Detection of the synoptic southeastward-extending Siberian cold high during 1978–2017

Yumin Zhang, Chuhan Lu

Key Laboratory of Meteorological Disaster, Ministry of Education, Nanjing University of Information Science and Technology, Nanjing, China

Keywords:Cold high Anticyclone Anticyclone identification Cold air

ABSTRACT To better understand the relationship between anticyclones in Siberia and cold-air activities and temperature changes in East Asia, this study proposes a 2D anticyclone identification method based on a deep-learning model,Mask R-CNN, which can reliably detect the changes in the morphological characteristics of anticyclones.Using the new method, the authors identified the southeastward-extending Siberian cold high (SEESCH), which greatly affects wintertime temperatures in China.This type of cold high is one of the main synoptic systems (45.7%)emerging from Siberia in winter.Cold air carried by SEESCH has a significant negative correlation with the temperature changes in the downstream area, and 52% of SEESCHs are accompanied by cold-air accumulation in North and East China, which has a significant impact on regional cooling.These results provide clues for studying the interconnection between SEESCHs and extreme cold events.

1.Introduction

Winter anticyclone activities in Eurasia are closely related to low-level cold-air activities ( Ioannidou and Yau, 2008 ).Surface cold high/anticyclone activities are often accompanied by large-scale coldair outbreaks, causing disasters such as local blizzards, freezing rain,and strong winds.The Siberian high is an important circulation system that affects weather changes in East Asia in winter.Previous studies have indicated that a significant temperature decrease may occur in East China when the Siberian high strengthens and expands to the southeast ( Lan and Li, 2016 ; Zhu et al., 2019 ).It is necessary to identify anticyclones efficiently and objectively to analyze their activities,as synoptic cold high/anticyclone activity is an important component of the Siberian high and its activity changes are closely related to the Siberian high ( Zhang et al., 2012 ).

Previous studies have proposed many objective identification methods using different variables, such as sea level pressure (SLP), pressure, geopotential height, vorticity, and their gradients, to identify cyclones/anticyclones ( Hanley and Caballero, 2012; Hoskins and Hodges,2002; Lim and Simmonds, 2002; Murray and Simmonds, 1991; Qin and Lu, 2017 ).These methods mainly identify system locations and paths after determining center locations ( Wang et al., 2009 ; Zhang et al., 2012 ;Zhi et al., 2019 ).Most methods are efficient but have limitations in identifying the regime of an individual system and the division or merger of a multicenter system.To enrich the characteristics of identified systems, 2D methods that detect system boundaries and spatial domains have been proposed in previous studies ( Wernli et al., 2006 ; Hanley and Caballero, 2012 ; Lu, 2017 ).In addition, Jiang et al.(2020) proposed a cyclone identification method based on the mean SLP/potential height, which can reduce the misidentification of troughs as cyclones.Fu (2020) suggested the use of restricted vorticity as a new metric for numerical mesoscale vortex identification, which had a high hit rate compared with that of manual analysis.However, in comparison to cyclone identification methods, anticyclone identification methods and their improvements are minimal.

said the Great, Huge Bear, in his great, rough, gruff voice. And when the Middle Bear looked at his, he saw that the spoon was standing in it, too. They were wooden spoons; if they had been silver ones, Goldilocks would have put them in her pocket.18

It is difficult to identify anticyclone structures owing to their complex shapes and weaker internal pressure gradients than those of cyclones.In addition, comparatively few studies have focused on the 2D activity characteristics of anticyclones ( Hoskins and Hodges, 2002 ; Bardin and Polonsky, 2005 ; Wernli and Schwierz, 2006 ).As the cold-core area in winter, Siberia is the most frequent activity area of winter cold highs in Eurasia.Ding (1990) indicated that the intensity of the Siberian high is closely related to cold surges.However, previous studies have mainly focused on the interannual and interdecadal variations of the seasonal Siberian high and its impacts on winter temperature in East Asia ( Lan and Li, 2016 ; Zhu et al., 2019 ; Liu and Zhu, 2020 ), with few having analyzed the impact of synoptic cold highs.The anticyclonic system in Siberia often appears as a strong cold high with multiple centers and asymmetric shapes.Under the influence of upper-level Rossby waves and low-level cold advection ( Takaya and Nakamura, 2005 ), anticyclones in Siberia often tend to intensify and stretch to the southeast, causing cold surges in North and East China ( Sun et al., 2010 ),while their maximum-value centers are almost all still over the Siberian–Mongolian Plateau.In this paper, we refer to this type of cold high as the southeastward-extending Siberian cold high (SEESCH).To better detect the extending feature of SEESCH, a reliable description of the shape of SEESCH is required.Therefore, this paper proposes an anticyclone identification method based on a deep-learning model, Mask R-CNN( He et al., 2017 ; Lu et al., 2020 ), to detect the 2D shape characteristics of anticyclones.

Fig.1.Southeastward-extending Siberian cold high.Color shading represents the SLP (units: hPa).The green frame covers (40°–60°N, 70°–120°E), which is the traditional region of activity for defining the Siberian high.The blue frame covers (30°–50°N, 100°–120°E), which is the downstream area of the Mongolian Plateau and mainly includes the Inner Mongolian Plateau, the Loess Plateau, and the North China Plain.

2.Data and methods

2.1. Data

The datasets used in this paper are the 6-hourly, 2.5° × 2.5° SLP, surface pressure, temperature, surface air temperature, and 850-hPa winds from ERA-Interim ( Dee et al.2011 ) for 40 winters from 1978 to 2017(the winter data of 1978 use data from January and February 1979).The anticyclone identification data are based on the 6-hourly SLP field.To improve the identification accuracy, the resolution of the identification data is 0.7° × 0.7°.

2.2. Methods

2.2.1.DefinitionofSEESCH

Every Sunday Jurgen went to church; and when his eyes rested onthe picture of the Virgin Mary over the altar as he sat there, theyoften glided away to the spot where they had knelt side by side

Through the above analysis, we have determined that during the early enhancement period of a SEESCH event, its scope of influence gradually expands to the south with accompanying southward movement of cold air.Thus, the CAM carried by cold highs increases in the downstream area, causing the regional temperature to drop.When a SEESCH event is stable and declining, although its central point is still over the Mongolian Plateau and the domain of influence covers the downstream area, the corresponding CAM does not increase significantly.At the same time, the weather conditions in the relevant area remain stable under the control of the high-pressure system, and a weak temperature increase may even appear.In previous studies, it has proved difficult to select SEESCH events and their corresponding periods of cold-air enhancement for focused analysis.However, by using our new anticyclone objective identification method based on Mask R-CNN, it is possible to reliably mark the 2D domain of influence of cold highs and then directly calculate the corresponding change in cold air to select cold highs and their corresponding periods of enhancement that have a substantial impact on regional weather, which is helpful for further analysis of the connection between cold-high activity and extreme cold weather.

The cards I bought, Topps, still included a piece of gum, unlike many of the newer brands. The smell of the gum, the cardboard, the ink, and the wax paper brought the baseball memories of my youth rushing back as surely as did that first whiff of mown grass in the spring. I flipped2 through the cards, recognized a few of the players names, and then put them away.

I have promised the Beast faithfully that I will come back, and he would die of grief if I did not keep my word! What would that matter to you? said the Prince Surely you would not care? Indeed, I should be ungrateful if I did not care for such a kind Beast, cried Beauty indignantly

Several years later, young Tom was rummaging1 around in the garage as only a five- or six-year-old can rummage2 when he came across the all-leather, NFL regulation, 1963 Chicago Bears-inscribed football. He asked if he could play with it. With as much logic3 as I felt he could understand, I explained to him that he was still a bit too young to play carefully with such a special ball. We had the same conversation several more times in the next few months, and soon the requests faded away.

2.2.2.NewanticycloneidentificationmethodbasedonMaskR-CNN

(1) The horizontal domain of the surface cold highs affecting the Mongolian Plateau region is manually analyzed based on the 6-hourly SLP field between 2008 and 2012, and 5-year winter 2D anticyclone domain data (label data) are obtained.The details of the manual analysis are given in the Appendix.

(2) Five-year winter SLP images and the subjective label data in step 1 are used as the training dataset.The training dataset is input into the Mask R-CNN model, and the corresponding model parameters and hyperparameters are set ( Table 1 ) for training.The model is trained 10 000 times.A series of network layers, such as CNNs, RPN, and RolAlign, are used to filter and determine the positions of targets in input images, and then the range of targets are marked through the mask branch of the model.Finally, the result of the target identification range is obtained.

Table 1 Parameters and hyperparameters of Mask R-CNN.

(3) The 40-year winter SLP images (from 1978/79 to 2017/18) are input into the trained Mask R-CNN model, and the machine-learning mask data are obtained as the output.These mask data are the 2D cold high identification results of Mask R-CNN.

40.Went to a corner of the wood, and there died: While it may be harsh to call her evil, the eldest sister is definitely rude and antagonistic76. Antagonists77 usually suffer just deaths at the end of fairy tales. The eldest sister s antisocial behavior causes her to die miserable and alone in the woods.Return to place in story.

2.2.3.Coldairmass

Each time a SEESCH appeared in the 40 winters from 1978/79–2017/18, it was statistically significant, and the results are shown in Table 3 .SEESCH events essentially appeared approximately 100–200 times per year, at an average of 165 times per year, which amounts to approximately 45.7% of all events in the winter of one year, showing that SEESCH events appear frequently in Eurasia in winter.To further investigate the connection with regional cooling events in China, the CAM carried by each SEESCH in North and East China (NEC) was quantitatively calculated.Furthermore a SEESCH that carried a 24-h increase in CAM in NEC was defined as an intensified SEESCH (I-SEESCH).Statistically, I-SEESCH events accounted for 52% of SEESCH events on average,and 74.1% of I-SEESCH events led to a regional decrease in temperature in NEC, indicating that a considerable proportion of SEESCH events can affect the changes in weather in downstream areas induced by intensified cold-air activities, leading to cooling weather in North China.

A deep-learning model based on Mask R-CNN ( He et al., 2017 ;Lu et al., 2020 ) is proposed to objectively detect the 2D domains of influence of anticyclones.The process of the new objective anticyclone identification method is shown in Fig.2 and described as follows:

Fig.2.The process of the new objective anticyclone identification method based on Mask R-CNN.“Subjective mask ” is the manually analyzed anticyclone identification result; “Machine learning mask ” is the output anticyclone identification result of the model; and the structure of the Mask R-CNN model is given in brief in the upper dashed box.

2.2.4.Cold-airoutbreak

To demonstrate that the cold air carried by a SEESCH event is closely related to the temperature of the downstream area, the number of ISEESCH events causing a regional drop in temperature in NEC in the winters from 1978/79 to 2017/18 was counted (2210 events in total).As shown in Fig.5 , the correlation coefficient between the increase in the CAM carried by intensified non-SEESCH events (I-non-SEESCH) and the decrease in regional temperature in NEC is just ? 0.21 ( Fig.5 (a)),which is lower than the one with SEESCH ( ? 0.39).The closer connection between SEESCH events and the change in regional temperature in NEC is consistent with the higher frequency of influential grids over induced by SEESCH with more accompanied CAM ( Fig.3 ).When CAMs carried by I-SEESCH events increase in the area of influence,the regional average temperature decreases significantly.The correlation coefficient between them is ? 0.39, and the regression coefficient is ? 0.014 ± 0.0006 K·hPa?1( Fig.5 (b)).After seasonal averaging, the correlation coefficient of the two is ? 0.68 ( Fig.5 (c)), at the 0.01 significance level based on the Student’st-test, indicating that cold-high activities can directly affect the temperature changes in an area of influence through cold air.Furthermore, the correlation coefficient between the CAM carried by SEESCH events and the duration of the cold-air outbreaks is 0.37 ( Fig.5 (d)), which is notably higher than that with non-SEESCH events (0.03).Therefore, SEESCH events could lead to more severe weather in winter in China compared to non-SEESCH events.

3.Results

3.1. Identification of cold high

To examine the ability of Mask R-CNN to identify cold highs, we randomly selected a SEESCH event that took place at 1800 UTC 16 February 1996 to show the identified 2D domain of the system ( Fig.3 (a)).A strong cold high in the SLP field in Fig.1 has the characteristic of a southeastern extension and is classified as a SEESCH, and its outer shape is identified by our new identification algorithm ( Fig.3 (a)).The domain of influence of the cold high identification result is consistent with the actual pressure distribution, and the outer shape essentially corresponds to the outer contours of the pressure distribution.The characteristic of this cold high extending to northern China is also shown in the result,which shows that the new identification method reliably identified this event.

In addition, the Siberian cold high can be divided into non-SEESCH and SEESCH events through the definition of SEESCH.The proportion of SEESCH is notably higher than non-SEESCH (Fig.3(b, c)).However,about 22.4% of Siberian highs still belong to the non-SEESCH category.As shown in Fig.3 (b), the influential areas of non-SEESCH are mainly concentrated in the core area of the classic Siberian cold high.In contrast, a pronounced southeastern extension, including the north and east parts of China, emerges in the 2D frequency of SEESCH ( Fig.3 (c)).This notable contrast suggests more CAMs resulting from SEESCH events could invade the north and east parts of China.

3.2. Synoptic analysis of SEESCH evolution

To demonstrate the ability of our method to recognize SEESCH consecutively, the evolution of SEESCH and the accompanying weather conditions during 16–19 February 1996 is shown in Fig.4 .During this process, a strong cold high system existed over the Mongolian Plateau.The center of the system was stabilized in the Sayan Mountains, and its domain of influence (the area of green dots) gradually expanded to the south over time.Affected by the expanding cold high, the SLP values in downstream East Asia increased gradually, and the East Asian monsoon was significantly enhanced.The cold high and cold air carried by its southeastern front edge moved southwards under the influence of strong northerly winds at the periphery, causing dramatic cooling over North China, the Jianghuai region, and South China, successively.In particular, there was strong cooling within 24 hours in Jiangnan and southern China on 16 and 17 February, with a local drop in temperature of 6–10 K.Finally, the cold air passed, East China was under relatively stable weather conditions under the control of the cold high, and then,a slight temperature increase followed in North China at 0000 UTC 19 February ( Fig.4 (d)).Therefore, our method offers a natural description of the evolution of the regimes of this SEESCH and its impact on regional cooling during its southeastern extension process.

Fig.3.(a) Machine-learning mask data of a cold high.The green shading is the identified cold high range, and the red dot is the maximum SLP value in the cold high.(b, c) Average 2D frequency distribution of the Siberian high in 40 winters from 1978 to 2017 for (b) non-SEESCH and (c) SEESCH events.

3.3. Multi-year identification ability

To further assess the reliability of our method, the number of SEESCH events identified by the model was compared with those identified by the manual analysis results (details of the manual analysis are given in the Appendix), and their matching rate is summarized in Table 2 .In the winters from 2008 to 2012, Mask R-CNN identified 97.3% of the manually analyzed SEESCH events on average, and the average matching rate of the five years was 97.1%.These data show that the machine-learning results can identify most of the manually recognized cold highs.In addition, the model has a high consistency in terms of shape descriptions compared with the those of the manually analyzed systems.

Fig.4.A case of SEESCH affecting the downstream area.Color shading is the 24-h temperature change (units: K); the isobars denote the SLP (units: hPa); the vector arrows are the 850-hPa wind; the area of green dots is the cold-high range recognized by Mask R-CNN; and the red dot is the maximum SLP point in the cold high.

Table 2 Comparison of the number of SEESCH events identified by Mask R-CNN and from manual analysis in winters from 2008 to 2012.The column labeled“Both ” represents the number of events that both machine-learning and manual analysis could identify.The identification rate is the ratio of “Both ” to “Manual ”.The “Same ” column represents the number of events that identification range results of machine-learning and manual analysis are consistent, and the matching rate is the ratio of “Same ” to “Manual ”.

3.4. Relationship between SEESCH and regional temperature anomalies

To quantify the cold-air activity accompanied by cold highs, the cold air mass (CAM) defined by Iwasaki et al.(2014) is used in this study to quantitatively calculate the cold air carried by each anticyclone.The definition of CAM is the difference between the surface pressure and the pressure at the 280-K isothermal surface ( Iwasaki et al., 2014 ).The calculation formula is as follows:

Table 3 Statistics of winter SEESCH events affecting the downstream area from 1978 to 2017.The column labeled “SEESCH ” represents the number of times SEESCH events appeared each year; the column labeled “I-SEESCH ” shows the number of times SEESCH events that carried a 24-h increase in CAM in NEC occurred; P1 is the ratio of I-SEESCH to SEESCH; and P2 is the probability of I-SEESCH causing the 24-h decrease in temperature in the downstream area.

Stable and strong cold highs that may have weather impacts on the surrounding regions and the downstream areas are often formed over the Mongolian Plateau of Eurasia in winter.As shown in Fig.1 , a significant anticyclone exists in Siberia.Although the central area of this system is located west of Lake Baikal, an obvious southeastern extension feature is displayed in the downstream areas of Mongolia that could bring a notable decrease in temperature over North China.We defined this type of cold high as SEESCH.In particular, the system center of SEESCH should be located over the Mongolian Plateau (green frame;40°–60°N, 70°–120°E), and its domain of influence accounts for more than 25% of the southeastern stretch area (blue box; 30°–50°N, 100°–120°E).

The definition of a cold-air outbreak event refers to the work of Zhang et al.(2018) .When the area-averaged daily temperature in eastern China (blue box in Fig.1 ; 30°–50°N, 100°–120°E) is lower than the December–January–February mean temperature by one standard deviation for two days or more consecutive days, it is regarded as a cold-air outbreak event.The standard deviation is the average of the daily standard deviation of area-averaged daily temperature in eastern China in 40 winters.

I fell in love with the minister s son the winter I turned fourteen. He was not Chinese, but as white as Mary in the manger. For Christmas I prayed for this blond-haired boy, Robert, and a slim new American nose. When I found out that my parents had invited the minister s family over for Christmas Eve dinner, I cried. What would Robert think of our shabby Chinese Christmas? What would he think of our noisy Chinese relatives who lacked proper American manners? What terrible disappointment would he feel upon seeing not a roasted turkey and sweet potatoes but Chinese food?

Fig.5.Regression of the increase of CAMs carried by intensified Siberian cold highs in NEC (units: hPa) and the decrease in regional temperature (units: K) during 1978/79–2017/18: (a) I-non-SEESCH; (b) I-SEESCH; (c) seasonal average of I-SEESCH; (d) regression of the CAM carried by SEESCH events in 40 winters from 1978 to 2017 (units: 10 6 hPa) and the duration of the cold-air outbreak (units: d).

4.Conclusions

This paper proposes a 2D anticyclone identification algorithm based on a deep-learning model, Mask R-CNN.We applied this method to the detection of SEESCH activity affecting the downstream area, especially in NEC.

Through case studies of SEESCH events, we found that the new identification method performs well in identifying the synoptic features of the southeastward expansion of the Siberian cold high.This type of cold high transports cold air from high latitudes to the south and is notable in downstream East Asia, especially in NEC.Furthermore, the statistics of all SEESCH events identified in the winters from 1978/79 to 2017/18 showed that SEESCH is one of the main active systems (45.7%) emerging from Siberia and occurs frequently in Eurasia in winter.Fifty-two percent of SEESCH events are accompanied by cold-air accumulation in NEC.The activity of cold air carried by these systems has a significant negative correlation with temperature changes in the downstream areas.When the cold high intensifies, the CAMs carried by these systems will increase in the downstream areas and significantly influence regional cooling.Compared to non-SEESCH events, SEESCH events could lead to more severe weather in winter in China.

Beautiful flowers, round which little bells were tied, stood in the corridors: what with the running to and fro and the draught, these bells tinkled so loudly that no one could speak to be heard

The results of this analysis may shed light on anticyclonic behavior in Siberia via the newly detected 2D cold high atlas derived from this deep-learning model, which could be beneficial to further studying the relationship between cold highs and extreme cold events.

Funding

This work was supported jointly by the National Key Research and Development Program of China [grant number 2019YFC1510201 ]and the National Natural Science Foundation of China [grant number 41975073 ].

Acknowledgments

The authors would like to thank the Center of Atmospheric Data Service, Nanjing University of Information Science & Technology, under the Geoscience Department of the National Natural Science Foundation of China, and the European Centre for Medium-Range Weather Forecasts for providing data.

Appendix

Manual analysis procedure for cold highs:

1) Check SLP images and search closed high-pressure systems that exist near the Mongolian Plateau (40°–55°N, 80°–120°E) or its downstream area in East Asia (20°–55°N, 100°–140°E).When multiple closed subsystems exist in a closed high-pressure system, they are regarded as a unified whole.

I kept closing the gap between my car and my brother s truck, inching toward the reassuring13 glow of the taillights like a frightened sheep following a shepherd. I could see Mac watching me in his rearview mirror , and that night at supper in Kingman, Arizona, he said, Leigh, you re tailgating . You re much too close for these mountain roads. He studied my face for a moment, then added, Tomorrow will be the last day of high country. Just try to hang in there. We ve got this far okay. You know there s nothing to be afraid of. ，，，，。。，：“，。!”，：“。，，。。”

2) Record the location of the center maximum SLP of the high-pressure system.Then, the grids within the system’s outermost enclosed boundary are labeled as influential grids by this system.The specific peripheral SLP value of a single high-pressure system is determined according to the distribution of SLP at that time.

3) Finally, the following characteristics can be obtained: the SLP value of the outermost boundary of the high-pressure system, the position information of all grids within the boundary of the high-pressure system (range of influence), the SLP value, and the latitude and longitude information of the center point.