For nearly a century, the study of atmospheric dynamics in the midlatitudes has presented dichotomic perspectives on one of its focal points: the birth and life cycle of cyclones. In particular, the role of fronts has driven much of the historical discourse on cyclogenesis. In the 1910s–20s, the Bergen School of Meteorology postulated that cyclogenesis occurs on a preexisting front. This concept was later replaced by the baroclinic instability paradigm, which describes the development of a surface front as a consequence of the growing cyclone rather than its cause. However, there is ample observational evidence for cyclogenesis on well-marked fronts (frontal-wave cyclones) as well as for cyclogenesis in the absence of fronts in broader baroclinic zones. Thus, after a century of research on the link between extratropical cyclones and fronts, this study has the objective of climatologically quantifying their relationship. By combining identification schemes for cyclones and fronts, the fraction of cyclones with attendant fronts is quantified at all times during the cyclones’ life cycle. The storm-track regions over the North Atlantic are dominated by cyclones that form on preexisting fronts. Over the North Pacific, the result more strongly depends on the front definition. Cyclones that acquire their fronts during the life cycle dominate over the continents and in the Mediterranean. Further, cyclones that develop attendant fronts during their life cycle typically do so around the time they attain maximum intensity. At the time of cyclolysis, at least 40% of all cyclones are still associated with a front. The number of occluded fronts at lysis has not been considered.
Which comes first: the extratropical cyclone or the front? An attempt is presented to disentangle the historic “chicken or egg” relationship that puzzles meteorologists to this day.
The acceptance of two contrasting interpretations of cyclogenesis over the last century “has been antiphonal with the two streams alternating in holding ascendency” (Davies 1997, p. 261). The first perspective emerged as a mainstream theory of cyclogenesis after the Bergen School of Meteorology promoted it in the 1910s–20s. They attributed cyclogenesis to the intrinsic instability of the hemispheric-spanning “polar front” upon which frontal cyclones develop as wavelike disturbances (Bjerknes and Solberg 1922). The key elements of today’s weather charts and visualizations of extratropical cyclones in introductory textbooks build upon the concepts developed in the 1910s, demonstrating the success of the Bergen School’s concept. A contrasting view emerged in the middle of the last century when Hsieh (1949), who studied a North American cyclone, concluded that “this type of surface cyclogenesis is evidently different from the usual type associated with waves on a surface front” (Hsieh 1949, p. 401).
According to the Bergen School’s cyclone model, during the life cycle of the surface cyclone, warm air pushes poleward while cold air advances toward the equator (Bjerknes 1919). The two air masses are separated by narrow transition zones, that is, the warm and cold fronts. Later, the highly idealized concept of the polar front as a discontinuity was replaced by a narrow transition zone (Bjerknes and Palmén 1937). According to this polar front paradigm, the surface front is considered to be the origin of the cyclone development, which occurs as an instability of the preexisting front. The fundamental question that emerges from this paradigm is why and when do fronts become unstable (e.g., Solberg 1928; Kotschin 1932; Orlanski 1968; Kasahara and Rao 1972; Sinton and Mechoso 1984; Sinton and Heise 1993; Parker 1998; Patoux et al. 2005). Schär and Davies (1990) and Joly and Thorpe (1990) identified an intrinsic frontal instability of surface fronts that are either associated with a warm-air precursor or a positive low-level potential vorticity (PV) anomaly, respectively. These studies provided, decades after the Bergen School presented its concepts, a theoretical basis for the primarily observation-based frontal-cyclone model. For more comprehensive literature reviews, the reader may consult Parker (1998) and Davies (1999).
The shift away from the frontal discontinuity as the origin of cyclogenesis was supported by the theory of baroclinic instability (Charney 1947; Eady 1949), which explained the generation of extratropical cyclones in initially broad baroclinic zones without well-marked surface fronts. Early nonlinear simulations of baroclinic instability (Phillips 1956; Mudrick 1974) revealed the generation of frontlike features during cyclone intensification and further supported the baroclinic instability concept, in which surface cold and warm fronts are a consequence of cyclogenesis and not the cause. Palmén (1951, p. 616), though investigating processes at the jet stream level, noted, “Obviously, a well-marked surface front is not so essential for the development [of the cyclone] as was generally assumed formerly.”
Our current view on extratropical cyclones is still characterized by this duality of concepts. Ample observational evidence and idealized studies support that, for certain cyclones, a preexisting front is the seat of cyclogenesis, and the instability appears to be released when an upper-level trough approaches the front (e.g., Petterssen 1955; Petterssen et al. 1962; Thorncroft and Hoskins 1990; Thorncroft et al. 1993; Appenzeller et al. 1996), whereas for other cyclones, the intense surface fronts develop only during cyclone intensification in an initially broad baroclinic zone (e.g., Mudrick 1974; Hobbs et al. 1996; Lackmann et al. 1997; Wernli et al. 1999; Thomas and Martin 2007; Schemm et al. 2013). The former category is sometimes referred to as “secondary cyclones” if the cyclones form on fronts trailing previous (primary) cyclones embedded in larger-scale baroclinic zones (Eliassen 1966). In the analytical model of Joly and Thorpe (1990), cyclones forming on a preexisting front deepen during their incipient growth phase because of barotropic conversion and only later transform into a baroclinic system.
Most studies on frontogenesis have noted that fronts should be regarded as transient rather than semipermanent phenomena (e.g., Sawyer 1956; Eliassen 1962). They proposed a two-step character of frontogenesis where, first, geostrophic deformation increases the temperature gradients in the growing baroclinic wave and, second, the ageostrophic frontal circulation strongly amplifies this gradient (Hoskins and Bretherton 1972). For example, Sanders (1999b) provides observational evidence for the transient character of fronts.
Clearly, some cyclones are already associated with fronts at the time of cyclogenesis, whereas cyclones growing in a broad baroclinic zone acquire their fronts during the life cycle. Thus, we distinguish in this study the following two categories of cyclones:
life cycles with attendant fronts at genesis (initial-front cyclones);
life cycles without fronts at genesis but with late-forming fronts (late-front cyclones).
Furthermore, we identify no-front cyclones, which have no front at genesis nor during their life cycle. Using climatologies of surface cyclones and fronts based on the European Centre for Medium-Range Weather Forecasts (ECMWF) interim reanalysis (ERA-Interim) dataset (Dee et al. 2011), these three extratropical cyclone categories are identified, and based on the obtained data, we attempt to answer the primary question of this study: When during their life cycle are extratropical cyclones attended by fronts?
A few more comments complete this conceptual overview. Arguably, fronts are not the only ingredient that can play an important role during extratropical cyclogenesis. It is widely accepted that extratropical cyclogenesis is generally associated with an upper-level flow anomaly (e.g., Sutcliffe 1947) or fostered by moist diabatic processes (e.g., Graf et al. 2017). Numerous previous studies illuminated the role of flow anomalies at the tropopause level in shaping the life cycle of extratropical cyclones [see Uccellini (1990) for a review]. Important early studies on this topic include contributions by Ficker (1920a,b), concisely summarized by Davies (2010), and Petterssen et al. (Petterssen 1955; Petterssen et al. 1962; Petterssen and Smebye 1971). The focus of this study, however, is exclusively on the cyclones’ surface structure and the historic “chicken or egg” relationship between (surface or near-surface) fronts and extratropical cyclones, without quantifying the importance of processes at the jet stream level.
The purpose of this study requires automated identification procedures for cyclones and fronts. To detect large-scale fronts in reanalysis data, different approaches have been proposed (often developed in different parts of the world and inspired by the researchers’ experience in the local meteorology), and the next section discusses their shortcomings and advantages.
DISCUSSION: HOW TO DETECT FRONTS AUTOMATICALLY?
Meteorologists have yet to settle on a single definition of fronts, and contrasting opinions have led to the development of different concepts of how to correctly locate a front on a weather map (e.g., Renard and Clarke 1965; Clarke and Renard 1966; Mass 1991; Uccellini et al. 1992; Sanders and Doswell 1995; Hewson 1998; Sanders 1999a; McCann and Whistler 2001). This disagreement complicates the development of an automated front identification scheme, and the discussion inherent to frontal analysis is concisely summarized in Uccellini et al. (1992). However, an automated detection of fronts is mandatory because we aim to process more than 35 years of 6-hourly data in the Northern Hemisphere.
Much of the discourse on frontal analyses centers on two basic questions: which variable to analyze and on which vertical level? More fundamentally, what in dynamical terms is understood as a front, and what are the consequences of such a definition for automated detection? Is it a tilted isentrope, a horizontal density gradient, or a combination of different variables? While most meteorologists will agree on a list of essential front characteristics (e.g., enhanced baroclinicity, anomalous relative vorticity, and static stability), historic debate has shown that we must accept a certain degree of freedom in the practical definition of a front.
The work of Bjerknes and Solberg (1922) suggests that a front may broadly be defined as a discontinuity in the properties between air masses of different origins. In fact, they describe a cyclone as consisting “of two essentially different air masses, the one of cold and the other of warm origin” (Bjerknes and Solberg 1922, p. 4). This “Lagrangian flavored” perspective calls for the analysis of a variable that is conserved along parcel motion and hence is able to properly separate air masses of different origins. According to earlier studies, a useful choice for such a variable is equivalent potential temperature θe, which, in contrast to potential temperature θ or virtual potential temperature θυ, is conserved in a reversible moist adiabatic process. Renard and Clarke (1965) recognized this in their early work on automated front identification. Inspired by the studies of Godson (1951) and Taljaard et al. (1961), they noted that “frontal-zones are considered as first-order thermal and moisture discontinuities” (Renard and Clarke 1965, p. 547). However, they dismissed the role of moisture and proceeded with θ because of “certain deficiencies in adequately depicting hemispheric moisture fields” (Renard and Clarke 1965, p. 551). Today, more than 50 years later, this poses much less of a challenge, and several studies have discussed or used θe (or wet-bulb θ) in their frontal analyses (e.g., Cahir and Lottes 1982; Steinacker 1992; Hewson 1997, 1998, 2009; Hewson and Titley 2010; Jenkner et al. 2010; Berry et al. 2011; Hope et al. 2014; Schemm et al. 2015). In fact, 850-hPa θe (or wet-bulb θ) is used in operational practice at the Met Office (UKMO; Hewson 1998; Hewson and Titley 2010; Mulqueen and Schultz 2015) and the German Weather Service [Deutscher Wetterdienst (DWD); Kurz 1998] to assist meteorologists with drawing surface fronts.
One aspect that is often regarded as a shortcoming of θe is that its gradient can be dominated by moisture contrasts instead of temperature contrasts, which add “little to density contrasts” (Sanders 1999a, p. 946). In such a case, a θe front no longer necessarily indicates a zone of enhanced baroclinicity. However, this can be regarded as an advantage, for example, when it comes to forecasting deep convection (Sanders and Doswell 1995). Lackmann (2011) presented another example that reflects the complications inherent to frontal analysis: Differential heating on a preexisting cold front causes the surface temperature of dry postfrontal air to warm more rapidly than humid, cloudy prefrontal air. As a result, this causes a reduction in the cross-frontal density contrast that may even vanish during the day. However, a zone of important weather activity remains even in the absence of a marked density gradient (wind convergence, precipitation, and convection). Lackmann (2011, p. 132) concluded that “if we can agree that this feature represents a true front,” θe will identify a front throughout the day. A rigorous density definition, in contrast, would identify this front only at certain times.
In agreement with this perspective of fronts as large-scale airmass boundaries, θe is typically evaluated on an elevated level that has to be a good compromise between “being sufficiently close to the ground for nonvertical fronts to be located with reasonable accuracy, and being sufficiently far away [from the surface] not to be dominated by differential surface heat and moisture exchanges” (Hewson 1998, p. 52). Petterssen (1940), who was worried about losing the large-scale character of air masses because of various diabatic surface processes, similarly rejected surface variables for (large scale) frontal analyses, as they are “often neither representative nor conservative” (Petterssen 1940, p. 7).
Sanders and Doswell (1995) and Sanders (1999a) advocated the use of surface θ for the analysis of fronts. One motivation was the work of Sanders (1955), who showed that fronts are strongest at the surface and weaker at higher levels. Sanders and Doswell (1995) were mainly concerned with the mesoscale development over the continental United States rather than the large-scale movement of cyclones. The lack of agreement on surface frontal analyses (Young and Fritsch 1989; Mass 1991; Uccellini et al. 1992) is, according to Sanders and Doswell (1995), the result of a widespread disregard of surface θ. They argued that “what formerly was considered nonrepresentative [they refer here to Petterssen (1940)], may now be viewed as representative of a mesoscale system” and that “the bases for declining to deal with surface temperature analyses, therefore, do not now seem to be valid” (Sanders and Doswell 1995, p. 505). Sanders and Hoffman (2002) therefore used station-based θ in their frontal analyses over the United States. The recommendations by Sanders reflect years of experience in analyzing mesoscale and submesoscale fronts over the United States [see also Bosart and Bluestein (2008)]. Since then, other studies have made use of the approach suggested by Sanders et al. (e.g., Payer et al. 2011), and their recommendations have been widely and successfully used in operational forecasting, mainly within the United States. To account for the role of moisture in modifying buoyancy, Sanders (1999a) suggested the use of θυ but did not attempt to use it. One advantage of θυ is that it is a “dry variable” and can be used in dry analytical formulations of frontogenesis. However, in case of latent heating, it is not conserved. In contrast, θe is conserved, but it is a function of the mixing ratio and therefore does not align with a rigorous density definition of a front (Glickman 2000). Baldwin and Logsdon (2011) found that (for automated methods) “experimentation with θυ has shown very similar results to that of θ.” Because θe identifies “distinct boundaries and generally has a much stronger signal than any of the other variables tested,” it “provides a superior performance as compared to simply potential temperature alone” (Baldwin and Logsdon 2011, p. 6).
It seems that the controversial discussion regarding the detection of fronts using thermal fields and levels depends on whether we approach it from a large-scale or mesoscale perspective. Hewson (1998) argued that “it seems preferable to refer to such regions [of enhanced gradients at the surface] as mesoscale boundaries as in Young and Fritsch (1989), and not to use them for identifying ‘surface fronts’ per se…Many of the 1000 hPa fronts fall into the mesoscale boundary category, as they are unrelated to fronts at higher levels” (Hewson 1998, p. 52). This argument is in line with that of Petterssen (1940). However, according to Sanders and Doswell (1995), such mesoscale boundaries can serve as the origin of severe weather and mesoscale cyclones; excluding them from frontal analyses requires proper justification. The arguments of, for example, Hewson (1998) and Petterssen (1940) are acknowledged but also questioned by Sanders (1999a), who noted that “there is an unstated inference that in order to be significant, a temperature contrast must extend through some minimum depth” (Sanders 1999a, p. 945). The appropriate level and thermal variable therefore depends largely on the specific research question. We close this survey by noting that several other methods exist to detect fronts, based on, for example, the 2D frontogenesis function (Petterssen 1936; Keyser et al. 1988), the wind field (Simmonds et al. 2012), or a combination of temperature gradients and vorticity (Solman and Orlanski 2010).
Based on the arguments and the historic discourse presented above, we decided to use two methods to detect fronts on their standard levels and to interpret the results without prioritizing one method over the other. In short, the first method is based on θe at 850 hPa following the strategy outlined by Hewson (1998) with some refinements made by Jenkner et al. (2010). The second method follows the recommendation by Sanders and Hoffman (2002) and is based on 2-m θ. We recognize that the first identification scheme does not comply with a rigorous density definition but, given the discussion above, regard both methods as pragmatic and well justified. As it will turn out, the answer to our focal question raised in the title is only weakly sensitive to the underlying front definition.
DETECTING FRONTS AND CYCLONES: AN EXAMPLE.
The required analysis steps and specific details of the used algorithms are best explained when honing in on a case example. At 0000 UTC 9 January 2010, two cyclones shape the surface sea level pressure (SLP) distribution over the North Atlantic (Fig. 1): a mature cyclone east of Newfoundland and a second one south of Nova Scotia. According to the cyclone detection scheme, the latter is a cyclogenesis event. More precisely, cyclogenesis is detected at 41°N, 62°W with a minimum SLP of 992 hPa in the National Oceanic and Atmospheric Administration (NOAA) surface analysis (996 hPa in ERA-Interim).
On the surface chart, the forecaster identified a cold and a shorter warm front attending the western cyclone (Fig. 1). Downstream, the fronts trailing the mature cyclone (58°N, 40°W) are partly drawn as occluded. South of the triple point, the cold front extends southwestward to subtropical latitudes and the warm front southeastward to 30°N. An additional feature associated with this cyclone is a surface trough northeast of its center (thin yellow dashed line). Farther downstream, a north–south-oriented stationary front is marked on the chart.
Our cyclone detection scheme computes a 2D cyclone mask (Fig. 2a; orange opaque area), which is defined as the area inside the outermost closed SLP contour (in this example, 1,000.5 hPa). The cyclone tracking follows the minimum SLP using 6-hourly time steps (the track of the cyclone that formed south of Nova Scotia is indicated by black asterisks in Fig. 2a). Along the cyclone track, we search for an overlap of the 2D cyclone masks with a 1D front object. Let us see how the front identification methods perform in this case.
The first method identifies frontal zones, which are areas of enhanced [>4 K (100 km)−1] θe gradients at 850 hPa (Fig. 2a; blue contours). Inside this frontal zone, a frontal line (solid black) is drawn using the thermal front parameter, and for cosmetic reasons, the resulting line is smoothed using a cubic spline interpolation.1 This front line is designed to indicate the existence of a front (i.e., it is a front marker) and to test for an overlap between the frontal zone (blue contours in Fig. 2a) and the cyclone area (orange opaque area in Fig. 2a) but not to identify the exact position of a front as done on surface weather maps. The key to answering our question is the overlap between the 2D cyclone area with the front. Because the frontal zones are typically narrow compared to the diameter of the 2D cyclone mask, the precise location of the frontal line inside the frontal zone is noncritical. Overall, the 850-hPa θe method (Fig. 2a) identifies the frontal regions similarly to the manual chart (Fig. 1). In particular, the cold front associated with the emerging low and the cold front trailing the eastern cyclone are well captured. The surface trough is identified as a front, while the occluded front is not captured. The difficulty in identifying occlusions is a characteristic of all methods we looked at. Most importantly for the purpose of this study, cyclone “G” forming south of Nova Scotia at this time is, because of its overlap with a front, classified as an initial-front cyclone, in agreement with the NOAA surface analysis (Fig. 1).
The cold front attending the emerging cyclone (G in Fig. 2a) is also well captured if we base the identification on 2-m θ gradients (Fig. 2b), and therefore also with the 2-m θ front approach, this cyclone is attended by a front at genesis. The frontal zone is detected using a threshold of 2.5 K (100 km)−1, which is weaker compared to the recommendation for 2-m θ fronts by Sanders and Hoffman (2002; for choice of the parameters, see also appendix “Technical details—Front and cyclone detection schemes and parameters”). Gradients of 2-m θ weaken to the east of this cyclone, and the frontal zone is fragmented between the two cyclones (blue contour in Fig. 2b). The trailing cold front of the mature system “L” (cf. Fig. 2a) is not identified. This is also the case if θυ is used instead of θ and even if the minimum threshold is reduced to 1.5 K (100 km)−1. Of course, an experienced forecaster will take several variables into account before placing a front on a chart. In this case, for example, the 2-m dewpoint temperature could help to identify the warm sector of the cyclone (green contour in Fig. 2b). But it is rather difficult to translate years of forecasting experience (and intuition) into an automated method. In summary, cyclone L is attended by a front at this particular time when using the 850-hPa θe method but not when using the 2-m θ method, while there is agreement on the fronts attending cyclone G.
Finally, we also have a brief look at θ-based fronts at 850 hPa when using the same thresholds as for the 2-m θ fronts [>2.5 K (100 km)−1]. The cold front of cyclone G is detected similarly as at the surface (Fig. 2c). Downstream over the Atlantic, a fraction of the stationary front is also identified. The cold front trailing the mature system is however missed, mainly because θ gradients at 850 hPa are below the threshold of 2.5 K (100 km)−1. Yet, θ gradients with values of approximately 1 K (100 km)−1 (thinner blue contours in Fig. 2c) are apparent, in particular southeast of the triple point as indicated on the surface chart (cf. Fig. 1). A cold front signature is vaguely discernible. Accordingly, θ on 850 hPa could detect the mature cyclone's trailing cold front only if we reduced the threshold to 1 K (100 km)−1. However, we hesitate to reduce the front threshold to such a low value, since Sanders and Hoffman (2002) recommended using 3.5 K (100 km)−1 for moderate fronts near the surface.
The fourth option, that is, θe-based fronts at 2 m, results in frontal gradients that are, as expected, overly dominated by humidity gradients (not shown). Therefore, we decided to perform our climatological analysis with two common choices presented in the literature, based on 850-hPa θe and 2-m θ, respectively. Both methods further remove stationary thermal boundaries with a minimum advection threshold; otherwise, thermal methods are flawed along steep topography, for example, south of Greenland or along the Norwegian coast. The results presented below are, in particular over the oceans, only marginally influenced by this removal. The relative change due to the removal of stationary boundaries is largest only in isolated regions in the vicinity of high topography, for example, the Altai ranges. Further, fronts need to extend at least 500 km. For further technical details, see appendix “Technical details—Front and cyclone detection schemes and parameters.” In the next section, results are shown from applying both methods to an ERA-Interim cyclone climatology for the winters 1979–2014 (with about 61,000 cyclone tracks).
GEOGRAPHICAL OCCURRENCE AND RELATIVE FRACTIONS.
Climatologies based on 850-hPa θe fronts.
The three cyclone categories populate different regions in the Northern Hemisphere. Initial-front cyclones have the highest relative frequencies in the storm-track region over the North Atlantic and North Pacific (Fig. 3a). The preferred cyclogenesis regions for this category are over the Gulf Stream, along the southern tip of Greenland, and above the Kuroshio. The absolute cyclone frequencies are highest in the central North Atlantic and North Pacific. In contrast, late-front cyclones primarily dominate over continental North America, the Mediterranean, East Asia, and over the Sea of Japan (Fig. 3b). The absolute frequencies of late-front cyclones are highest in the central North Atlantic and North Pacific, over northeastern Canada, and in the Norwegian–Barents Seas, similar to the initial-front cyclones, but also in the Mediterranean. The preferred cyclogenesis regions for the late-front cyclones are in the Gulf of Alaska, the leeward side of the Rocky Mountains, south of Greenland, in the Nordic seas, and in the Gulf of Genoa. Hence, late-front cyclones form preferentially at the end of the storm tracks (Gulf of Alaska and Nordic seas) or as leeward cyclones (downstream of the Rocky Mountains and the Alps). Leeward cyclones can deviate strongly from the classical Norwegian life cycle concept (e.g., Hobbs et al. 1996). Finally, the relative frequencies of the no-front cyclones (Fig. 3c) are highest in the Atlas region, the Altai range (where, however, the 850-hPa surface frequently intersects the topography), in the East Siberian Sea, and farther poleward in the Arctic Ocean. The absolute frequencies are highest in the Nordic–Barents and Mediterranean Seas, and no-front cyclones form preferentially over Scandinavia, in the Beaufort Sea area, and in the Gulf of Genoa. Hence, this set of cyclones consists of quasi-stationary thermal lows (Atlas and Altai), lee cyclones (Mediterranean), and polar lows (Arctic latitudes).
Climatologies based on 2-m θ fronts.
Based on the 2-m θ front identification method with a threshold of 2.5 K (100 km)−1, initial-front cyclones still dominate the North Atlantic (Fig. 4a). However, in the North Pacific storm track, late-front cyclones have the highest frequencies (Fig. 4b), in contrast to the results based on the 850-hPa θe fronts. More precisely, in the North Pacific, around 60%–65% of all cyclones are late-front cyclones, and 30%–35% are initial-front cyclones. These numbers are almost exactly reversed when using the 850-hPa θe fronts. Climatologically, 2-m θ gradients are higher in the western and central North Atlantic, in particular over the Gulf Stream, compared to the gradients above the Kuroshio in the western to central North Pacific (not shown). Further, cyclogenesis events leeward of the Rockies, over the Gulf Stream, and along the southern tip of Greenland , which feed into the North Atlantic storm track, are predominantly classified as initial-front cyclones based on the 2-m θ method. In contrast, in the North Pacific, cyclones formed in a broader region over the Kuroshio and east of Kamchatka primarily feed the North Pacific storm track with initial-front cyclones. Accordingly, in the North Atlantic, three cyclogenesis areas contribute to the initial-front cyclone category, while only one region contributes to this category in the North Pacific. Hence, it appears plausible that, based on a 2-m θ front definition, initial-front cyclones dominate the North Atlantic, but late-front cyclones dominate the North Pacific. Of course, this result is sensitive to the front threshold: A lower threshold would increase the initial-front cyclone fraction in the North Pacific. However, note that the chosen threshold is already below what Sanders and Hoffman (2002) referred to as a moderate front strength [3.5 K (100 km)−1].
Subcategory of initial-front cyclones: Frontal-wave cyclones.
Initial-front cyclones can be further divided into two branches: a) cyclones that form on a trailing front and b) cyclones that form on a nontrailing front. A trailing front is attached to a parent cyclone, and these cyclones forming on a trailing front are typically referred to as frontal-wave cyclones (e.g., Bjerknes 1919; Appenzeller and Davies 1996; Renfrew et al. 1997; Rivals et al. 1998; Parker 1998; Schemm and Sprenger 2015). We demonstrate this subdivision of the initial-front cyclones based on the 850-hPa θe fronts. Initial-front cyclones with genesis on a nontrailing front (Fig. 5a) and frontal-wave cyclones (Fig. 5b) occur in different regions of the North Atlantic and North Pacific. Frontal-wave cyclones are most frequent in the central and eastern parts of both ocean basins (∼25%; Fig. 5b). Initial-front cyclones with genesis on a nontrailing front account for 50% of all detected cyclones in the eastern and central parts of the North Atlantic and North Pacific and up to 60% in the western parts (Fig. 5a).
Frontal-wave cyclogenesis occurs throughout the central North Atlantic and North Pacific and is not confined to the regions above western boundary currents, which is the preferred cyclogenesis region of the nontrailing subcategory. Because frontal-wave development requires a mature cyclone with an elongated trailing front, it is not surprising that frontal-wave cyclogenesis occurs preferentially toward the end of the storm track rather than in its entrance region.
WHEN ARE EXTRATROPICAL CYCLONES ATTENDED BY FRONTS?
Finally, we address this question: When during their life cycle are cyclones attended by fronts? We perform this analysis separately for the 850-hPa θe and the 2-m θ fronts, keeping in mind that the latter method tends to put Gulf Stream and leeward cyclones in the same category.
We start with some technical notes up front. First, we label all times during the cyclone’s life cycle when it is attended by a front. Consideration is given only to cyclone tracks that are located at least for one 6-hourly time step north of 30°N (in total, about 45,000 tracks). To compare this multitude of tracks with different lifetimes, they are normalized to obtain a dimensionless track time with three fixed points: −1 at cyclogenesis, 0 at the time of minimum SLP, and +1 at cyclolysis (for details, see the appendix “Technical details—Normalizing cyclone track length”). Afterward, we average over all normalized cyclone life cycles to obtain the fraction of cyclones that are attended by fronts along their life cycle (Fig. 6). We start by discussing the results of the 850-hPa θe method (Fig. 6a).
By definition, all cyclones of the initial-front category are attended by a front at cyclogenesis (Fig. 6a). Subsequently, a decrease to a fraction of approximately 80% is observed, and this fraction remains fairly constant until the time of minimum SLP is reached. Hence, a large proportion of initial-front cyclones are attended by a front until decay begins. For late-front cyclones, the fraction of cyclones with attendant fronts strongly increases during the phase of cyclone intensification between normalized times −1 and 0. The fraction peaks (∼57%) shortly before maximum intensity is reached. During the decay phase, the fraction of late-front and initial-front cyclones with attendant fronts decreases to approximately 40% until cyclolysis for both categories (Fig. 6a). These fronts at cyclolysis potentially serve as the source of new cyclogenesis events.
Most late-front cyclones acquire a front shortly before maximum intensity. Initial-front cyclones follow more closely the conceptual model of the Bergen School, with a well-marked front at the initial time and a joint evolution of the front cyclone system. In fact, this is the case for almost 80% of all cyclones in this category. Reducing the frontal threshold to 3.5 K (100 km)−1 introduces a moderate increase in the fraction of front-attended cyclones in both categories (dashed lines in Fig. 6a). However, qualitatively, the results based on the two different thresholds lead to similar conclusions.
Some aspects of the results change if we base our analysis on 2-m θ fronts, but the overall picture remains unchanged (Fig. 6b). Around 50% of all cyclones that acquire fronts during their life cycle are attended by a front at maximum intensity, although, in contrast to θe fronts, there is no clear peak at maximum intensity. Rather, the fraction of cyclones with attendant fronts continues to increase weakly up to 53% and decreases shortly before lysis back to 50%. This decrease after maximum intensity is less pronounced compared to the 850-hPa θe fronts. Again, the qualitative picture does not change if the frontal threshold is changed from 2.5 to 1.5 K (100 km)−1.
To summarize, most cyclones that form in the absence of a front acquired a front around maximum intensity regardless of which front definition is used. Overall, we argue that the qualitative agreement between the results based on the two thermal front definitions is large, with a small difference discernible during the cyclones’ decay phase. Because of the difficulty in detecting occluded fronts, many decaying cyclones are likely associated with an occlusion at cyclolysis. However, with our approach, this fraction cannot be quantified.
SUMMARY AND CRITICAL DISCUSSION.
The investigation of the relationship between extratropical cyclones and attendant fronts was performed to identify three cyclone categories. The first category leans on the Bergen School cyclone model, which contains extratropical cyclones with a well-marked front at cyclogenesis (initial-front cyclone) and a joint evolution of cyclones and fronts. The second category encompasses cyclones that acquire a front during their life cycle but without a clear front at genesis (late-front cyclone). The third category contains cyclones that are never attended by a front according to the underlying front definition. Using two automated identification schemes for fronts and one for cyclones allowed us to partition the total cyclone climatology into the three categories and to analyze 35 years of cyclone tracks.
Based on 850-hPa θe fronts, we find that initial-front cyclones grow preferentially over the western boundary currents and along southern Greenland and dominate the storm tracks in the North Atlantic and North Pacific. Late-front cyclones grow preferentially along the leeward side of mountains (Rocky Mountains and Alps) and dominate the cyclone climatology over the continents and in the Mediterranean.
A front definition based on 2-m θ classifies, in agreement with the 850-hPa θe fronts, the majority of cyclones over North America and in the North Atlantic as initial-front cyclones. However, the majority of cyclones in the North Pacific are classified as late-front cyclones and not as initial-front cyclones. While most cyclogenesis events leeward of the Rocky Mountains, over the Gulf Stream, and at the southern tip of Greenland are associated with a 2-m θ front, and therefore contribute jointly to the high number of initial-front cyclones in the North Atlantic, it is only genesis over the Kuroshio that contributes to the initial-front cyclone category in the North Pacific. The fact that 2-m θ gradients are climatologically higher over the Gulf Stream compared to the Kuroshio likely explains the asymmetry between the two basins.
When are cyclones attended by fronts?.
Based on the results using the 850-hPa θe fronts, we conclude that the majority (∼60%) of all extratropical cyclones, which start without a preexisting front at genesis, acquire a front shortly before they reach maximum intensity. The majority (80%) of cyclones that formed on preexisting fronts are attended by fronts until they reach maximum intensity. The fraction of front-attended cyclones clearly decreases during the phase of decay for both categories. At the time of cyclolysis, about 40% of all cyclones are still attended by a front. Using 2-m θ fronts, the overall picture does not change substantially. Again, most cyclones without a preexisting front at genesis acquire a front when reaching maximum intensity (∼50%). After reaching maximum intensity, the fraction remains at a high level (∼53%) for a longer period compared to the θe fronts. At lysis, around 50% are still attended by fronts. This fraction might be even higher if occluded fronts were included.
In summary, our analyses confirm the dichotomy in the relationship between cyclones and surface fronts. Both initial-front and late-front cyclones contribute essentially to the overall cyclone climatology in the Northern Hemisphere in winter, indicating that nature promotes both types of scenarios: cyclones that form on preexisting fronts and fronts that develop during and because of cyclone intensification. Future work will show whether the two types of cyclones systematically differ, for example, in terms of their upper-level forcing and the role of diabatic processes.
Limitations of our study.
As it is the case with any automated detection of cyclones or fronts, the underlying assumptions introduce an element of uncertainty in the results. For example, the variables and global thresholds used for identifying fronts are necessarily based on (forecasting) experience and to a certain degree subjective. Also, one might question the use of globally uniform thresholds. More generally, because no overall consensus exists in the community about how to define a (weather) front (see the “Background” section), our results will need to be revised once an ultimate front definition would be available. Occluded fronts and airmass boundaries complicate the issue. Analyzing manually 35 years of 6-hourly data in the Northern Hemisphere is not only impractical, but because of the lack of agreement among trained experts, it is also not desirable. Therefore, our results, which address a long-standing question of dynamical meteorology, rely on the assumption that our automated approaches to identify cyclones, frontal zones, and fronts are meaningful. The sensitivity studies performed (different frontal identification methods with different gradient thresholds) indicate that the main findings presented above and visualized in Fig. 6 are reasonably robust.
“It has been said that synoptic meteorology is imbued with as much art as science” (UKMO 1964, p. 19). As demonstrated in our study, automated detection schemes of frontlike features are an essential tool for analyzing longtime reanalysis or climate datasets. But we feel that the basic strategies for automated frontal analysis stand at a crossroad. While it becomes more and more challenging to accurately identify front lines in high-resolution “noisy” datasets without additional filtering (and thereby removing the benefits of high-resolution data), front identification schemes that identify fronts as 3D volumes provide promising alternatives. These schemes may deviate even more from the classical concept of a front line on a surface weather map than the techniques used in this study. More fundamentally, when considering the themes of frontal analysis and detection, it would be appropriate to ask questions such as the following: What in today’s operational forecasting is the purpose of frontal analysis, what were traditional purposes, and how do they differ from modern research purposes and needs required to answer questions similar to the one posed in this study?
We wish to thank three anonymous reviewers and the editor Jeff Waldstreicher (NOAA) for their constructive comments, in-depth discussions on frontal analyses, and their assessment of our manuscript. The historical overview on automated front detection was triggered by the review process. We also thank Huw Davies for providing further comments on our manuscript. Figures 2–5 were produced using NCL (Computational and Information Systems Laboratory Technology Development Division 2016). The monotonic spline interpolation and Fig. 6 were calculated using R. We acknowledge the ECMWF for providing the ERA-Interim dataset. Funding for this study was provided by Swiss National Science Foundation (P300P2_167745), ETH Zürich, and University of Bergen.
APPENDIX: TECHNICAL DETAILS—FRONT AND CYCLONE DETECTION SCHEMES AND PARAMETERS.
For all analyses, we focus on Northern Hemispheric winter [December–February (DJF)] and use 6-hourly time steps of the ERA-Interim dataset on a 1° × 1° grid (Dee et al. 2011). The monthly mean cyclone and 850-hPa θe front data are publicly available [at http://eraiclim.ethz.ch; more details on the available climatologies are presented by Sprenger et al. (2017)], and the higher-resolution data and tracks are available on request from the corresponding author.
Fronts based on θe at 850 hPa.
The first front identification scheme relies on the detection of θe fronts at 850 hPa. Fronts are localized using the thermal front parameter, which is a directional derivative that measures changes in the gradient of a pertinent thermodynamic variable in the direction of the gradient (Renard and Clarke 1965). The strategy is as outlined in Hewson (1998) and applied in this study with further refinements as presented in Jenkner et al. (2010) and Schemm et al. (2015). These refinements encompass i) placing the front line in agreement with the original version of the thermal front parameter of Renard and Clarke (1965) at the maximum front gradient, ii) a minimum length criterion such that only fronts longer than 500 km are accepted, and iii) an advection filter that eliminates quasi-stationary frontal zones that occur along coastlines or steep topography. For the filtering, a minimum advection criterion of 3 m s−1 is applied. The minimum accepted θe gradient of a front is set to 4 K (100 km)−1 at 850 hPa. This threshold is 1.0–1.5 K (100 km)−1 higher than its climatological value in the Gulf Stream region during winter. The grid points are joined into a 1D frontal object, and we use a cubic spline interpolation to draw the fronts.
Fronts based on 2-m θ.
The second method relies on the recommendations by Sanders and Hoffman (2002). Because we study the entire Northern Hemisphere, and to remain consistent with the other methods, we rely on 2-m temperature obtained from ERA-Interim and convert it to 2-m θ rather than using station data. According to Sanders and Hoffman (2002), a strong front gradient is 7 K (100 km)−1, and a moderate gradient is 3.5 K (100 km)−1. To account for the coarse resolution of the reanalysis, and after a series of manual experiments, we decided to reduce the minimum front threshold to 2.5 K (100 km)−1. As for the 850-hPa θe fronts, we accept only fronts exceeding 500 km in length, we remove quasi-stationary gradients using a minimum advection threshold, and we place the front according to the thermal front parameter similar as for the 850-hPa θe fronts.
Cyclone detection: Initial-front, late-front, and no-front cyclones.
Cyclones are detected as closed isobars in the mean SLP field following the original concept of Wernli and Schwierz (2006), with a contour-search interval of 0.5 hPa and further refinements presented in Sprenger et al. (2017). A tracking scheme is used that accepts cyclone tracks if they persist at least 24 h. The location of cyclogenesis, which is defined as the occurrence of a first closed SLP contour along a track, is given by the SLP minimum at the starting point of a cyclone track. More details are given in Wernli and Schwierz (2006). A cyclone track is regarded as attended by a front at a given time if the outermost closed contour of the 2D cyclone object overlaps with a 1D front object at this time.
APPENDIX: TECHNICAL DETAILS—NORMALIZING CYCLONE TRACK LENGTH.
For every cyclone track, the time relative to the time of maximum intensity, ∆t = t − tmin(p), is divided by the time period between cyclogenesis and maximum intensity: ∆tn = |tcyclogenesis − tmin(p)|, for t < tmin(p). After the maximum intensity is reached, that is, for all t > tmin(p), we divide by the time between cyclolysis and maximum intensity: ∆tn = |tcyclolysis − tmin(p)|. To turn this into a continuous normalized track lifetime between −1 and 1 for every life cycle, a monotonic cubic spline (Fritsch and Carlson 1980) is computed.
The spline interpolation can introduce slight deviations from the thermal front parameter or introduce crossings of the isotherms.