The climatology of cold‐air damming in the Kanto Plain, Japan

The climatological characteristics of cold‐air damming (CAD) were statistically investigated on the Kanto Plain throughout the year. We detected 397 CAD events from 1991 to 2020 using hourly surface observation data from the Japan Meteorological Agency. The statistical analyses revealed the following characteristics: (1) CAD is most frequent in autumn followed by spring; (2) strong CAD is more frequent in the cold season than in the warm season; (3) The long‐term trend in CAD frequency is not obvious; (4) 53% of the events finish in 12 h or less, and 17% of them continue for 24 h or more. CAD with precipitation tended to be stronger than that without precipitation, and the latter was more frequent in the nighttime than in the daytime, indicating the contribution of diabatic processes to the development of CAD. Analysis of synoptic fields in CAD cases using long‐term reanalysis data revealed six patterns that differed in the intensity and position of surface highs and lows, corresponding to the typical seasonal sea level pressure fields.


| INTRODUCTION
Cold-air damming (CAD) is a mesoscale feature characterized by a shallow layer of cold air entrenched along the eastern slopes of a north-south mountain range (Araki, 2015a;Bailey et al., 2003;Bell & Bosart, 1988;Ellis et al., 2018;Kitabatake, 2019).The cold air is usually a dome-shaped and referred to as a 'cold air dome' (Bell & Bosart, 1988;Rackley & Knox, 2016) with a thickness of less than 1 km (Araki, 2019a) and a width corresponding to the internal Rossby's radius of deformation, which is of the order of 100 km (Araki, 2019a;Kitabatake, 2019).CAD to the east of the Appalachian Mountains (Appalachian CAD) usually persists for a timescale of hours to a few days (Rackley & Knox, 2016).CAD can be identified by a surface pressure ridge extended from north to south and a northerly wind prevailing in the boundary layer to the east of the mountain (called 'damming region', Bell & Bosart, 1988;Rackley & Knox, 2016).In the surface weather map, a 'U'or 'wedge'-shape isobar pattern appears over the damming region (Bailey et al., 2003;Bell & Bosart, 1988;Rackley & Knox, 2016).
CAD exerts a severe impact on sensible weather, including a significant temperature decrease, changing the precipitation type in winter, triggering convection along the periphery of the cold air dome in the summer, and visibility impairment (Bailey et al., 2003;Forbes et al., 1987;Kitabatake, 2019;Rackley & Knox, 2016).CAD is occasionally accompanied by coastal fronts (Appel et al., 2005;Araki, 2015b;Fujibe, 1990).Despite these significant impacts, the currently used numerical models face challenges in the reproducibility of CAD behaviour and often predict erosion too early (Lackmann & Stanton, 2004;Rackley & Knox, 2016).
Previous studies have demonstrated that two factors play crucial roles in CAD development.The first factor is the north-high and south-low (NhSl) sea level pressure (SLP) field, which causes a statically stable easterly wind around the mountains (e.g., Araki, 2015a;Bailey et al., 2003;Bell & Bosart, 1988;Kitabatake, 2019).When this wind is blocked by mountains, it turns equatorward, namely southward in the Northern Hemisphere, and maintains a cold air dome on the eastern slopes according to the force balance between the pressure gradient and friction (Bell & Bosart, 1988;Xu, 1990).In the eastern Appalachian Mountains, an anticyclone passing to the north ('parent high' or 'parent anticyclone') is often accompanied by CAD (Bailey et al., 2003;Ellis et al., 2018;Rackley & Knox, 2016).The second factor is the diabatic processes that intensify cold air in the lower troposphere on eastern slopes, including evaporative cooling and insolation sheltering by clouds (e.g., Araki, 2019a;Bailey et al., 2003;Bell & Bosart, 1988;Fritsch et al., 1992;Kitabatake, 2019;Rackley & Knox, 2016).At night, clouds play an opposite role because they prevent radiative cooling (Fritsch et al., 1992).Conversely, solar heating during the day can cause CAD erosion (Lackmann & Stanton, 2004).
CAD can occur in many other regions in the midlatitudes with a long north-south mountain range (Bailey et al., 2003).Araki (2015a) pointed out that the Kanto Plain, in southeastern Honshu, the main island of Japan (Figure 1a), is one such region.It is a highly populated area centred in Tokyo and is surrounded by mountains over 1000 m in elevation in the west (Figure 1b).Araki (2015a) noted that CAD in the Kanto Plain often appears when a developing extratropical cyclone approaches the southern coast of Honshu ('southcoast cyclone, SCC') in the winter.Subsequently, Honda et al. (2016) and Araki (2019b) noted the contribution of CAD to snowfall in the Tokyo metropolitan area.However, few studies in the Kanto Plain have focused on CADs in other seasons, even though Appalachian CAD occurs in all seasons and under various synoptic conditions (Bailey et al., 2003;Rackley & Knox, 2016).
An example of CAD events other than those associated with the winter SCC is presented in Figure 2. At 18:00 UTC on 22 November 2019 (03:00 Japan Standard Time, JST on 23 November 2019), a migratory anticyclone was located east of Honshu, and a weak low was located south of Honshu (Figure 2a).The SLP field around the Kanto Plain was north-high and south-low (NhSl), which is ideal for CAD development.The wind and SLP fields based on hourly observation data at the Japan Meteorological Agency (JMA) observatories and the Automated Meteorological Data Acquisition System (AMeDAS) stations around the Kanto Plain are shown in Figure 2b.The high SLP and northerly wind area extended along the eastern side of the mountain range, from northeast to southwest.Figure 3 shows the vertical cross section of the potential temperature and wind based on the Meso Analysis data of the JMA.East of the mountains, a cold air dome was conspicuous below the 900 hPa level.The cold air trapped along the eastern side of the mountain range had a typical CAD structure.
As in this case, CAD in the Kanto Plain is not limited to those associated with SCC.In order to understand the climatology of CAD, this study statistically analysed CAD throughout a year and discussed its seasonal frequency, hourly frequency, duration, effects of diabatic processes associated with precipitation, and the synoptic field patterns causing CAD.The overall approach is shown in Figure 4.

| CAD detection
We utilized surface data from the JMA observatories to detect CAD events.Considering that CAD occurs on a time scale of hours to a few days (Rackley & Knox, 2016), we set the analysis period to be 30 years from 1 January 1991 to 31 December 2020 (10,958 days; 262,992 h), for which hourly data at the JMA observatories were available.We used data on SLP, station-level pressure, and temperature at Fukushima (FKS), Takada (TKD), Kumagaya (KMG), Choshi (CHS) and Shizuoka (SHZ, Figure 1b) to detect low-temperature and pressure ridges in the damming region.The periods that contained missing data at any of the five sites were excluded from the analysis.The missing periods totalled 113 h, which was sufficiently small.We calculated the potential temperature (θ [K]) from the station-level pressure (p stn [hPa]) and temperature (T [ C]) using the following equation: where R d is the gas constant for dry air [J/kgÁK] and C p is the constant pressure specific heat [J/kgÁK].Referring to Bailey et al. (2003), Rackley andKnox (2016), andAraki (2019b), we attempted to detect CAD based on a surface pressure ridge with low temperature extending from the north along the eastern slopes.The Laplacians of SLP and potential temperature normal to the mountain range enable quantification of the cold ridge (Bailey et al., 2003).We employed the Laplacian (r 2 x) calculated from the hourly data at CHS, KMG, and TKD using the following (Araki, 2019b;Bailey et al., 2003): where x is either SLP (p) or potential temperature (θ), d is distance between observation points (d 2 − 3 : 146.8 km, d 1 − 2 : 140.8 km), and the subscripts 1, 2, and 3 correspond to CHS, KMG, and TKD, respectively.Smaller r 2 p (larger r 2 θ) value indicates higher pressure (colder) in the inland than in the coastal area.
To confirm the presence of a cold ridge from north to south, following requirements must be met: (1) p 2 >p 1 , and p 2 >p 3 to ensure the presence of the ridge, (2) r 2 p<r 2 p, where r 2 p is calculated as one standard deviation below the average of all negative r 2 p values in the dataset, in order to avoid detecting very weak events (Bailey et al., 2003), (3) r 2 θ>0 to detect the cold air accumulation in the inland, and (4) p 4 >p 2 >p 5 , where subscripts 4 and 5 correspond to FKS and SHZ, respectively, to ensure that the ridge extends from north to south.Finally, a CAD event was defined by a period which satisfied above four requirements for six consecutive hours or more.If the time difference between the two events was within 3 h, they were regarded as a single event in order to allow for the temporary weakening of CAD.These lengths were set according to Bailey et al. (2003) and Rackley and Knox (2016).Bailey et al. (2003) identified the strong CAD events using a linear combination of normalized measurements of duration, the magnitude of the maximum r 2 p, r 2 θ, and the magnitude of the along-barrier pressure gradient.In addition, they defined the CAD peak time with the r 2 p value.However, in this study, the hourly CAD intensity was measured based on the r 2 θ value, which can directly represent the intensity of cold air in the plain.The results did not change significantly between the definition with r 2 p and r 2 θ as shown in Appendix A. The peak time of an event was defined as the time of maximum r 2 θ in the same way.Following Bailey et al. (2003) and Rackley and Knox (2016), the top 50% of cases with maximum r 2 θ were defined as strong CAD events.

| Determination of precipitation
To examine the effects of diabatic processes associated with precipitation, we determined the presence of precipitation in the Kanto Plain for each event using hourly precipitation data from AMeDAS, following the procedures described by Fujibe (1990), and compared the frequency and intensity of CAD with and without precipitation.Considering that CAD is a low-level feature, stations in the Kanto Plain at heights below 300 m were used (Figure 5).Parameter R was defined as the percentage of stations where precipitation of at least 1 mm was observed during an event among all stations except stations with missing data.All events were classified into three categories in R as shown in Section 3.2.

| SLP field classification and composite analyses
We investigated the characteristics of the synoptic-scale field around Japan during CAD events in the Kanto Plain based on a classification scheme applied to SLP patterns.We utilized the Japanese 55-year Reanalysis data (JRA-55, Kobayashi et al., 2015;Harada et al., 2016), which were available in a grid format with a horizontal resolution of 1.25 × 1.25 and a time step of 6 h (0000UTC, 0600UTC, 1200UTC, and 1800UTC; UTC is 9 h behind JST).We set the analysis domain from 20 N to 60 N and 120 E to 160 E, and thinned out some grid points to compensate for the uneven grid point density by latitude.The analysis domain and grid points are shown in Figure 6.The classification was based on the SLP deviation from the average over these grid points at the JRA-55 analysis time closest to the CAD peak time.If the CAD peak was at 0300UTC, 0900UTC, 1500UTC, or 2100UTC, the JRA-55 data at 0000UTC, 0600UTC, 1200UTC, and 1800UTC, respectively, were used.
The analysis was targeted only on events whose peak time was 96 h or more after those of the previous event to ensure that the synoptic field had shifted sufficiently from the previous event, as suggested by Appel et al. (2005).This procedure resulted in a total of 346 events.
Cluster analysis directly applied to data from many stations tends to be affected by local variations (Kato et al., 2013).To focus on representative large-scale patterns, our analysis was conducted for the scores obtained from principal component analysis (PCA), which was applied to the SLP grid point data using a variancecovariance matrix.The m-th principal component score of i-th event (z mi ) is represented by where N is the total number of grid points, x in is the SLP at the n-th grid point in the i-th case, and w mn is the m-th principal component loading and assumes The similarity between the events is expressed quantitatively using the Euclidean distance (D) calculated using where M is the number of principal components.Note that the scores (Z), eigenvalues, and distances are dimensional quantities because they are purposely not standardized to maintain a variable scale for each region.
Finally, clustering was performed by applying Ward's method to D.
After classifying the SLP fields, composite analyses at the peak time were conducted to clarify the synoptic scale structure of each cluster.We investigate the structure of warm moist air advection around the CAD at the 850 hPa, ridge and trough at the 500 hPa, and the jet at the 300 hPa, respectively.These factors are likely to be related to the development of CAD (Bailey et al., 2003).

| Basic statistics
Using the above-mentioned procedure, 397 CAD events were detected for the period 1991-2020, equivalent to 13.2 events per year.Figure 7 shows the basic statistics.The number of CAD events did not show a clear long-term trend, but a large year-to-year variation with a maximum in 1993 (Figure 7a), when the summer was extremely cool in Japan.Indeed, the numbers of events and total duration in July and August 1993 was the highest of the 30 years (Figure S1 on Supporting information).The frequency distribution of duration (Figure 7b) shows that approximately half of the events (53%) were completed in 12 h or less, whereas some events (17%) persisted for 24 h or more.The monthly frequency of all events and strong events, and the percentage of strong events are shown in Figure 7c.The frequency is high in October, September, and April, in descending order, and lowest in July.This result is basically consistent with the Appalachian CAD  (Bailey et al., 2003).The percentage of strong events were higher in the cold season compared to those in the warm season, which is similar to the Appalachian CAD (Bailey et al., 2003).

| Differences in CAD characteristics depending on the precipitation
The frequency distribution of R is shown in Figure 8.Based on this, all events were classified into three categories: events with R greater than 80% as 'Precipitation Cases (PC)'; those with R less than 20% as 'Non-Precipitation Cases (NPC)'; and others as 'Other Cases (OC'.The numbers of PC, NPC, and OC were 243, 86, and 68, respectively. Figure 9a shows the hourly frequency rate, defined as the ratio to the daily average, for PC, NPC, and OC.The NPC have a clear diurnal cycle in which CAD was more frequent from midnight to dawn, while PC have only a small diurnal change.The lowest frequency of PC is at 17:00, which occurs later than that of NPC.The frequency distribution of CAD intensity (r 2 θ at the peak time) for PC, NPC, and OC is shown in Figure 9b as a box plot.61% of the PC belonged to the strong CAD group, and PC were clearly stronger than those of NPC.This difference was statistically significant evaluated by Welch's t-test at the 1% significance level.However, 29% of the NPC belonged to the strong CAD group, indicating that precipitation is not necessary, although favourable, for the development of a strong CAD.

| Results of the SLP field classification
Figure 10 shows the average SLP field for all cases.A high and a low are located east of Hokkaido and on the southern coast of Honshu, respectively.Thus, the NhSl SLP field pattern is evident as an average feature.The northern high corresponds to the 'parent high (anticyclone)' for Appalachian CAD (Bailey et al., 2003;Ellis et al., 2018;Rackley & Knox, 2016).
Table 1 shows the contribution and cumulative contribution rates for each principal component, respectively.We regarded the top seven principal components, which accounted for 80.4% of the total variance, as The contribution ratio (%), cumulative contribution ratio (%) of n th principal component (PCn).representing the main features of the SLP fields.
Figure 11 shows a dendrogram of the cluster analysis results.We selected six groups separated by a distance of 400 hPa for the subsequent analysis.
Figure 12 shows the SLP fields for the six clusters.All clusters indicate the NhSl SLP field, although there are some differences among them.In Cluster 1 (C1), the southern low is relatively significant, whereas the northern region extends from a dominant high on the Eurasian Continent toward the east of Hokkaido (Figure 12a).C2 was similar to C1, except that the northern high had a centre to the east of Hokkaido (Figure 12b).C3 was similar to C2, except that the northern high was located at higher latitudes (Figure 12c).In C4, the southern low is located in western Japan and the northern high is located at higher latitudes (Figure 12d) than the average field (Figure 11).In C5, the northern high was dominant, whereas the southern low was not (Figure 12e).In C6, the southern low is relatively intense, and the northern high is located southeast of Hokkaido (Figure 12f).The numbers of events of C1, C2, C3, C4, C5, and C6 are 38, 72, 77, 42, 78, and 39, respectively.

| Characteristics of each cluster
The monthly frequencies of each cluster are shown in Figure 13.C1 was frequent in winter, with 82% of events appeared between December and February.C2 frequently occurred during the cold season, from November to March.C3 was found during all seasons expect from late spring to early summer.In contrast, C4 was predominant during the warm season, especially in June and July.C5 was mainly observed from September to October, when CAD was most frequent in the year (Figure 7c).C6 mainly appeared in the spring (from March to May) and August.
The results of the composite analyses of the upper layers around Japan are shown in Figures 14-16.At the 850 hPa level (Figure 14), southerly warm advection from the Pacific Ocean to the damming region is obvious in all clusters, which has also been observed in Appalachian CAD (Bailey et al., 2003;Bell & Bosart, 1988;Ellis et al., 2018;Forbes et al., 1987).A cyclonic circulation centred on the Sea of Japan brings the warm moist air except C5 (Figure 14a-d,f).In C5, instead, an anticyclonic circulation to the east of Honshu plays a role (Figure 14e).
At the 500 hPa level (Figure 15), there was a trough around the Korean Peninsula in all clusters.The trough was more conspicuous in C1, C2, C3, and C6 than that in the other two clusters, corresponding to a relatively intense low in the SLP field (Figure 12).In C2, C3, C4, and C5, a clear ridge appeared near Hokkaido or north of it (Figure 15b-e).Both the trough and ridge are situated to the west of the well-developed low and the northern high, respectively.Thus, they are involved in sustaining the NhSl SLP fields in each cluster.
At the 300 hPa level (Figure 16), there were two strong wind cores to the southwest and east of Honshu in C1, C3, and C6 (Figure 16a,c,f).This feature was observed in many individual cases, as shown in Figure 17.It seems to be the 'double-jet structure' (Araki et al., 2019), in which updrafts tend to prevail in the region between the two strong-wind areas, because of ageostrophic vertical circulation (Araki et al., 2019;Bailey et al., 2003;Uccellini & Kocin, 1987).In this composite field, this area corresponds to the warm advection area at the 850 hPa level.This result is consistent with the field surrounding the CAD with precipitation east of the Appalachian Mountains (Bailey et al., 2003).C2 and C5 only have a clear core to the east of Japan (Figure 16b,e).In C4, while the strong wind core was not pronounced, unlike in other clusters, the contour lines had a clear anticyclonic curvature in northern Japan (Figure 16d).

| The synoptic scale field pattern
Our analysis revealed that CADs in the Kanto Plain correspond to the NhSl SLP field around Japan, however, they include various patterns in addition to those associated with a south-coast cyclone (SCC) in winter.In this section, we discuss the climatological meaning of each pattern as a synoptic CAD environment.
Overall, the SLP field patterns differed in the intensity and position of the southern low and northern high.C1 and C2, which are predominant in winter, had a low on the southern coast of Japan (Figure 12a,b).These SLP fields correspond to the SCC pattern in winter and are occasionally accompanied by snowfall over the Tokyo metropolitan area (e.g., Araki, 2015a;Araki, 2019b;Honda et al., 2016).Unlike the Appalachian CAD, which is mainly accompanied by a parent high, C1 does not have a distinct high centre around northern Japan.Although a significant SCC also appears in C6 (Figure 12f), C6 is distinguished from these winter types in that the SLP is not high in the Eurasian Continent.
In summer, a strong surface high occasionally appears over the Sea of Okhotsk and brings a cold northeasterly wind on the Pacific Ocean side of northeastern Japan (called 'Yamase'), typically corresponding to the blocking of westerlies on the upper layer near Far East (Nakamura & Fukamachi, 2005).C4 corresponds to this scenario.
F I G U R E 1 3 The percentage of frequency of each cluster for each month.
C5 indicates a migratory anticyclone passing northern Japan (Figure 12e) supported by a ridge at the 500 hPa level (Figure 15e).This pattern indicates that CAD can be developed only by a prominent migratory anticyclone passing through northern Japan, even if the southern low is weak.
C3 did not exhibit an obvious seasonal trend.The SLP field (Figure 12c) and characteristics of the upper layer of C3 have features close to the average pattern over all CAD cases.We estimated that C3 represents an intermediate situation among the cases.4.2 | Seasonal CAD frequency Yoshino and Kai (1975) defined daily synoptic scale SLP field patterns, focusing on the shape, location, and type of high and low around Japan for 30 years , and described the monthly frequency of each pattern.The NhSl SLP field corresponds to their IIc type in which a cyclone passes northeast or east-northeast from Taiwan to the southern coast of Japan, and the IIIa type in which a migratory anticyclone passes north of Japan or northern Japan.Figure 18 shows the monthly frequency percentages of the IIc and IIIa types based on the results shown in their Table 1.In autumn, when CAD is most frequent, the IIIa type is more frequent than in any other season.This is consistent with the high frequency of C5 where CAD is associated with a remarkable migratory anticyclone in northern Japan.In spring, a cyclone frequently develops and passes along the southern coast of Honshu (Chen et al., 1991) corresponding to the high frequency of the IIc type.This tendency was also observed in this study as the high frequencies of C6, in which a significant SCC is involved in CAD development in spring.
Thus, the monthly CAD frequency was consistent with the seasonal tendency of the SLP field around Japan.The high frequency of CAD in autumn is due to the high frequency of migratory anticyclones passing through northern Japan or north of Japan; in spring, the high frequency of CAD is due to the passage and genesis of cyclones along the southern coast of Honshu.

| The roles of diabatic processes
Our statistics for NPC and PC revealed that diabatic processes (evaporative cooling and insolation sheltering by clouds) are effective for CAD development.The diurnal cycle of the NPC is clearly reflected in CAD erosion by solar heating (Lackmann & Stanton, 2004).In contrast, the lack of diurnal variation in PC suggests that the PC is hardly affected by solar heating and tends to persist during the daytime.In addition, the difference in CAD intensity (r 2 θ) between PC and NPC showed that the diabatic processes intensified CAD.Fritsch et al. (1992) showed that the diabatic processes contribute to intensifying CAD by supplying cold air, increasing the static stability of the damming region, and enhancing southward cold air advection.Similar processes are likely to contribute to CAD in the Kanto Plain.

| SUMMARY
To reveal the climatological characteristics of CAD in the Kanto Plain, we detected 397 CAD events using hourly surface observation data over 30 years and conducted statistical analyses.The results are summarized as follows: 1. CAD is most frequent in early autumn, followed by spring, while strong CAD is more frequent in the cold season than in the warm season.2. Nearly half of the CAD events (53%) lasted for 12 h or less, and 17% of events continued for 24 h or more.3.While CAD events without precipitation have a clear diurnal cycle with high frequency from midnight to dawn and low frequency in the afternoon, those with precipitation appear almost constantly throughout the day.Additionally, the latter was significantly stronger than the former.These results indicate that the diabatic processes associated with precipitation contribute CAD to develop more strongly and sustain during daytime.4. The synoptic scale fields during CAD events were classified into six patterns using SLP field data.All patterns generally indicate the NhSl SLP field, but the intensity and position of the northern surface highs and southern lows were different from each other.5.The seasonal characteristics of the synoptic scale field in CAD events correspond well with the seasonal tendency of the SLP field pattern described in previous studies.In autumn, migratory anticyclones frequently pass through northern Japan or north of Japan, resulting in a high frequency of CAD.In spring, a high frequency of cyclones generated and developing on the southern coast of Honshu corresponded to frequent CAD cases.
This study discussed the basic climatological characteristics, the contribution of diabatic cooling, and the background field of CAD in the Kanto Plain.Future F I G U R E 1 8 The percentage of monthly frequency of IIc (a cyclone passes northeast or east-northeast from Taiwan to the southern coast of Japan) and IIIa (a migratory anticyclone passes north of Japan or northern Japan) types demonstrated in Yoshino and Kai (1975).The values were based on the Table 1 of the Yoshino and Kai (1975).[Colour figure can be viewed at wileyonlinelibrary.com] works will focus on the mesoscale and elucidate the structure and its temporal variation.

F
I G U R E 1 (a) Map of Japan and observation station locations around Japan.The center frame indicates the domain of (b).(b) Locations of the observation stations, Kumagaya (KMG), Choshi (CHS), Takada (TKD), Fukushima (FKS), and Shizuoka (SHZ), which were used to detect the CAD events and the terrains around the Kanto Plain.[Colour figure can be viewed at wileyonlinelibrary.com]

F
The vertical profile of potential temperature [K] (shade) and horizontal wind [m/s] (barbs) with Japan Meteorological Agency's Meso Analysis data at 18:00 UTC 22 November 2019 (03:00 JST November 23 2019) along the Line AB (showed in Figure 15).▲ indicates the Pacific coast.[Colour figure can be viewed at wileyonlinelibrary.com]

F
I G U R E 5 The position of Japan Meteorological offices (circles), Automated Meteorological Data Acquisition System stations (AMeDAS stations, triangles), and name of prefectures.[Colour figure can be viewed at wileyonlinelibrary.com]F I G U R E 4 Overall approach and configuration of this study.

F
I G U R E 6 The grid points of Japanese 55-year Reanalysis (JRA-55) data, which was used in the classification procedure.[Colour figure can be viewed at wileyonlinelibrary.com]

F
I G U R E 8 The frequency distribution of R. [Colour figure can be viewed at wileyonlinelibrary.com]F I G U R E 9 (a) Hourly frequency rate (the ratio of the hourly frequency to the daily average of it) of PC, NPC, and other cases.(b) The distribution of intensity (the value of r 2 θ at the CAD peak time) of PC, NPC, and other cases.the cross mark represents the average, the dots represent outliers (the values more than 1.5 times the interquartile range from the 1st or 3rd quartile).[Colour figure can be viewed at wileyonlinelibrary.com]F I G U R E 7 Results of statistical analyses.(a) the long-term trends of the number of the CAD, (b) the frequency distribution of duration of CAD every 1 h, and (c) the monthly frequency (the number of CAD events divided by the number of days for each month) of total CAD (black bars) and strong CAD (grey bars), and the percentage of strong CAD (polyline).[Colour figure can be viewed at wileyonlinelibrary.com]

F
I G U R E 1 0 The average sea level pressure (SLP) field (deviation from the area average).The solid (dashed) line indicates the positive (negative) value of the deviation of SLP every 1 [hPa].The white area corresponds to the target area of the principal component analysis for the classification.[Colour figure can be viewed at wileyonlinelibrary.com]F I G U R E 1 1 A dendrogram produced by the cluster analysis.The number of Cluster n (Cn) is noted in parentheses.

F
I G U R E 1 4 The average temperature [K] (deviation from the area average, shade), wind [m/s] (barbs), geopotential height [m] (solid lines, every 30 m) and dew point depression [K] (dots) field at the 850 hPa level at the CAD peak time in (a) Cluster 1 (C1), (b) C2, (c) C3, (d) C4, (e) C5 and (f) C6.The grey area corresponds to the areas where the altitude is below the terrain.[Colour figure can be viewed at wileyonlinelibrary.com]

F
I G U R E 1 7 The distribution of wind speed [m/s] (shade) and geopotential height [m] (solid lines, every 120 m) at the 300 hPa level at 00:00 UTC 18 January 2016 (09:00 JST 18 January 2016; a case of CAD events).[Colour figure can be viewed at wileyonlinelibrary.com]