TPSC, which is also revealed in the LBM simulation. The northerly wind on the southwest side of the cold air column collides with the southerly wind on the west side of the subtropical high over the YRB. Coupled with the increased upward movement of the air above, more rainfall is formed over the YRB. Since the winter TPSC may act as the new predictability source for the EASM, we further inspected the prospective predictability by establishing empirical prediction models in which ENSO, the NAO, and the TPSC are used as the predictors. The results from the leave-five-out cross-validation approach for the period of 2001-2020 display a more considerable skill by introducing the TPSC as an additional predictor than the previous method using ENSO and NAO as factors. In particular, the TPSC has dramatically improved the prediction of extreme summer monsoon in the YRB in 2020. Therefore, the close linkage between the winter TPSC and the EASM provides another physical background for the seasonal prediction of the EASM. This work suggests that the TPSC has greatly enhanced the driving capacity of the EASM since the beginning of the twenty-first century. However, under the background of global warming, the extreme events caused by monsoons gradually increase, and most of them occur on a sub-seasonal time scale. Can this cross-season relationship between the TPSC and the EASM still play a role on the sub-seasonal scale? If not, what are the sources of predictability for the sub-seasonal extreme events? Li et al. (2020) pointed out that subseasonal-to-seasonal (S2S) models can skillfully forecast TPSC within a lead time of two weeks but show limited skill beyond three weeks, influencing the forecasted SAT in the S2S models. Compared with the climate forecast discussed above, the demands of operational forecasts in the S2S range are growing rapidly, which is a great challenge to the simulation of land-atmosphere coupling. In addition, except for the TPSC, whether other external factors contribute to the change of the EASM is unclear and beyond the scope of the current work. Future work should be done to understand the EASM variability further.

Disclosure statement
No potential conflict of interest was reported by the author(s).