Decadal trend of synoptic temperature variability over the Northern Hemisphere in winter

Synoptic temperature variability gives rise to cold waves and extreme cold events in winter. Based on four reanalysis datasets, this study investigates the decadal trend of synoptic temperature variability in boreal winter during the period from 1980 to 2019, with particular focus on the sharp drops in synoptic-scale temperature, which are associated with cold waves. The result shows that the synoptic-scale standard deviation of temperature decreases significantly with a trend of − 0.15 K/decade (− 0.09 to − 0.21 K/decade among reanalysis datasets) over continental regions in mid-high latitudes. Correspondingly, the rapid cooling events (RCEs), defined based on the day-to-day temperature decrease exceeding 6 K, also show a general decreasing trend in terms of their frequency and intensity. The strongest decrease occurs over eastern North America (ENA) and western Eurasia (WE). The weakening of the RCEs is closely connected to the decreased trend of eddy kinetic energy (EKE), suggesting that the weakened transient eddy activities may have mitigated the synoptic-scale temperature variability and the associated RCEs over mid-high latitudes. This study highlights that the decreased synoptic temperature variability leads to fewer and weaker RCEs on the synoptic scale over mid-high latitudes in winter though the mean state of winter temperature continues to warm.


Introduction
Most meteorological disasters in winter are associated with cold waves. Cold waves bring cold air from polar regions to mid-latitudes, resulting in an abrupt drop in temperature and fierce wind on the synoptic scale (Lau and Lau 1984;Chang and Chen 1992;Zhang et al. 1997), such as the cold wave that invaded East Asia in late January 2016 and the cold spell that swept much of the northern USA in winter, 2021 (Herring et al. 2018;Ma and Zhu 2019;Yamaguchi et al. 2019;Doss-Gollin et al. 2021). During the cold wave outbreak, the abrupt decrease in temperature raises the risk of cardiovascular diseases and relative mortality (McGeehin and Mirabelli 2001;Shi et al. 2015;Ballester et al. 2016;Wang et al. 2016). In addition, the sustained low-temperature condition may aggravate the burdens of traffic and energy supply (Zhou et al. 2011;Kim and Lee 2019) and reduce agricultural production (Wheeler et al. 2000;Liu et al. 2006;Lobell et al. 2011). Moreover, under appropriate moisture conditions, cold waves may also bring snowstorms or freezing rain, leading to catastrophic effects on human society, like the persistent freezing rain in southern China in January 2008 (Hong and Li 2009;Zhou et al. 2011).
Given the importance of cold waves to human society, many studies have been devoted to long-term change in cold waves. Most have adopted metrics based on the extreme cold temperature, such as frost days (FD), cold nights (TN10p), daily minimum temperature (TNn), and cold spell duration index (CSDI) (Meehl and Tebaldi 2004;Alexander et al. 2006;Anandhi et al. 2016;Sheridan and Lee 2018;van Oldenborgh and Mitchell-Larson 2019). Most previous studies have consistently shown a decreasing trend in winter cold extremes under the background of global warming based on both observational records (Park et al. 2011;Vavrus et al. 2006;Smith and Sheridan 2020) and model simulations under anthropogenic forcing (Bell et al. 2004;Boo et al. 2006;Yang et al. 2020). Indeed, weakened cold extremes and enhanced hot extremes have been reported by previous studies (Schar et al. 2004;Meehl and Tebaldi 2004;Griffiths et al. 2005). Horton and Johnson (2015) showed that the occurrence of cold extremes, which are defined as the daily minimum temperature below the 5th percentile of the distribution, decreased by nearly 0.2 times/year over North America in winter from 1979 to 2013. Park et al. (2011) comprehensively investigated the change of cold days (daily mean temperature anomalies below 1.5 σ) after the mid-1980s over South Korea and found a significant negative trend of − 7.52 times/decade in the occurrence of cold days, though the seasonal mean minimum temperatures increased significantly by 0.54 K/decade. Over Europe and China, extreme low daily temperature has also decreased significantly (Gong and Ho 2004). A decreasing trend in extreme low temperature under a continuous warming background is not surprising, but the trend of synoptic-scale temperature variability has been less studied.
In mid-high latitudes, the synoptic temperature variation is controlled by horizontal advection associated with weather systems, and the advection depends on temperature gradient and wind (Schneider et al. 2015;Holmes et al. 2016). The northerly (southerly) wind usually brings cold (warm) advection, which leads to the decrease (increase) in the local temperature on synoptic scale. In addition, the change in temperature variability may be primarily independent of the increasing mean state temperature under a changing climate. Previous studies have focused more on long-term changes in daily minimum temperature or seasonal mean temperature (Wang et al. 2005;Fischer et al. 2012;Lee et al. 2013;Lim and Kim 2013). During winter, a synoptic-scale temperature drop within several days also has a great impact on human health (McGeehin and Mirabelli 2001;McMichael et al. 2006). Some current studies found that interannual variability of synoptic cold wave activity could be influenced by the El Niño/Southern Oscillation (ENSO) (Chen et al. 2004(Chen et al. , 2013. ENSO can induce the northward displacement of the East Asian trough and modulates the seasonal mean temperature variability over the East Asian-Western Pacific region indirectly (Leung and Zhou 2016;Leung et al. 2017). However, it is unclear whether the synopticscale decrease in temperature is becoming stronger or weaker under the warming condition. In this study, we aim to answer the following question: has the synoptic temperature variability been increasing or decreasing in recent decades and what caused its decadal trend? We focus primarily on synoptic-scale cooling events because they have a much stronger societal impact than synopticscale warming during winter.
The remaining sections of this paper are organized as follows. The data, metrics, and methods are described in Section 2, and the climatology and decadal trends of temperature variability are presented in Section 3, along with a discussion of the possible mechanism. The major conclusions are summarized and discussed in Section 4.

Data
Four reanalysis datasets are adopted in this study, which include (1) the NCEP/NCAR Global Reanalysis Product 1 (NCEP/NCAR) (Kalnay et al. 1996), (2) the NCEP-DOE Reanalysis version 2 (NCEP-DOE) (Kanamitsu et al. 2002), (3) the Japanese 55-year Reanalysis (JRA-55) (Kobayashi and Iwasaki 2016), and (4) the ERA5 Global Reanalysis (ERA5) (Hersbach et al. 2020). The four datasets cover a common period ranging from January 1, 1979 to November 30, 2019, and only the period after 1979 is used. For each dataset, the following variables spanning from December 1to February 28 in each winter are adopted: the daily mean surface temperature at 2 m above ground (T 2m ) and the zonal (u) and meridional (v) components of wind at 500 h Pa. The winter of a year is represented as the 3-month period from December of the previous year to February of the current year. For example, the winter of 1980 means December 1979 to February 1980. Therefore, 40 winters from 1980 to 2019 are available for all four datasets, and the climatology and trends are also calculated based on these 40 winters.

Definition of rapid cooling events (RCEs)
Synoptic-scale decreases in daily temperature have much stronger societal impacts in winter and are directly associated with cold waves (Ryoo et al. 2005;Cattiaux et al. 2010). However, long-term changes in cold waves may be masked by the background global warming because the low-temperature-related metrics are adopted (Alexander et al. 2006;Park et al. 2011). Following the previous studies (Schneider et al. 2015;Blackport et al. 2021), the synoptic-scale standard deviation of daily temperature (SSD), calculated as the standard deviation of 2-6 day bandpass-filtered time series of daily temperature, is adopted to measure the overall amplitude of synoptic-scale temperature variability. However, the SSD does not directly describe the frequency and intensity of sharp temperature drops in winter. For example, the two idealized 12-day time series of daily temperature have the same standard deviations, but there are 5 (0) days with a sharp drop in temperature (the difference of two adjoint days less than -6 K) based on the dashed black (solid red) curve ( Fig. 1). Some new metric is needed since the SSD of temperature may not fully describe the frequency and intensity of the sharp temperature drops.
To accurately describe the sharp temperature drops during winter, this study also uses day-to-day temperature difference to measure the occurrence and intensity of synoptic-scale variability in temperature, motivated by previous studies (Lau and Lau 1984;Ryoo et al. 2005;Park et al. 2011). Based on the T 2m data, the day-to-day temperature difference can be calculated as where ∆T represents the difference in daily mean temperature between any two consecutive days of a winter. Here, yr stands for the index of a year (yr = 1980, …, 2019) and i stands for the index of a certain day within the winter of a year (i = 1, …, 89). The daily mean temperature is adopted rather than daily minimum temperature or daily maximum temperature (T min and T max ), because daily mean temperature is more closely connected to synoptic variability such as frontal activity and less affected by clouds and radiative processes on sub-daily scale (Tang et al. 2012;You et al. 2018). For example, a sunny day following a cloudy day may experience an increase in T max and a decrease in T min due to increased daytime shortwave heating and increased nighttime long wave cooling even without the activity of weather systems (Curry et al. 1996;Stephens 2005). Synoptic-scale decrease in temperature is associated with cold frontal systems, and previous studies also used daily mean temperature to measure frontal activities (Serreze et al. 2001;Hayasaki et al. 2006).
The occurrence of a rapid cooling event (RCE) is defined as the day-to-day temperature difference (∆T) below − 6 K for 2 consecutive days, which matches the threshold adopted  (Zhang et al. 1997;Hayasaki et al. 2006). We also tried to adjust the threshold to − 4 and − 8 K, and percentile thresholds (5th and 10th percentiles), and we find that the selection of threshold does not alter the main conclusion (Fig. S1). Based on the definition of RCEs described above, the frequency, mean, and extreme intensity of RCEs are calculated. The RCE frequency is defined as the times of the occurrence of RCEs at each grid point during each winter, and the mean intensity and the extreme intensity of RCEs are defined as the average and minimum of ∆T for all RCEs at each grid point during each winter. Here, we designate a symbolic function, (i, yr) , to judge whether an RCE occurs, in which (i, yr) = 1 for days with an RCE and (i, yr) =0 for days without an RCE. The frequency (FRQ), mean intensity (MIT), and extreme intensity (EIT) of RCEs at a grid point for year yr are expressed as where ∑() indicates the sum and Min() indicates the minimum value among the 89 days (i = 1, 89) of a winter.

Eddy kinetic energy (EKE)
The EKE is employed to measure the strength of transient Rossby waves and is related to the weather variability in the mid-latitudes, like the cold frontal activities (Lau and Lau 1984;Schubert et al. 2011). In this study, low frequency oscillations of wind are removed, and synoptic fluctuations are retained by applying a 2-6 day bandpass filter, similar to Coumou et al. (2015). The EKE is calculated based on the filtered time series of zonal and meridional winds as where u ′ and v ′ stand for the bandpass-filtered zonal and meridional wind components.
(4) EIT(yr) = Min( (i, yr) * ΔT(i, yr)) Fig. 3 The spatial averaged climatology of SSD of temperature, frequency of RCEs, mean, and extreme intensity of RCEs within 40-80°N based on the four datasets (NCEP/ NCAR, NCEP-DOE, JRA55, and ERA5). The black dashed line represents the mean value of different datasets Figure 1a shows the climatology for the synoptic-scale standard deviation of temperature (SSD) in winter (Fig. 2a). The frequency, mean and extreme intensity of RCEs based on the JRA55 dataset are given in Fig. 2b-d. As Fig. 2a shows, the continental synoptic temperature variability is much greater over the mid-high latitudes within 40 to 80°N, with two high-variability centers located over eastern North America and inland Eurasia. Like the spatial pattern for SSD, the frequency of RCEs shows a maximum over mid latitude North America and inland Eurasia (Fig. 2b). The mean intensity and extreme intensity of RCEs also show a similar spatial pattern to the SSD, with larger intensity over east-central North America and inland Eurasia in the midlatitudes ( Fig. 2c and d). In general, the spatial patterns for synoptic temperature variability are consistent among the four metrics above (Fig. 2a-d) and are consistent with the other three reanalysis datasets (figures not shown).

Climatology
The spatially averaged climatology for SSD, the frequency, mean, and extreme intensity of RCEs based on all continental grids within 40-80°N are presented in Fig. 3.
Based on four datasets, the SSD ranges from 5.67 to 6.56 K over the mid-high latitudes (Fig. 3a), and about 5.19 to 9.68 RCEs may occur in each winter (Fig. 3b). Meanwhile, the mean intensity of RCEs ranges from 8.07 to 9.05 K (Fig. 3c), and the extreme intensity of RCEs may reach 10.27 to 13.90 K, as averaged among the 40 winters from 1980 to 2019 (Fig. 3d). The four individual datasets all show a convergent climatology of SSD and the mean intensity of RCEs (Fig. 3a, c). The frequency and extreme intensity of RCEs do show some uncertainty among datasets, where values based on the NCEP/NCAR and NCEP-DOE datasets are greater than those based on the JRA55 and ERA5 datasets by 20-30% (Fig. 3b, d).

Decadal trend
We show the spatial pattern of the trend in SSD, which measures the overall decadal change in synoptic temperature variability (Fig. 4). It is found that the SSD decreases significantly in most of North America by 40°N in all datasets ( Fig. 4a-d), though it shows the weaker and ambiguous trend over Eurasia. Increased SSD is seen in a fraction of the continental grid points over Eurasia within 30-60°N based on the NCEP/NCAR dataset (Fig. 4a). Averaged among all Fig. 4 The decadal trend of SSD of temperature (unit: K/decade) based on the four datasets (NCEP/NCAR, NCEP-DOE, JRA55, and ERA5). Data on the ocean has been masked. The trend of grid point statistically significant at the 90% level is stippled the continental grid points within 50-70°N, the trend of SSD ranges from − 0.09 to − 0.21 K/decade over the mid-high latitudes in the four datasets; all of which are statistically significant at the 99% confidence level (Fig. 5b). However, the trend is weak and insignificant for the averages among the continental grid points within 70-90°N (Fig. 5a), 30-50°N (Fig. 5c), and 10-30°N (Fig. 5d). In general, the synoptic temperature variability decreases significantly in winter over the land areas of mid-high latitude, though some uncertainty exists among datasets (Duan et al. 2013;Screen 2014).
We also calculate the frequency, mean intensity, and extreme intensity of RCEs (see Section 2) for each winter and show the decadal trends in Figs. 6, 7, and 8. As consistently suggested by the four reanalysis datasets, most areas in the mid-high latitudes show a decreasing trend in the frequency of RCEs (Fig. 6), particularly over the eastern North America (ENA) region and western Eurasian (WE) region, whereas the increasing trend in the frequency of RCEs is seen over western North America region and North Africa. Unlike the pattern of SSD, the decreasing trend of RCEs almost covers the whole ENA and WE, and the signal is relatively robust. The spatially averaged trends over WE and ENA regions (the domains within the two boxes in Fig. 4a) are − 0.59 and − 0.31 times/decade based on the NCEP/ NCAR dataset (Fig. 4a) and range from − 0.56 to − 0.86 (− 0.62 to -0.71) times/decade over ENA (WE) region in the other three datasets (Fig. 6b-d).
In addition to the reduced frequency, the mean and extreme intensity of RCEs also show an overall negative trend, as seen in Figs. 7 and 8. For both the mean and extreme intensity of the RCEs, the number of grid points with the negative trend is much higher than the number of grid points with the positive trend ( Figs. 7 and 8). In Fig. 7, a slight but robust decreasing trend is seen in the mean intensity of RCEs over ENA region, which reaches about − 0.15 K/decade averaged among four datasets, with  70-90°N, b 50-70°N, c 30-50°N, and d 10-3 0°N. The * and ** mean the trends are significant at the 95 and 99% confidence levels high consistency ranging from − 0.14 to − 0.19 K/decade (− 0.16 K/decade in NCEP/NCAR, − 0.19/decade in NCEP-DOE, − 0.14/decade in ERA5, and − 0.16 K/decade in JRA55). However, there exists no significant trend over all of Eurasia except in the NCEP-DOE dataset (Fig. 7b). The spatially averaged trend for the mean intensity of RCEs reaches − 0.13 K/decade over WE region in the NCEP-DOE dataset, which is about twice as much as the trend in the other three datasets. The trend of the extreme intensity of RCEs, as shown in Fig. 8, shares a pattern similar to that of the trend of mean intensity. The regional averaged trend of the extreme intensity ranges from − 0.56 to − 0.70 (− 0.25 to − 0.50) K/decade over ENA (WE) region.
The regional mean trends over the ENA and WE regions for the frequency, mean, and extreme intensity are shown in Fig. 9. Multiple metrics consistently show fewer and weaker RCEs for both regions. The frequency of RCEs decreases at a rate of − 0.66 (− 0.58) times/decade over ENA (WE) region averaged among four datasets (Fig. 9a). The trend of the mean intensity is − 0.17 (− 0.091) K/decade over the ENA (WE) region (Fig. 9b), and the trend of the extreme intensity is − 0.62 (− 0.30) K/decade over the ENA (WE) region (Fig. 9c). The frequency and intensity of RCEs both decrease over the ENA and WE region in recent 4 decades, although the trend of mean intensity of RCEs over the WE region does not pass the 95% confidence level. The decreases in RCE frequency and intensity over ENA are all statistically significant at the 99% confidence level, suggesting that fewer and weaker RCEs occurred in winter during the past 4 decades. The decreasing trend of RCEs frequency over WE region is also statistically significant at the 99% confidence level.

Causes for the decadal trend
Mid-latitude weathers are modulated by frontal activities, which are controlled by the fast-moving Rossby waves at the mid-upper troposphere, i.e., transient eddies (Eady 1949;Hovmöller 1949;Charney 1990). The Rossby waves travel along the westerly jet and lead to the synoptic cyclone (anticyclone) near the surface, generating cold advection to the east (west) of the surface anti-cyclones (cyclones) (Hoskins et al. 1985;Hoskins and Ambrizzi 1993). To investigate the change in the energy of transient eddies and its possible Fig. 6 As in Fig. 4 but for the trend of RCE frequency (unit: times/ decade). The two boxes in a are two domains for eastern North American continent (ENA, 45-60°N, 50-100°W) and western Eurasian continent (WE, 45-70°N, 20-90°E). The regional averaged trends for these two domains (unit: times/decade) are marked on the top of each panel. The * and ** indicate that the spatial averaged trend pass the statistically significant test at the 95 and 99% levels Z. Cui, C. He effect on the RCEs, the EKE for each winter is calculated, and the decadal trend of EKE is shown in Fig. 10. In Fig. 10, the negative trend of EKE is found over a large part of the mid-high latitudes, especially in the North Atlantic Ocean and northwestern Eurasia, while a band-shaped positive trend occupies the continent in the middle latitudes near 40°N. All four datasets show a similar pattern of EKE trend (Fig. 10a-d), which confirms a robust weakening trend of EKE ranging from − 1.16 to − 1.83 m 2 s −2 decade −1 in the continental region north of 50°N. The zonally averaged EKE shows a positive trend over 35-50°N and negative trends on its northern and southern flanks. The negative zonal mean trend reaches − 3.23 (− 5.97) m 2 s −2 decade −1 at 60°N (20°N). Based on the average among the four datasets, the positive trend reaches 3.42 m 2 s −2 decade −1 within 35-50°N.
This pattern for the EKE trend is highly consistent with the pattern for the trend of RCE frequency (Fig. 6), with a spatial correlation coefficient of 0.30 (significant at 99% confidence level) averaged over the four datasets. It seems that the activity of transient eddies has weakened over the mid-high latitudes in recent decades, which may decrease the synoptic temperature variability of temperature and reduce the occurrence of RCEs.
To further explore the relationship between the decadal change in RCE frequency and EKE, Fig. 11 shows the values of the trend in RCEs frequency as a function of the trend in EKE for all continental grid points within 50-70°N (gray markers) and 10-30°N (red markers). The decadal trends of RCE frequency have a positive linear correlation with the change in EKE over the mid-high latitudes (black fitting lines), and the correlation coefficient ranges from 0.18 to Fig. 9 The time series and the associated trend of spatial averaged RCE frequency (a unit: times/decade), mean, and extreme intensity (b, c unit: K/decade) over eastern North American continent (ENA, red crosses) and western Eurasian continent (WE, blue triangles). The domains for the ENA and WE are shown as boxes in Fig. 6a. The dashed lines in (a-c) indicate the averaged trend among the four datasets. The * and ** of labels indicate that the averaged trends pass the statistically significant test at the 95 and 99% confidence levels 0.40 among the four datasets, with an average of 0.32; all of which are statistically significant at the 99% confidence level (Fig. 11). The above relationship only exists in the mid-high latitudes but not over the low latitudes (pink fitting lines). This shows that the spatial pattern for the EKE trend well explains the spatial pattern for the trend of RCE frequency in the mid-high latitudes, and it suggests that the weakening of the transient eddy activity may be responsible for the decrease in the occurrence of RCEs.

Summary and discussion
In this study, the decadal trend of synoptic temperature variability is evaluated based on four reanalysis datasets covering 40 winters with a common period from 1980 to 2019. Synoptic temperature variability is measured in terms of the synoptic-scale standard deviation of temperature (SSD), as well as the frequency and intensity of the synoptic-scale decrease in daily mean temperature (i.e., RCEs). To unravel the possible cause for the trend of synoptic temperature variability, EKE is calculated to explore the relationship between the intensity of weather activity with the synoptic temperature variability.
The SSD and RCEs show a consistent pattern of climatology for synoptic temperature variability. They both capture the much greater climatological temperature variability over eastern North America and inland Eurasia and the relatively weak climatological temperature variability over the tropics and subtropics. During the past decades, SSD decreased significantly in most of North America but shows relatively weaker trend over Eurasia. The zonal averaged SSD decreased by about − 0.09 to − 0.21 K/decade over mid to high latitudes among the four datasets. As a manifestation of narrowed synoptic temperature variability, the RCEs are becoming less frequent and weaker in recent decades, particularly over the eastern North America and western Eurasia. Over these two regions, the annual number of RCEs in winter decreases by about − 0.6 times/decade, and the extreme intensity of the RCEs decreases by about − 0.6 K/ decade over eastern North America. Our results are consistent with previous studies that suggested a reduction of subseasonal temperature variability over the northern high-latitudes in autumn and over North America in both winter and summer (Screen 2014;Schneider et al. 2015;Rhines et al. 2017;Herring et al. 2018), in contrast to increased variability over tropic and subtropic regions (Santer et al. 2005). The observational trend in past decades is also consistent with model results, which suggested a robust decreasing trend of subseasonal temperature variability over mid-high latitudes under future emission scenarios (Barnes and Screen. 2015;Bathiany et al. 2018;Blackport et al. 2021).
Most previous works focused on cold extremes in winter, and they used metrics like T min or cold days (Ryoo et al. 2005;Park et al. 2011). These indices are strongly controlled by the mean state of temperature (Alexander et al. 2006).

Fig. 10
The spatial pattern of the trend for EKE among the four datasets in winter (unit: m 2 s −2 /decade). Stippling indicates the grids where the trends are significant at the 90% confidence level. The meridional profile for the zonal averaged trend of EKE is shown on the right of each panel Global warming may simply lead to more hot spells in summer and fewer extremely low temperature events in winter, even if the synoptic-scale temperature variability remains unchanged (Schar et al. 2004;Ryoo et al. 2005;Park et al. 2011). The metric adopted in this study focuses on synopticscale decrease in temperature rather than cold extremes, and it shows that the rapid cooling events in winter also have reduced during the past decades.
In winter, the synoptic-scale temperature variability is closely associated with the transient eddies, which are also referred to as fast-moving Rossby waves propagating eastward with a phase speed of ~ 6-12 m/s (Petoukhov et al. 2013;Ali et al. 2021). Previous works suggested a weakening of the transient eddies manifested as reducing EKE over the Eurasian continent in summer during recent decades, which brought more heat waves (Coumou et al. 2015). In this work, we identified that the EKE also shows a significant decreasing trend in winter, suggesting an overall weakening of the synoptic-scale transient eddy activities across seasons in recent 4 decades. The pattern of trend in EKE shows a pattern similar to that of RCE frequency, which indicates that the weakening of transient eddy activities may be responsible for the trend toward fewer RCEs in winter in recent decades.
Author contribution Zhenyuan Cui performed all the data analysis and wrote the initial draft of this paper. Chao He designed the study and improved the quality of writing. All authors read and approved the final version of the manuscript.
Funding This work was supported by National Natural Science Foundation of China (41875081) and the Fundamental Research Funds for the Central Universities (11621049).

Data availability
In this study, four reanalysis datasets were obtained and as follows: The NCEP/NCAR and NCEP-DOE datasets are Fig. 11 Scatter diagram for the trends in RCE frequency as a function of the trends in the EKE for all the continental gird points over the mid-high latitudes (50-70°N, black circles) and mid-low latitudes (10-30°N, pink triangles). The leastsquares fitting line is shown as the black (pink) dashed line in each panel. The correlation coefficients are marked on the upper right of each panel, and the ** indicates that the correlation coefficient is significant at the 99% confidence level download from https:// psl. noaa. gov/ data/ gridd ed/ data. ncep. reana lysis. html and https:// psl. noaa. gov/ data/ gridd ed/ data. ncep. reana lysis2. html. TheJRA55 dataset is obtained from https:// rda. ucar. edu/ datas ets/ ds628.0/. The ERA5 dataset is downloaded from https:// cds. clima te. coper nicus. eu/ cdsapp# !/ datas et/ reana lysis-era5-single-levels? tab= eqc.
Code availability NCL codes used in this study are available from the corresponding author Zhenyuan Cui upon reasonable request.

Declarations
Ethics approval This work complied with all the relevant ethical approval processes.

Competing interests
The authors declare no competing interests.