The North Pacific Storm-Track Suppression Explained From a Cyclone Life-Cycle Perspective

Surface cyclones that feed the part of the North Pacific storm track experiencing a midwinter suppression originate from three regions: the East China Sea (∼30◦N), the Kuroshio extension (∼35◦N), and downstream of Kamchatka (∼53◦N). In midwinter, in terms of cyclone numbers, Kuroshio (45%) and Kamchatka (40%) cyclones dominate in the region where eddy kinetic energy is suppressed, while the relevance of East China Sea cyclones increases from winter (15%) to spring (20%). The equatorward movement of the baroclinicity and the associated upper-level jet toward midwinter influences cyclones from 5 the three genesis regions in different ways. In January, Kamchatka cyclones are less numerous, less intense and their lifetime shortens; broadly consistent with the reduced baroclinicity in which they grow. The opposite is found for East China Sea cyclones, which in winter live longer, are more intense, and experience more frequently explosive deepening. The fraction of explosive East China Sea cyclones is particularly high in January when they benefit from the increased baroclinicity in their environment. Again, a different and more complex behavior is found for Kuroshio cyclones. In midwinter, their number 10 increases, but their lifetime decreases; on average they reach higher intensity in terms of minimum sea level pressure, but the fraction of explosively deepening cyclones reduces and the latitude where maximum growth occurs shifts equatorward. Therefore, the life cycle of Kuroshio cyclones seems to be accelerated in midwinter with a stronger and earlier but also shorter deepening phase followed by an earlier decay. Once they reach the latitude where eddy kinetic energy is suppressed in midwinter, their baroclinic conversion efficiency is strongly reduced. Together, this detailed cyclone life-cycle analysis reveals 15 that the North Pacific storm-track suppression in midwinter is related to fewer and weaker Kamchatka cyclones and to more equatorward intensifying and then more rapidly decaying Kuroshio cyclones. The less numerous cyclone branch from the East China Sea partially opposes the midwinter suppression.

(i) downstream of Kamchatka, (ii) over the Kuroshio extension, and (iii) over the East China Sea. So far, it is unclear if the suppression affects the cyclones from these genesis regions in a similar way.
In this study, we investigate midwinter changes in surface cyclone life cycles over the western North Pacific according to their genesis region. Surface cyclones are an important subcategory of the wide distribution of flow features collectively termed "eddies". Upper-level cyclonic eddies, some of them shallow, correspond to troughs. Once a trough interacts with a surface eddy, they mutually amplify and propagate in tandem poleward (Hoskins et al., 1985, Fig. 21). The combined system develops into a mature low-pressure system corresponding to a deep cyclonic eddy. With surface cyclone tracks we thus identify particularly strong cyclonic eddies that play an essential role for the overall storm track climatology. Over the North Pacific, we expect to find different life-cycle characteristics in midwinter compared to November and March, because in midwinter the cyclones typically form on the poleward flank of a strong subtropical jet, whereas in November and March they usually 70 develop on the equatorward flank of a more poleward located jet. Schemm and Schneider (2018) have already shown that for the entire North Pacific the lifetime of surface cyclones decreases. Here we study in detail all surface cyclones that affect the region of the midwinter suppression in the western North Pacific between October and April and quantify their frequency, lifetime, intensity, baroclinic conversion rates, and other characteristics according to their genesis region. This approach will serve to address the following questions:

75
-What is the relative contribution of different genesis regions to the surface cyclone frequency in the region affected by the midwinter suppression?
-Are there any differences in the character of the surface cyclones of different origin between midwinter and the shoulder months? For example, how does their number, lifetime and time to maximum intensity vary during the cold season?
-Is the suppression of the baroclinic conversion during midwinter equally strong for cyclones of different origin? 80 To answer these questions, we use an object-based surface cyclone tracking algorithm and evaluate baroclinic conversion rates obtained from bandpass-filtered data along individual cyclone tracks. With this approach, we combine two complementary perspectives on storm track dynamics.
Our study is organized as follows. In section 2 we introduce the used data and methods. In section 3 we describe the midwinter evolution of eddy kinetic energy (EKE) over the North Pacific and define specific target regions characterized by an 85 increase and decrease in EKE during winter, respectively. The surface cyclone tracks and the genesis regions of cyclones that propagate through the target regions are presented in section 4. section 5 presents a detailed analysis of changes in different life cycle characteristics, including maximum deepening rates. Baroclinic conversion along cyclone tracks of different origin are studied in section 6. We conclude our study in section 7.

90
The analysis period is October to April 1979-2018. All diagnostics rely on 6-hourly ERA-Interim data that are interpolated to a 1 • grid. ERA-Interim is publicly available for download via ECMWFs archive at https://apps.ecmwf.int/datasets/.

Surface-cyclone tracks and surface cyclogenesis
For the identification and tracking of surface cyclones, we make use of the algorithm introduced by Wernli and Schwierz (2006) and refined by Sprenger et al. (2017). The detection of surface cyclones is based on a contour search in the mean sea 95 level pressure (SLP) field at intervals of 0.5 hPa. To obtain a cyclone mask at each time step, all grid points inside the outermost closed contour, which must not exceed 7500 km in length, are labelled with 1, all others with 0. The obtained binary cyclone fields are used to compute cyclone frequencies. Cyclone centers are defined as the grid point with minimum SLP inside the outermost closed contour. The cyclone centers are tracked using 6-hourly cyclone center positions, and a track is accepted if it exists for a period of at least one day. The first time step along each track is defined as the genesis time step and the SLP 100 minimum defines the genesis location. The algorithm contributed to the cyclone identification and tracking intercomparison project of Neu et al. (2013).

Baroclinicity and baroclinic conversion
The background baroclinicity and the corresponding baroclinic conversion are defined based on tendency equations for eddy kinetic and available potential energy (Lorenz, 1955;Orlanski and Katzfey, 1991;Chang, 2001). A detailed derivation of both 105 tendency equations using high-pass-filtered input data is given in Schemm and Rivière (2019). The baroclinic conversion to eddy energy, the sum of eddy kinetic and eddy available potential energy, is the scalar product between the eddy heat flux 1 √ S θ v and the background baroclinicity − ∇θ √ S , where v denotes the high-pass-filtered horizontal wind, θ the low-pass-filtered potential temperature, and S the static stability 110 in pressure coordinates S = −h −1 ∂θ R ∂p . The reference potential temperature θ R is computed from monthly mean data and h denotes the scale height. Background baroclinicity is defined as the horizontal gradient of the low-pass-filtered potential temperature divided by the static stability, B = − ∇θ √ S . The background baroclinicity is closely related to the Eady growth rate (Lindzen and Farrell, 1980). The data used in this study is similar to that used in Schemm and Rivière (2019). For the low-pass (2018) for a more detailed discussion of month-to-month EKE variations]. From February to March, EKE increases again in a meridionally confined band between ∼40 • and 50 • N and decreases equatorward of ∼40 • . Based on theses patterns of 125 intraseasonal changes in EKE, we select in the following surface cyclones that propagate through one of the regions with a midwinter (December to January) EKE decline or increase, respectively. The two target regions are indicated as gray boxes in Fig. 1 and are located at the entrance of the storm track (EKE maximum). One of our aims is to assess if and how the EKE tendency dipole seen in Fig. 1b is linked to characteristics of the surface cyclone tracks. We focus in particular on cyclones that propagate through the northern target region, which exhibits a decline in EKE during midwinter and is therefore essential for 130 understanding the midwinter suppression phenomenon. 4 Surface-cyclone view on the North Pacific storm track in midwinter

Surface cyclogenesis
Consideration is first given to cyclogenesis associated with surface cyclone tracks that propagate through the northern target region, i.e., the region where EKE decreases during midwinter (Fig. 1b). In November, these surface cyclones originate from

140
The southern target region, i.e., the region where EKE increases during midwinter (Fig. 1b), is fed exclusively by surface cyclones with genesis over the Kuroshio extension during November (Fig. 2b). In January, a second but weaker cyclogenesis region emerges southwest of Japan over the East China Sea. In March (Fig. 2f), the two genesis regions exhibit similar cyclogenesis frequencies and contribute equally to the cyclone tracks in the southern target region. Notably, also for this region there is no midwinter suppression in the cyclone frequency. This result is in agreement with the findings of Schemm and Schneider 145 (2018) that the suppression is connected to a reduction in the cyclone intensities rather than their frequencies.
Surface cyclones over the North Pacific are known to be triggered by two upper-level seeding branches: a northern branch over Siberia and a southern branch along the subtropical jet across southern Asia (Chang, 2005). We briefly report about the upper-level seeding associated with the three preferred regions of surface cyclogenesis (  Figure S1). Cyclogenesis over the East China Sea is associated with the southern seeding branch and an upper-level trough downstream of the genesis location (not shown). This behavior was already recognized by Chang (2005), who noted that "cyclogenesis for these cases is probably not triggered by the [upper-level] 155 wave packet" (Chang, 2005(Chang, , p.1998. The genesis of these cyclones seems to be connected to a bottom-up development, as is the case for diabatic Rossby waves (e.g., Boettcher and Wernli, 2013).

Relative surface cyclone frequencies
In the previous section, we showed that surface cyclogenesis downstream of Kamchatka, over the Kuroshio extension and, in late winter, over the East China Sea, contribute to the surface cyclone tracks in the northern target region, where EKE exhibits a 160 midwinter suppression. The latter two enter the target region from the south, while Kamchatka cyclone have their genesis inside 7 https://doi.org/10.5194/wcd-2020-33 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License.

Kamtschatka cyclogenesis
Kuroshio cyclogenesis November November January January

March March
Northern Target Region the target region. To study the relative importance of the different genesis regions for the suppression, we group the cyclone tracks that affect the northern target region into two categories. The first contains tracks with genesis over the Kuroshio or the East China Sea, which enter the target region from the south. The second category contains tracks with genesis near Kamchatka.
The cyclone frequency fields for the two categories are then divided by the total cyclone frequency field obtained from all tracks 165 that propagate through the northern target region. These relative contributions are shown for November, January and March in Fig. 3. During all months, Kamchatka cyclones contribute up to 40-50% along the poleward side of the target region and only 10-20% along the equatorward side. The relative contribution of cyclone tracks entering the target region from the south is 80-90% along the equatorward side and 50-60% along the poleward side of the target region. This further corroborates that the midwinter suppression is related to a change in the characteristics of these cyclones and not in their frequencies. In the 170 following, we study statistics of cyclone characteristics for the different genesis regions in greater detail. In the previous section, we showed that the northwestern Pacific surface storm track is fed by three preferred cyclogenesis regions: (i) East China Sea, (ii) Kuroshio and (iii) Kamchatka; and neither of the three exhibits a midwinter suppression in terms of cyclogenesis frequency (Fig. 2). Next, we investigate several life cycle characteristics. As in the previous sections, our 175 focus is on the northern target region (gray box in Fig. 1), where EKE decreases during midwinter.
As mentioned in the previous section, the total number of cyclones and the number of cyclones per day in the northern target region (first column in Tab  22 % to the total cyclone number (Tab. 1). Kuroshio cyclones are also most intense in January, but the change in minimum SLP between November and January is small. The equatorward movement of the baroclinic zone in midwinter seems to be beneficial for the intensification of East China Sea cyclones in January. Kamchatka cyclones, however, become less intense, a result that can, at least qualitatively, be expected from the equatorward retreat of the baroclinic zone. For Kuroshio cyclones the situation is complex. While there is a weak reduction in the minimum SLP from November to January, the fraction of life cycles 205 that satisfy the Sanders and Gyakum (1980) criterion for explosive deepening, known as "bomb cyclogenesis" 1 , is reduced in midwinter. In contrast, for the East China Sea cyclones the bomb fraction peaks in January, which is in agreement with the increase in the background baroclinicity. Kuroshio cyclones thus appear to deepen rapidly in a short time period (Tab. 3), in agreement with the midwinter peak in baroclinicity, but after reaching their strongest intensity they also decay rapidly, as indicated by the shortest life time during midwinter (Tab. 2). This suggests that for Kuroshio cyclones the peak in baroclinic 210 conversion occurs earlier during their life cycle and short deepening is also more intense, but thereafter they move relatively soon out of the zone of high baroclinicity, resulting in less intense cyclones at higher latitudes. This could also explain the dipole pattern in EKE shown in Fig. 1b, because in January approximately 50% of all cyclones tracks that propagate through the northern target region also propagate through the southern target region. Thus, to better understand the intensification, in the next section we investigate baroclinic conversion first over the two target regions from an Eulerian viewpoint and afterward 215 along the different tracks from a quasi-Lagrangian viewpoint. 6 Baroclinic conversion and its relationship with surface cyclone tracks

Baroclinic conversion over target regions (Eulerian perspective)
EKE has a baroclinic and barotropic source, and both are known to be affected by midwinter suppression (Schemm and Schneider, 2018). In general, however, the dominant source of EKE is baroclinic conversion. In the following, we first diagnose in November and March. However, the monthly differences in the southern target region are smaller than in the northern target region. Furthermore, the difference between January and March is less clear in the southern target region.
The above findings suggest that the characteristics or the synoptic systems that propagate through the two target regions shown in Fig. 1 clearly differ between the three months. In January, the associated baroclinic conversion is lower than in of all cyclones that enter the northern target region from the south, have an earlier baroclinic conversion peak (in the southern target region) and reduced baroclinic conversion later in the northern target region. As we show below, the first scenario ap-240 plies to Kamchatka cyclones and the second one to Kuroshio cyclones. But first, we explore how baroclinic conversion changes during days when a cyclone propagates through the northern target region.

Baroclinic conversion in the northern target region associated with surface cyclones
In order to quantify the contribution of surface cyclones to the climatological monthly mean baroclinic conversion in the northern region, we split all days into cyclone and non-cyclone days using the surface cyclone tracks. Thereby, we essentially 245 separate deep cyclonic eddies from shallow upper-level eddies (e.g., troughs or ridges without surface low-or high-pressure systems) and from deep anticyclonic eddies. Technically, a surface cyclone track may propagate outside of the northern target region but nevertheless affect the baroclinic conversion inside the target region. We therefore define a cyclone day as a time step when 25 % of the northern target region is covered by a cyclone mask (see section 2 for details). This results in about 50 % cyclone and 50 % non-cyclone days.

250
affects baroclinic conversion during cyclone and non-cyclone days. Yet, the two distributions differ significantly from each other, in particular in January during midwinter suppression. To test this statistically, we compute 10'000 distributions, each of which consists of randomly selected cyclone and non-cyclone days with replacement. Each randomized distribution is of equal size as there are days in the original cyclone-day distribution. For each randomized distribution, we compute the mean baroclinic conversion, and from the 10'000 mean values the 97.5th and 2.5th percentiles, which are shown as red confidence 260 intervals in Fig. 5. The mean baroclinic conversion values of the cyclone (non-cyclone) day distribution is above (below) the 97.5th (2.5th) confidence intervals in each month. We therefore conclude that the two distributions significantly differ from each other and from a randomized selection. Based on Fig. 5, we conclude that baroclinic conversion in the target region is reduced in midwinter both during cyclone days and non-cyclone days. However, since the baroclinic conversion during cyclone days is higher than during non-cyclone days, the cyclone days contribute much more to the total conversion in the Pacific storm 265 track and accordingly to the reduction in the conversion in midwinter. It is therefore reasonable to focus now only on cyclone days and the associated tracks. In a next step, we investigate the cyclone tracks that enter the northern target region from the south and compare the baroclinic conversion along these tracks before and after entrance.
6.3 Baroclinic conversion along cyclone tracks outside and inside the northern target region (Lagrangian perspective)

270
Cyclones that feed the northern target region are Kamchatka cyclones, with genesis inside the target region, and Kuroshio and East China Sea cyclones, which enter the target region from the south. We therefore group all time steps along Kuroshio and East China Sea cyclone tracks into two periods, before and after entering the northern target region. The idea is to see whether the maximum in baroclinic conversion occurs earlier during the life cycle in January, as suggested in section 5 and based on Tab. 3, and therefore outside the northern target region. In the following, we discuss box-and-whisker diagrams of baroclinic 275 conversion and of the background baroclinicity (Fig. 6) separately for cyclones that enter the target region from the south (Kuroshio and East China Sea cyclones) for time steps before and after entering the target region, and for Kamchatka cyclones, which reside inside the target region. For all cyclones, baroclinic conversion is averaged within a 1000 km radius around the cyclone center.
Before entering the northern target region from the south, baroclinic conversion along cyclone tracks is larger in January 280 than in November and March (black boxes in Fig. 6). The distribution of baroclinic conversion outside of the target region exhibits a seasonal cycle that is qualitatively in agreement with the seasonal cycle of the mean baroclinicity equatorward of the target region. The difference between January and March is small, which is a result of the increasing influence of East  China Sea cyclone tracks towards late winter and early spring. East China Sea cyclones deepen on average more rapidly than Kuroshio cyclones (Tab. 4) and they are particularly frequent in March ( Fig. 2 and Table 1).

285
After entering the northern target region, baroclinic conversion associated with Kuroshio and East China Sea cyclones is reduced in January compared to November and March, which reflects the midwinter suppression (gray boxes in Fig. 6).
Kuroshio cyclones spend most of their life cycle in the northern target region (percentage of time steps in each category is shown below each box in Fig. 6). In January, the fraction of time steps outside the northern target region is lower than in November and March suggesting that in January cyclones propagate faster poleward and hence out of the zone of high baroclinicity.

290
The strong reduction of baroclinic conversion in the northern target region in January occurs despite the fact that the mean baroclinicity along the tracks of Kuroshio cyclones is only marginally reduced compared to other months (horizontal red bars on top of the gray box-and-whiskers in Fig. 6). This confirms the conversion budget discussed in Schemm and Rivière (2019), who argued that the reduced conversion in this region primarily results from a reduction of the eddy conversion efficiency, with only minor contributions form a change in mean baroclinicity.

295
In summary, maximum baroclinic conversion along the surface cyclone tracks with genesis over the Kuroshio extension is climatologically largest in January, but occurs equatorward of the northern target region and therefore earlier during the cyclone life cycle. The larger conversion is thus in agreement with an overall higher mean baroclinicity over the North Pacific, but the

Conclusions
This study presents a systematic analysis of the characteristics of cyclone life cycles over the North Pacific, with a particular focus on surface cyclones that propagate through the northwestern Pacific, where EKE decreases during midwinter (referred to as northern target region shown in Fig. 1). The goal of this study is to enrich the existing literature on the midwinter suppression change in the subtropical jet regime, which typically occurs during midwinter in the western North Pacific.
The surface cyclone tracks feeding the storm track in the northwestern Pacific originate from three preferred regions: (i) downstream of Kamchatka, (ii) over the Kuroshio extension and (iii) over the East China Sea (Fig. 2) (Chang, 2005).
Our key findings can be summarized as follows. The equatorward movement of the baroclinic zone in midwinter affects the life cycles of cyclones from all three genesis regions, but in a different way: -Kamchatka cyclones develop in midwinter in a region of reduced baroclinicity. Compared to November, their lifetime decreases, the time to maximum deepening since genesis reduces and they become less intense. They contribute by about 320 40 % to the total cyclone number in winter over the northwestern Pacific, where EKE is suppressed. The weakening of Kamchatka cyclones is thus a crucial contribution to the suppression. Interestingly, despite the reduced baroclinicity in January, the number of Kamchatka cyclones is not reduced in midwinter. Kamchatka cyclones do not re-intensify during March, thus they do not benefit from the poleward movement of the baroclinic zone in spring.
-East China Sea, benefit from the equatorward movement of the baroclinic zone in midwinter. Compared to November, 325 they become more intense, the fraction of explosively deepening cyclones increases and their lifetime extends. They become weaker in March, but the fraction of explosively deepening life cycles remains higher than for Kuroshio and Kamchatka cyclones. In addition, their lifetime is longer in March compared to January. In March, East China Sea cyclones contribute by nearly 22% to the total cyclone number over the northwestern Pacific, while in fall and winter their contribution is approximately 15%. Thus, they seem to play a role in the re-intensification of the storm track during 330 spring.
-The changes in the life cycles of Kuroshio cyclones are the most complex, but understanding these changes is crucial, because Kuroshio cyclones contribute strongest to the total cyclone number in the northern target region in midwinter (45% ). Compared to the shoulder months, in January the lifetime of Kuroshio cyclones and the time to maximum deepening are shortest. The fraction of explosively deepening cyclones first reduces from November to January but then 335 remains at similar levels until March. Highest values in baroclinic conversion are found during midwinter, but these occur at lower latitudes, south of the northern target region, and they are sustained for a reduced number of time steps.
In terms of minimum sea level pressure, Kuroshio cyclones are most intense in January.
Overall, it seems as if during midwinter the life cycle of a Kuroshio cyclone is best characterized by a short and intense early deepening, in agreement with the higher baroclinicity, followed by a fast decay and poleward propagation away from the more 340 equatorward located baroclinic zone. According to this interpretation, we observe an acceleration of the Kuroshio life cycle during midwinter. This interpretation is in agreement with the idea of a reduced baroclinic conversion efficiency because the efficiency is dictated by the vertical tilt of a cyclone (Schemm and Rivière, 2019). Acceleration of the life cycle with intense early growth results in cyclones that acquire a rather inefficient vertical tilt earlier in the life cycle. Kuroshio cyclones are thus in different months in different stages of their life cycle at similar latitudes. The stronger but earlier deepening followed by an 345 earlier decay is the quasi-Lagrangian perspective on the equatorward shift seen in EKE from an Eulerian perspective (Fig. 1b).

Caveats
Our results are based on a single object-based cyclone detection scheme. It is known that cyclone tracks are sensitive to the identification and tracking scheme Neu et al. (2013), this holds also true for the genesis location. While they typically agree on deep systems, a higher sensitive must be expected for shallow systems, as is the case for Kamchatka cyclones in January.

350
Further, we ignore short-lived systems with a lifetime of less than 24 hours. Such systems might become more frequent in midwinter because the lifetime of all systems is reduced in January. Baroclinic conversion occurs also in the absence of surface cyclones, for example by the propagation of an upper-level trough, like the second baroclinic conversion peak in Fig. 4.
Baroclinic conversion during non-cyclone days is also affected by the midwinter suppression and this reduction is not explained by our study.

355
Author contributions. All three authors contributed to the discussion and final interpretation of the results. HB and SeS performed the analyses. SeS wrote, supported by HB and HW, the publication.
Competing interests. The authors declare no competing interests.