Extratropical cyclones(ECs)are baroclinic lowpressurevortices over mid and high latitudes of thetwo hemispheres. They play an important role in theatmospheric general circulation. The transfer of heat, moisture, and kinetic energy between the tropics and polar regions largely relies on the movement and developmentof ECs. In addition, ECs are importantsynoptic systems that affect weather variability over awide area of mid and high latitudes. Driven by baroclinicinstability, ECs with frontal systems may causesignificantly intense weathers(Ding, 2005). ECs oftenoccur in the strong baroclinic zones and result inthe enhancement of synoptic scale disturbances. Vorticity and thermal advection often trigger the initialdevelopment of ECs. As the rainfall increases, the effectof condensational heating becomes more and moreimportant, finally leading to explosive development ofcyclones(Ding and Zhu, 1994). On the other h and , the explosive development of ECs is conducive to occurrencesof s and storms over northern China(Liu et al., 2003; Zhao and Zhao, 2004a, b; Tu et al., 2007), and there is also a close relationship between blizzards and explosively-deepening cyclones in NortheastChina in winter and spring(Cai et al., 2009).

In the early times, detecting and tracking ECs reliedon time-consuming manual synoptic analysis(Reitan, 1974; Zishka and Smith, 1980; Zhang, 1984; Zhu et al., 2000). Nowadays, cyclone detection has beendigitalized, which eliminates the artificial r and om errorsin the manual analysis. Since the mid 1980s, objectivealgorithms based on sea level pressure, geopotentialheight, or vorticity fields have been developed and applied to detecting and tracking ECs(Lambert, 1988; Alpert et al., 1990; Murray and Simmonds, 1991a, b; König et al., 1993; Hodges, 1994; Serreze, 1995; Haak and Ulbrich, 1996; Blender et al., 1997;Sinclair, 1997; Lionello et al., 2002; Pinto et al., 2005;Mendes et al., 2010). In most cases, cyclone cores aredefined in terms of local sea level pressure minima, orlocal minima of 1000-hPa geopotential height(Lambert, 1988). Alternatively, cyclones can be defined interms of local vorticity maxima at low levels(König et al., 1993). The latter approach tends to identify moresmall scale systems with strong positive vorticity thatmainly emerge in the fast moving cyclone processes inthe early or late stages of their life cycles(Murray and Simmonds, 1991a, b). However, these small scale cyclonesoften do not have minimum pressure. Trackingof ECs is usually performed by assigning an individualsystem identified at one particular date to a successorat the subsequent date, and the assignment ofthe most likely successor involves a search in a certainarea(Pinto et al., 2005). Alpert et al.(1990), König al.(1993), and Blender et al.(1997) searched thelikely successor position over an area chosen as a circlearound the original position. Geng and Sugi(2001)chose the nearest identified system within a certain radiusas the successor position. Additionally, Murray and Simmonds(1991a, b) and Pinto et al.(2005)usedthe local steering flow to search the likely successor. Asa rule, cyclolysis occurs if no likely successor is found.

The activities of ECs are most frequent over severaloceans in the Northern Hemisphere. Serreze(1995)statistically analyzed the climatic aspects ofArctic cyclones during the period of 1973–1992 interms of cyclone sea level pressure tendency, maximumdeepening rate, and frequencies of cyclogenesis and cyclolysis. Serreze et al.(1997)also noted thatthere are some connections between the Arctic cycloneactivity and the North Atlantic Oscillation(NAO). Asfor Atlantic cyclones, Blender et al.(1997)analyzedAtlantic/European cyclone tracks from 1990 to 1994using a Lagrange method based on the high resolution1000-hPa geopotential height data. Cluster analysisyielded three types of cyclone tracks characterizingstationary, zonally travelling, and northeastwardmoving ECs. North Pacific is another main regionwhere cyclone activities are frequent. In the past 50years, the frequency and intensity of ECs over NorthPacific have increased markedly, with associated upwardtrends in extreme surface winds between 25° and 40°N(Graham and Diaz, 2001). Favre and Gershunov(2006)also found that ECs over northeasternPacific had been intensifying with their trajectoriesmore southward since the mid 1970s. Compared withother low pressure systems, ECs over the Mediterraneanare sub-synoptic scale systems(Trigo et al., 1999; Maheras et al., 2001), which have smaller spatialscale and shorter lifetime(i.e., 2–2.5 days on average).Bartholy et al.(2009)analyzed climatic characteristicsof western Mediterranean cyclones for a long period(1958–2002)based on reanalysis data, and examinedin particular their interannual and interdecadalvariations.

Objective analysis of ECs over East Asia(EA)didnot start until 2000. Yao et al.(2003) and Wang etal.(2007)investigated the frequency characteristics ofspring cyclones over northern EA in association withweather/climate in China using the objective method.Wang et al.(2009)showed that there is a dipole structurein the EC changes, with increasing frequency inthe north and decreasing in the south part of northernEA during the period 1958–2001. This characteristicis related to the northward movement of the baroclinicfrontal zone around 110°E. Based on the previousstudies of ECs in EA, this study uses an improvedalgorithm to detect and track ECs(see Section 2). TheLaplacian of pressure is considered as the EC intensitythreshold, which reserves more lows with strongpositive vorticity that often appears in the formativeor dying stages of ECs. Both steering flow and extrapolationof low systems are used to track ECs, and some constrains are prescribed to eliminate thermallows and tropical cyclones. In addition, we analyzeseasonal features(i.e., cyclogenesis, cyclolysis, systemintensity, movement, and explosive deepening)of ECsfor the purpose of forming a foundation for furtherstudies on the interannual and interdecadal variationsof ECs over EA.2. Data and methods

In this paper, the European Centre for MediumrangeWeather Forecasts(ECMWF)reanalysis data(ERA40), with a 2.5°×2.5° horizontal resolution and a 6-h time interval, are used. The objective algorithmfor detecting and tracking ECs is based on the methodof Murray and Simmonds(1991a, b), Simmonds et al.(1999), and Pinto et al.(2005), with some modificationsmade according to the features of ECs over EA.

Cyclones are identified objectively in the followingprocedure. First, gridded sea level pressure dataat 6-h intervals are scanned for local minima to obtainpossible cyclone cores, the pressure of which is lowerthan that of any of the eight grids around. Second, themean of Laplacian of pressure(∇²p(x_i, y_i)= p_xx+p_yy)around the cyclone center over a radius of 4 latitudedegrees is calculated, and the cyclones with ∇2p lessthan 0.4 hPa(deg.)^-2 are removed. Note that takingLaplacian of pressure as intensity threshold reservesmore cyclones with strong positive vorticity in cyclogenesis and cyclolysis stages than does taking pressureas intensity threshold. Third, when a cyclonesystem is strong enough, a family of cyclones is likelyto occur in the associated frontal zone, so we stipulatethat only the strongest one within five degrees is included, whose mean Laplacian of pressure is the maximum.According to Wu(1999), the Tibetan Plateau(20°–40°N, 60°–110°E)is such a tremendous topographythat cyclone seldom occurs over there. Thus, thestudy area in this paper covers 20°–75°N, 60°–160°E, excluding the Tibetan Plateau.

On the basis of identified cyclone systems at eachdate mentioned above, the next procedure is to trackeach cyclone process. First, predict the likely cycloneposition at the subsequent date r_pred(t + 1)by usingcyclone position at current date and the previous date,

Click Browse formula

Click Browse formula

The cyclone processes tracked above include cyclonesof all life cycles, while only ECs having a lifetimeof at least 24 h are studied. This considerationgets rid of local warm season thermal lows(Wang et al., 2009). Also, there are a few false paths in whichcyclones move upstream over five longitude degrees indistance during their lifespan(occupying about 5% ofall tracks). One possibility is that the algorithm mayfind the cyclone n upstream whose qmn is the maximumwhen searching likely successors over a certainradius; the other possibility is that some tropical cyclonesmoving westward from Northwest Pacific to theEA continent may be tacked. In the study of Zhu etal.(2000), EC movements are well correlated withthe airflow at 500- or 700-hPa levels, which is basicallywesterly in EA, so the spurious paths of cyclonesmoving upstream/westward over five longitude degreesare eliminated for the above reasons. Finally, cycloneprocesses are classified as motionless cyclones(movingwithin five longitude degrees in distance) and mobilecyclones(else). Motionless cyclones are removed wheneliminating some local thermal lows. After completingthe elimination step above, we output all elementsof each cyclone process, such as cyclogenesis date, cyclolysisdate, cyclone position, center pressure, cycloneintensity(the mean of pressure Laplacian over a radiusof four latitude degrees), and then establish thedatabase of ECs over EA for the period 1958–2001.3. Test of the objective identification and tracking algorithm3.1 Objective identification

To test the objective detection method, a comparisonbetween results of objective identification and surface weather charts of the Meteorological InformationComprehensive Analysis and Processing System(MICAPS)is performed. We give two examples toexamine the performance of the objective detectionmethod. Figure 1a shows some cyclone systems by themanual analysis at 0000 UTC 22 April 1976. We findtwo closed cyclones in Northeast China, one in downstreamareas of the Yellow River, two in NorthwestPacific, and one in Siberia. The algorithm has identifiedthese large and closed systems precisely, especiallythe two in Northeast China and the one in the lowerreaches of Yellow River(Fig. 1b). Thus, choosing thelocal pressure minima as the cyclone core is a feasiblemethod. The objective algorithm finds two cyclonesin Siberia(70°N, 90°–110°E)with the stronger one at70°N, 95°E(Fig. 1b). But there is only one systemidentified by the manual analysis at 70°N, 110°E. Thisis likely due to the differences between reanalysis data and observation data.

Fig. 1. Surface sea level pressure(hPa)charts of MICAPS(a, c) and ERA40(b, d)at 0000 UTC 22 April(a, b) and 1May 1976(c, d). The letters L and H in(a, c)denote locations of the low and high pressure systems identified by themanual analysis. The black solid dots in(b, d)denote the cyclone locations identified objectively.

Figure options

Another case is at 0000 UTC 1 May 1976. Wecan easily find one closed cyclone in the border areabetween Northeast China and Mongolia, one in JapanSea, one in the region to the north of Aral Sea, and oneat 70°N, 130°E by the manual analysis(Fig. 1c). InFig. 1d, all the aforementioned systems are identifiedsuccessfully. In most cases, these systems are mobile and can cause significant weather phenomena thereby.At the same time, we also find that the objective algorithmcan hardly identify some week systems withno associated fronts or no closed isobars, which can beidentified by the manual analysis.3.2 Tracking

By referring to historical weather charts publishedby the National Meteorological Center of the ChinaMeteorological Administration, cyclone processes havinga lifetime of at least 1 day in the area of 25°–65°N, 65°–135°E in 1965, 1971, 1986, and 1998 and trackedby the manual synoptic analysis are compared withthose tracked objectively. It is found that the numberof cyclone processes tracked by the objective algorithmis 10%–20% more than that tracked by themanual analysis. Although there are some differencesin the cyclogenesis and cyclolysis dates between objective and subjective results, the objective algorithmcan track more than 80% of real cyclone processes.

The following reasons may explain some of theabove differences. First, there are errors existing inthe statistics near the domain boundary. Second, themanual analysis is based on twice daily weather charts, but the objective method is based on four-time dailyreanalysis data, so cyclones of one-day lifespan trackedobjectively are very likely different from those in themanual weather charts. Third, tracking weak cyclonesof short lifetime differs in the two methods. Finally, tracking of cyclones on mutational routes for encounteringa large terrain or other reasons differs in the twomethods.

Some successful tracking cases are given here.The first case is a typical Mongolia cyclone(Fig. 2a), which moved northeastward, crossed the Sino-Russianborder, and finally reached the Okhotsk Sea, as revealedby the manual analysis. During its lifespan, cyclone pressure had little changes. This synoptic processis successfully tracked by the objective algorithm(Fig. 2b), with only small differences existing in theposition and time of cyclogenesis and cyclolysis betweenobjective and subjective methods. The secondcase(Fig. 2c)is a cyclone that formed in the middle and lower reaches of Yellow River, which moved northeastward, crossed North Korea, Japan Sea, Japan, and finally died out in Northwest Pacific. In this synopticprocess, the cyclone strengthened and its centralpressure lowered after it reached the Japan Sea. Thecyclone trajectory tracked objectively(Fig. 2d)beginsearlier than that tracked subjectively, and its formationposition identified objectively is in the Hetaoregion(35°–40°N, 110°E), which lies to the west ofthat identified subjectively. The last case is a Yangtze-Huaihe cyclone(Fig. 2e), which was generated at theYangtze River estuary, migrated northeastward, and reached northwestern Pacific via Korea and Japan.With the migration of the cyclone system, its corepressure lowered gradually. Comparing the resultsfrom the objective method(Fig. 2f) and the manualanalysis(Fig. 2e), we find that although the systemin the objective track died out a little later thanthat in the subjective track, it is still a good tracking.To sum up, the objective method can track individualcyclone trajectory successfully, and the timeerror between the objective algorithm and the subjectivemethod is about one day.

Fig. 2. Cyclone racks identified by the manual analysis(left panels) and the objective algorithm(right panels)beginningat 0000 UTC 26 August 1965(a), 0000 UTC 25 August 1965(b), 0000 UTC 16 February 1971(c), 0600 UTC 15 February1971(d), 0000 UTC 4 May 1971(e), and 1200 UTC 2 May 1971(f). The triangle in each panel denotes the startingpoint of the cyclone track. Circles of larger size represent lower pressure values in hPa and vice versa.

Figure options

In the preceding section, we introduced an objectivemethod for detecting and tracking ECs over EA.Next, we will analyze climatic characteristics of ECsover EA in terms of cyclogenesis, cyclolysis, lifespan, intensity, and movement, based on the objective detectionresults.4.1 Cyclogenesis, cyclolysis, and cyclone lifespan

At first, we define the number of formed cyclonesin each 2.5°×2.5° grid box as the cyclogenesis frequency.The cyclolysis frequency is defined similarly.Figure 3 represents the spatial distribution of cyclogenesisfrequencies for spring(MAM), summer(JJA), autumn(SON), and winter(DJF)over the period1958–2001. We find three major cyclogenesis areas, locatedin the central, eastern, and western parts of thewhole domain, respectively. The central area refers toMongolia, i.e., south of Lake Baikal; the eastern arearefers to the east coast of China–Northwest Pacific; and the western area refers to the western SiberianPlain(55°–75°N, 60°E). It is seen that cyclogenesisfrequencies are greater in central and eastern areasthan in the western area. The highest cyclogenesisfrequency occurs in spring. In summer, no cyclogenesisshows up in the western area, and the easternarea with cyclogenesis extending from central Japanto the oceanic area east of Japan is much smaller thanthat in spring. In addition, cyclogenesis frequency increasesonly in the region situated to the north of theOkhotsk Sea in summer. When autumn comes, cyclogenesisfrequency increases again. The eastern area ofcyclogenesis extending northward to the Okhotsk Seais much larger than that in summer. In winter, cyclogenesisfrequencies in central and western areas decreasegreatly, the scope of which shrinks accordingly, and the eastern area becomes the major cyclogenesisregion. Compared with the situation in autumn, theeastern area exhibits a notable southward shift in winter, extending from the east coast of China eastwardto Northwest Pacific, and northward to central Japan.

Fig. 3. Cyclogenesis frequency(×10−2)in(a)spring, (b)summer, (c)autumn, and (d)winter for the period 1958–2001. Areas with dots enclose values above 15×10−2.

Figure options

Wang et al.(2009)indicated that the main cyclogenesislocation is related to the mean meridional temperaturegradient. On average, two baroclinic frontalzones located at 40°–55°N, 80°–110°E and 30°–40°N, 110°–160°E(the dotted regions in Fig. 4b)are in correspondencewith the central and eastern cyclogenesisareas respectively. In Fig. 3, the central and easterncyclogenesis areas move northward or southward withseasons. This is probably due to the seasonal migrationof frontal zones. We also find that the easternarea is in front of the upper-level trough(Fig. 4a), where positive vorticity advection is active, and thisfacilitates the formation of ground cyclones. The central and eastern areas are respectively located in theeastern areas of the Mongolian Plateau and TibetanPlateau, where leeward slope is another important reasonfor the high frequency of cyclone formation.

Fig. 4. 1958–2001 mean fields of(a)500-hPa geopotential height(solid; gpm) and temperature(dashed; ℃) and (b)700-hPa meridional temperature gradient(℃(deg. lat.)−1). Areas with dots in(b)contain values above 0.8℃(deg.lat.)−1.

Figure options

Spatial distributions of cyclolysis frequency areshown in Fig. 5. There are several major cyclolysisregions. In spring, two cyclolysis regions appear inthe vicinity of 60°–70°N, 90°E and 60°–70°N, 120°E, where cyclones migrate from the West Siberian Plain and central Siberian Plateau, respectively. NortheastChina is another remarkable cyclolysis region, wherecyclones mainly come from the Mongolian Plateau(the central cyclogenesis area). Besides, Okhotsk Sea–Northwest Pacific is also a cyclolysis area, where cyclonespossibly migrate from the east coast of China and southern ocean of Japan(the eastern cyclogenesisarea). It is also found that the spatial distributionof cyclolysis frequency in summer is similar to that inspring, but with lower values and reduced coverage areas.Cyclolysis frequency is comparable in autumn and summer, so is the range of the cyclolysis area. Whenwinter comes, cyclolysis frequency decreases and cyclolysisarea shrinks again, especially over the continent.

Fig. 5. As in Fig. 3, but for cyclolysis frequency.

Figure options

The time duration from cyclogenesis to cyclolysisis defined as the cyclone lifespan. From Fig. 6, we can see that frequency distributions of ECs withvarious lifespans over EA in four seasons are similar, and the cyclone events of 1–2-day lifetime occur mostfrequently(above 12 %). The longer the lifespan, thelower the frequency. Cyclone lifetime is mainly 1–7days; few cyclones can survive more than 7 days. Relativelyspeaking, the average lifespan in winter(1.5 days)is a little shorter than that in other threeseasons.4.2 Intensity

Many indices and analysis methods are used todescribe the intensity of ECs(Jones and Simmonds, 1993). Here, we mainly discuss relative central pressure and Laplacian of pressure. As we all know, meancentral pressure(MCP)can be considered as a criterionfor measuring EC intensity. Simmonds and Wu(1993)defined a new index of relative central pressure(RCP), which is the deviation of MCP from thetime-averaged mean sea level pressure. Their studiesproved that RCP can describe system intensity ofa single cyclone process better than MCP. Figure 7shows that intense spring cyclones are mainly locatedin the area extending from Northeast China eastwardto the Okhotsk Sea and Northwest Pacific. Comparedwith that in other three seasons, intensity of ECs insummer is the lowest, and strong cyclones only occurin the east coastal area of China(25°–30°N, 120°–130°E). Autumn cyclones intensify over the WestSiberian Plain(north of 60°N and west of 90°E)as wellas the Okhotsk Sea. The RCP distribution in winteris similar to that in autumn, but the zone with strongwinter cyclones over the Okhotsk Sea and NorthwestPacific has a southwestward extension.

Fig. 6. Occurrence frequency distributions of cyclones with various lifespans during 1958–2001.

Figure options

Laplacian of pressure can be approximatelytreated as quasi-geostrophic relative vorticity, so herethe mean pressure Laplacian around the center of acyclone over a radius of four latitude degrees is consideredas an indicator to assess cyclone intensity. Acomparison of Figs. 7 and 8 indicates that the distributionof Laplacian of cyclone pressure(Fig. 8)issimilar to that of RCP(Fig. 7), e.g., cyclone intensityin summer is the smallest of the four seasons.In spring, cyclones are intense in Siberia(north of60°N and west of 120°E), Northeast China, and NorthwestPacific. But for summer cyclones, the range ofthe cyclone area, where the values of pressure Laplacianare greater than 1.0 hPa(deg.)⁻², is narroweddown greatly to Siberia north of 65°N. When autumncomes, the strong cyclone area extends, and Siberia and Northwest Pacific are two regions where cyclonesystems are intense. Furthermore, the spatial distributionof pressure Laplacian in winter is similar tothat in autumn, but with a larger coverage of intensecyclones over Northwest Pacific.

Fig. 7. Relative central pressure(hPa)of cyclones in(a)spring, (b)summer, (c)autumn, and (d)winter for the period1958–2001. Areas with dots contain values under –20 hPa.

Figure options

Fig. 8. Laplacian of cyclone pressure(hPa(deg.)−2)in(a)spring, (b)summer, (c)autumn, and (d)winter for the period 1958–2001. Areas with dots enclose values above 1.0 hPa(deg.)−2.

Figure options

The total travelling distance of cyclones duringthe whole cyclone cycle is analyzed and shown in Fig. 9. It is found that, on average, cyclones travel farthestin spring(about 3600 km), and nearest in winter(about 3100 km). The longest travelling distanceof cyclones can reach 7500 km. This is the same infour seasons. Besides, the travelling distances between1500 and 2000 km are most frequently observed in fourseasons, accounting for about 4.2%, 4.0%, 4.1%, and 2.5% of the total, respectively.

Fig. 9. Frequency distribution of cyclone travelling distance in(a)spring, (b)summer, (c)autumn, and (d)winter forthe period 1958–2001.

Figure options

Spatial distribution of seasonal cyclone travellingvelocities(Fig. 10)indicates that cyclones movemainly eastward in the region north of 60°N, and moveprincipally northeastward in the region south of 40°N, especially over the ocean area. Furthermore, cyclonesin the region to the west of Mongolia move northeastward and then southeastward after bypassing theMongolian Plateau. It is also found that ECs move thefastest in winter, and the slowest in summer. This isbecause the maximum speed of steering flow over 500hPa occurs in winter and the minimum occurs in summer.In spring, the region where cyclones travel fastwith velocities greater than 15 m s−1 is located to thenorthwest of Lake Baikal. In summer, cyclones movevery slowly, and fast-moving cyclones are only foundin the area of 65°N, 125°–145°E. Travelling speeds ofautumn cyclones increase rapidly. Spatial distributionof travelling velocities in autumn is similar to that inspring, but with the area of the fast-moving region alittle larger. In winter, the rapid-moving cyclone regionsituated to the northwest of Lake Baikal extendsnortheastward to the north of the Okhotsk Sea. Itis interesting to find that two new rapid-moving cycloneregions(with velocity greater than 15 m s−1)emerge over Northeast China and south of Japan. Inaddition, travelling speeds of cyclones in parts of theYangtze-Huaihe River region can increase to 15 m s−1in winter. The accelerated cyclone migration between20° and 40°N is probably due to the strengthening ofthe upper westerly jet in winter.

Fig. 10. Cyclone travelling velocity(m s−1)in(a)spring, (b)summer, (c)autumn, and (d)winter for the period 1958–2001. Areas with dots enclose values greater than 15 m s−1.

Figure options

In order to derive a 44-yr climatology of dominatingtrajectories of ECs in EA, especially in thearea south of 55°N and east of 80°E, we use a clusteranalysis named k-means procedure. The k-meansalgorithm(Mirkin, 1996; Hartigan, 1975)is designedto rearrange elements to k groups with minimum variabilitywithin clusters and maximum variability betweenthem. Since all displacement vectors undergoingthe analysis must have the same dimension, onlycyclone tracks within the first 48 h are taken into account and the tracks beyond 48 h of those long-livedcyclones are discarded. This means that 3612 detectedtracks with their 48-h cyclone positions, generated inthe area south of 55°N and east of 80°E, are considered.They take up about 60% of all cyclone tracksin the same region. The highest frequencies of cyclonelifetime are found between 1 and 2 days, whichis described distinctly in Fig. 6, so choosing cycloneprocesses having a minimum 48-h lifespan can includeas many samples as possible. In order to obtain fewerclusters, all sample tracks are classified into two sets, corresponding to cyclones formed in the region westof 125°E(first longitude ≤ 125°E) and east of 125°E(first longitude > 125°E). Before performing the kmeansalgorithm, all the coordinates within each setare st and ardized. As the variance in the zonal directionis larger, this normalization gives more weight tothe meridional displacements. The number of clustersk is considered suitable whenever there is no overlapbetween the mean positions of different clusters; meanwhile, the member of each cluster shares similar formationareas, velocities, and directions. Moreover, wecarry out an analysis of variance to test whether themean positions of tracks are significantly different(atthe 0.01 significance level)between clusters.

The statistical results in Fig. 11 show that thereare four dominating trajectories in the region westof 125°E and two east of 125°E, respectively. Themain trajectory b and s and main areas of activity ofeach cluster are estimated by one st and ard deviationof the latitude coordinates. Among the trajectories, two are initiated in Mongolia, one in the lower reachesof Yellow River, and the other in the lower reaches ofYangtze River, which consist with those found by themanual identification(Fig. 3.23 in Zhu et al., 2000) and can affect China powerfully. The occurrences ofspring s and storms in North China and winter blizzardsin Northeast China are probably due to the explosivedevelopment of cyclones in the two tracks in Mongolia.Furthermore, the development of cyclones alongthe two tracks in the lower reaches of Yellow River and Yangtze River might be a significant reason forthe heavy rainfall and offshore winds in these regions.We also find that the five tracks east of 110°E arebasically northeastward. The northeastward movementsof the cyclones along the north path are mainlydue to the obstruction of the north-south mountain inNortheast China. Cyclones in other four paths movenortheastward, possibly owing to the southwest steeringflow in front of the EA trough. To obtain an intuitionalview of the cyclone tracks, we plot all cyclonetracks including their 48-h positions used in the kmeanscluster analysis during the periods 1958–1967, 1968–1977, 1978–1987, and 1988–2001(figure omitted).It is seen that the real cyclone formation locationsin a few paths south of Japan are to the west ofthe mean position. Some cyclones form in the vicinityof Taiwan, China, and then move northeastward(thedotted line in Fig. 11).

Fig. 11. Cluster mean position of cyclone tracks. The inverted triangle denotes the starting point of each track, slashinglines represent areas of influence of each cluster, and the line connected with black solid dots denotes the track by thek-means method.

Figure options

When the central pressure of a cyclone at latitudeφ drops at least 24×sinφ/sin60°(in hPa)duringa 24-h period, we call it an explosively-deepening cycloneor a bomb(Sanders and Gyakum, 1980). Figure 12 describes that the preferred regions of explosivecyclone intensification are found in the east coast ofChina, Japan, and Northwest Pacific, while very fewover the inl and area. Inl and bombs mainly happen inMongolia and the region northwest to the BalkhashLake. Compared with those in North America(NA)(Fig. 3 in Gyakum et al., 1996), inl and bombs inEA are less likely to happen. Compared with about14 bombs from January to March every year during1989–1992 in NA(Fig. 5 in Gyakum et al., 1996), there are fewer explosively-deepening cyclones in EA, only about 8 during the same period. Furthermore, the occurrence frequency of bombs in EA has largeseasonal variations, i.e., winter bombs occur most frequentlywhile few bombs appear in summer. Why dothe bombs often happen over the oceans and rarelyoccur in summer? This is because the occurrence ofbombs has something to do with the background field.Most rapidly deepening cyclones form in the areas onthe left of the outlet area of the upper level westerlyjet, such as Japan and Northwest Pacific, especially inwinter and spring.

Fig. 12. Occurrence frequency(10−2 cyclones(2.5×2.5 deg.)−1 season−1)distributions of explosively-deepening cyclones in(a)spring, (b)summer, (c)autumn, and (d)winter during 1958–2001.

Figure options

Temporal variations of the central pressure ofexplosively-deepening cyclones in two selected years(1990 and 1992)are described in Fig. 13, in which 0h denotes the onset of the most rapid fall of centralpressure during the 24-h period. There are 19 and 18bombs in the two years. We can see that the centralpressure of EA bombs changes in the range of 1020 to960 hPa, but that of NA bombs can be as low as 930hPa(Fig. 5 in Gyakum et al., 1996). This indicatesthat, in these two years, rapid intensifying cyclones inNA are stronger than those in EA. However, by only atwo-year comparison, we cannot draw any conclusionabout the differences between the bombs in EA and NA. Much more detailed and in-depth work should bedone in the future.

Fig. 13. Temporal variations of central pressure of explosively-deepening cyclones in(a)1990 and (b)1992.

Figure options

In this study, climatic characteristics of ECs overEA are analyzed based on an improved objective identification and tracking method. Major conclusions canbe summarized as follows:

(1)The objective method finds cyclone locationsbased on local minima of sea level pressure usingLaplacian of pressure. Compared with the manualanalysis method, the objective algorithm is more objective and time saving, and can be applied to allkinds of observation, reanalysis, and model data. Theobjective detection method can successfully identifymature ECs which are intense and large scale, whilesome weak ECs without fronts or closed isobars cannotbe identified. This disadvantage is possibly connectedwith the low resolution of the data.

(2)We make use of steering flow and extrapolatethe cyclone movement to predict likely successors ofprevious cyclones by choosing a certain weighting factor.According to the features of the ECs in EA, someclassification and elimination are done. The number ofcyclone processes tracked by the objective algorithmis 10%–20% more than that tracked by the manualanalysis. More than 80% of real cyclone processes canbe tracked by the objective algorithm, and the timeerror between the objective algorithm and the subjectivemethod is about 1 day.

(3)There are three major cyclogenesis regions, i.e., Mongolia(central area), east coast of China–Northwest Pacific(eastern area), and western SiberianPlain(55°–75°N, 60°E)(western area). The central and eastern areas have a good accordance with twobaroclinic zones, where frontogenesis and cyclogenesisprefer to occur. The central and eastern areascover the lower reaches of Mongolian Plateau and TibetanPlateau where leeward terrain contributes tocyclogenesis. Spring is the high-frequency season forcyclogenesis, while winter is the low-frequency season, especially over the continent. Relatively speaking, cyclolysisfrequency in the region north of 60°N, NortheastChina, and Northwest Pacific is higher than thatin other areas.

(4)Cyclone lifespan changes mainly from 1 to7 days, and cyclones with 1–2-day lifetime occur themost frequently. Lifetime of winter cyclones is theshortest, and cyclones in other three seasons have alonger lifespan. Summer cyclones are weakest, and those in other three seasons are stronger. Strong cyclonesoften appear over the western Siberian Plain, Northeast China and Okhotsk Sea–Northwest Pacific.

(5)The k-means clustering analysis produces 6major cyclone trajectories in the area 20°–55°N, 80°–160°E. Among them, four trajectories affecting Chinaare probably associated with severe weather/climatephenomena such as spring s and storms, winter blizzard, heavy rainfall, and offshore winds, consistentwith those found by the manual analysis. High altitudeterrain and steering flow make some tracksnortheastward. On average, winter cyclones move thefastest while summer cyclones move the slowest. Theregion in which cyclones travel fast is located to thenorthwest of the Lake Baikal. In winter, two new fastmovingcyclone regions are found in Northeast China and the ocean area south of Japan. The strengtheningof the upper level westerly jet in winter is probablythe reason for the acceleration of cyclones travellingbetween 20° and 40°N.

(6)The preferred regions of explosively-deepeningcyclones(bombs)are found in Pacific coast and NorthwestPacific while very few bombs happen in the regionaway from the ocean. Northwest Pacific is on the leftof the outlet of the upper level jet, where cyclones preferto break out. Compared with their counterpartsin NA, fewer cyclones in EA develop explosively. Thenumber of bombs in 1990 and 1992 in EA is less thanthat in NA, and bomb intensity in EA is weaker thanthat in NA.

As can be seen from the above analysis, the objectivealgorithm on the basis of sea level pressureminima for detecting and tracking ECs over EA iseffective and feasible. The database of ECs in a longtime sequence(1958–2001)can supplement and improvethe previous manual analysis results over shortperiods. At present, most studies of ECs over EAfocus on cyclone frequency and intensity, while in thispaper, we gain insight on cyclone lifespan, travellingdistance, travelling velocity, dominating tracks, aswell as their seasonal differences. The interannual and interdecadal variability of ECs over EA has not beenwell investigated and the reason for EC changes overdifferent timescales is not clear, so further discussion and in-depth investigations along this line should bedone in the near future.

Acknowledgments. The authors thank theMeteorology Reference Room, National MeteorologicalInformation Center for supplying the historicalweather charts for this study.