1. Introduction

The coast of Alaska is experiencing serious impacts owing to erosion and inundation. The primary drivers are waves and surge generated by strong winds that are typically associated with storms. Understanding the trends and dynamics of coastal erosion in Alaska is grounded in grasping both the dynamics and the climatology of storms in the North Pacific (NP) and Bering Sea and how they vary spatially and temporally (Atkinson 2005; Rachold et al. 2005). This issue extends beyond the physical processes of erosion—environmental and societal impacts are equally severe (Graham and Diaz 2001)—and includes the maritime industries (fishing, shipping) and subsistence hunting activities. The motivation to improve regional-scale understanding of storm track dynamics and patterns is thus strong. Storms are also a means by which the atmosphere redresses sinks and sources of energy and momentum; analyses with these theoretical meteorological objectives in mind are important for aiding our understanding of large-scale climate dynamics (Murray and Simmonds 1991).

Various analyses of storm activity have appeared in the literature; most have focused on the Northern Hemisphere as a whole, with the Arctic being separately targeted in some instances (Hoskins and Hodges 2002; Keegan 1958; McCabe et al. 2001; Serreze 1995; Sorteberg and Walsh 2008; Whittaker and Horn 1984; Zhang et al. 2004). Relatively few, however, have focused specifically on the North Pacific and Bering Sea regions (Graham and Diaz 2001; Norris 2000) apart from identifying the prominent North Pacific storm track. Less common still are studies that consider seasons other than winter: a summer discussion is unusual and autumn/spring, the equinoctial seasons, are almost never explicitly considered. Autumn is examined in Key and Chan (1999); however, they do not explicitly focus on the North Pacific.

Trends in storm activity have been examined in several papers. McCabe et al. (2001), using Serreze’s sea level pressure (SLP)-based algorithm and National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis data, studied storm track trends in the Northern Hemisphere for the period 1959–97. They found an increase in high-latitude winter cyclone frequency and a decrease in midlatitude winter cyclone frequency over this period. Graham and Diaz (2001), using NCEP–NCAR reanalysis data and their own algorithm also based on SLP, calculated trends in the NP (defined only as far north as 50°N) and found an overall winter [December–February (DJF)] cyclone frequency trend of 0.21 cyclones per year over the period 1948–98. Key and Chan (1999), using NCEP–NCAR reanalysis data, studied storm tracks using the tracking algorithm of Bell and Bosart (1989), which is based on the geopotential height at two levels (1000 and 500 mb). Interestingly, even as far back as 1989, they found that the surface cyclone frequency in the Arctic had increased during all seasons [DJF, March–May (MAM), June–August (JJA), and September–November (SON)], while surface cyclone frequency in the midlatitudes had decreased in both the NH and SH. They also found opposite signs in trends at 1000 and 500 mb, which could indicate that the systems had become shallower.

Storm activity analyses are generally derived from climatologies that are developed by establishing a definition of a “storm,” accumulating counts of their occurrences, and extracting statistics to determine their climatological characteristics (e.g., wind speed, ground speed, and duration). Analyses of storm activity, such as reviewed above, have used different methods to diagnose the storms, which vary in perspective (Lagrangian, Eulerian), method (manual and/or automated tracking), variable (mean sea level pressure, geopotential height, relative vorticity, quasigeostrophic vorticity, etc.), and tracking method (finding the central pressure in an MSLP field, finding the maxima in the relative vorticity field, and/or finding the storm center by analyzing satellite imagery). Considering studies that have included the North Pacific the following range of approaches may be noted: Graham and Diaz (2001) used their algorithm based on SLP from a Lagrangian perspective; Norris (2000) used the root-mean-squared daily vertical velocity at 500 mb during the summer season to diagnose storm tracks in the NP; and Simmonds and Keay (2002) used the Melbourne University cyclone finding and tracking scheme (Murray and Simmonds 1991) based on the geostrophic vorticity computed from the MSLP.

The key starting point in any analysis of storm activity is the event-level definition of a storm. The most common meteorological variable used for this purpose has been mean SLP (MSLP), implemented in manual (Pickart et al. 2009; Whittaker and Horn 1984) and various automated tracking schemes (Murray and Simmonds 1991; Serreze et al. 1997; Zhang et al. 2004). Another meteorological variable that has been used is relative vorticity (ζ). Sinclair (1994), using the method of Murray and Simmonds (1991), makes an important point about the poleward displacement of the pressure minimum relative to the vorticity maxima in the presence of a background zonal wind field, which is an advantage of the method. In particular, the method defined by Hodges (1994, 1995, 1996, 1999) has also been found to yield satisfactory results for several reasons (Barry and Carleton 2001; Hoskins and Hodges 2002). First, contrasted with MSLP-based approaches, relative vorticity is more sensitive to small, rapidly moving systems because MSLP tends to preferentially identify larger, slower-moving systems. In a geostrophic sense, vorticity focuses more on the small spatial scale end of the synoptic range so that more systems are identified than by using MSLP, for example, which focuses more on the large-scale end (Hodges et al. 2003). This is important for high-latitude work where small-scale features can occur with greater frequency than in the midlatitudes (Twitchel et al. 1989).

Second, MSLP can be adversely influenced by the background as discussed by Sinclair (1994). Also, a low pressure center situated within a larger region of low pressure can possess a weak pressure gradient which means surface winds are weak (Fig. 1). An algorithm based on identifying pressure minima would identify and categorize the system as strong due to a low absolute pressure reading (e.g., 956 mb in Fig. 1); however, the pressure gradient in the vicinity of the low pressure center is not very strong. Third, the presence of relative vorticity (ζ) is an important mechanism in the initiation of extratropical storms. Its use thus enables identification of a storm earlier in its life cycle (Sorteberg and Walsh 2008), which can aid the understanding of genesis dynamics. Fourth, ζ is important to the dynamical maintenance and growth of a system, so its use also facilitates ongoing system monitoring. Finally, relative vorticity is also related directly to wind speed because its derivation is simply the vertical component of the curl of the relative velocity, that is,

View Expanded

Linking storm track information results directly to winds also improves their utility for secondary studies (e.g., coastal erosion) that depend upon wind forcing. For these reasons, therefore, storm climatology results based on Hodges’ algorithm form the bases of the analyses presented here.

An important point that has often been overlooked in storm climatology analyses, especially with regards to impacts in northerly latitudes, is autumn activity. Autumn is an important time of year in the Arctic because it is the time of least ice, and therefore greatest open water exposure (see Fig. 12, where the 0° isotherm provides a functional indication of the average sea ice extent). In the region of study for this paper, autumn is also a time when the jet stream position and the occurrence of remnant tropical cyclone activity can create a particularly strong storm response.

Thus, it is the objective of this paper to aggregate storm activity results for the North Pacific and Bering Sea regions obtained using a robust and proven storm tracking algorithm, focusing on the seasonal time frame, including autumn. It also explores trends in storm activity responses, including trend considerations of the pre- and post-1979 periods and discussions of changes in SST. Changes in horizontal gradients of SST affect baroclinic instability, which in turn influences storm tracks. The seasonal climate over the Pacific has also undergone changes in the mid to late 1970s (Graham and Diaz 2001). These analyses will be conducted via a thorough discussion of different track variables, making this effort one of the most complete and updated studies of storm tracks for the North Pacific/Bering Sea.

Hence, the present paper is aimed at describing and understanding the climatology and trends of the storms in the North Pacific region, including the Bering Sea and the Gulf of Alaska, for standard climatological seasons, that is, winter (DJF), spring (MAM), summer (JJA), and autumn (SON). The following questions will be addressed:

• Has there been a relative change in the seasonal patterns of storminess in these regions?

• What are the characteristics of equinoctial season storms? (This question is important since spring and autumn are not typically addressed, yet many of the most problematic storms in the Alaska region occur in the autumn.)

• What is the ground track speed of storms? (The threat presented by a storm is increased if it is slow moving. Knowing how fast systems move is an important objective.)

Section 2 discusses the methodology used; section 3 presents the results, focusing on a seasonal intercomparison of the climatologies and then on a discussion of the trends in the tracking variables; finally, even though a direct comparison with magnitudes from other climatologies is not immediately possible given the different normalizations in other studies (Simmonds and Keay 2002), a discussion will conclude the paper that places these results in light of the existing literature, whenever possible. Owing to space constraints, detailed consideration of these patterns in the context of other major circulation mode indices will be considered in another paper.

2. Methodology

Storm systems were identified and tracked using the algorithm developed by Hodges (1994, 1995, 1996, 1999), and run for the NH and subset for this study. In this algorithm relative vorticity is analyzed directly on the unit sphere, not a preprojected or unprojected Cartesian grid. This approach “reduces the likelihood of introducing bias into the tracking and statistics and makes the parameters required for the tracking and statistical estimation independent of any projection” (Hodges et al. 2003). Essentially, using a gridded dataset as input, the algorithm identifies a storm center to be a relative vorticity maximum that exceeds 1 × 10⁻⁵ s⁻¹. To solve the correspondence between the feature points in consecutive time steps, which Hodges identifies as the most difficult aspect of any storm tracking system, a specialized cost function is applied (Hodges 1995). Minimization of the cost function results in the smoothest set of tracks subject to constraints on displacement distance and the track smoothness.

Relative vorticity at 850 hPa is used as the analysis variable since 850 hPa represents a good tropospheric weather activity level for storm tracking (Hodges et al. 2003). Likewise, vorticity represents a good analysis variable because it does not depend on any extrapolation techniques, as does MSLP. The tracks have been filtered to retain only those that last longer than 2 days and travel farther than 1000 km (mobile storms).

The algorithm input data, four times daily u and υ for the relative vorticity calculation, were drawn from the NCEP–NCAR Global Reanalysis 1 (GR-1) (Kalnay et al. 1996; Kistler et al. 2001) from 1948/49 to 2008, provided by NOAA/National Weather Service (NWS)/NCEP [NCEP–NCAR reanalysis data provided by the NOAA/Office of Oceanic and Atmospheric Research (OAR)/Earth System Research Laboratory (ESRL)/Physical Sciences Division (PSD), Boulder, Colorado, from their Web site at http://www.cdc.noaa.gov/]. SST data were provided by the NCEP–NCAR reanalysis as well.

Six sets of spatial statistics derived from the tracking were retained for analysis:

• (i) genesis density (gdens): indicates the areas where the storm tracks start [units: number density of starting points (10⁶ km² season)⁻¹];

• (ii) track density (tdens): number density of storm tracks passing through a region per season [units: storms (10⁶ km² season)⁻¹];

• (iii) lysis density (ldens): indicates the areas where the storm tracks die out per season [units: number density of ending points (10⁶ km² season)⁻¹];

• (iv) mean intensity (int): intensity of the vortex [units: intensity in 10⁻⁵ s⁻¹];

• (v) mean lifetime (lft): number of days the low pressure systems lasted that pass through a region;

• (vi) mean speed (mspd): mean track speed (m s⁻¹).

Two major analyses were conducted on each of these variables. First, a general pattern analysis by season was conducted for the six tracking statistics. Second, assessment of trends using least squares fitting for selected variables was performed for three collective regions in and around Alaska (Fig. 2). Statistical significance was based on a t test of the linear regression coefficients. The regions studied were:

• (i) Alaska interior/Yukon (AIY) from 60° to 70°N, 160° to 120°W;

• (ii) Gulf of Alaska (GOA) from 50° to 60°N, 160° to 120°W;

• (iii) Bering Sea (BeS) from 50° to 70°N, 160°E to 160°W.

The next section will discuss the main results of these analyses.

3. Results

4. Discussion and conclusions

Winter emerged as the most active season for most of the storm parameters, including genesis density and track density. Simmonds and Keay (2002) found high winter mean cyclone occurrences (tdens) over the Sea of Okhotsk and the Gulf of Alaska, which is essentially in accordance with our results, except that we find even higher tdens counts during autumn (a season not considered in their study). Winter storm tracks are also located farther south, a result consistent with many studies (Nakamura 1992; Simmonds and Keay 2002; White 1982). This location is also influenced by the midwinter suppression (Nakamura 1992). One of the key determinants of storm activity is a strong thermal gradient. The north–south gradient of temperature is greatest in the winter. The sea ice extent is farthest south in late winter (Fig. 12). In addition, the Kuroshio Extension (Qiu 2000), also known as the North Pacific Current, acts as a southern anchor for the thermal gradient. The net effect results in a thermal gradient that is strongest in winter. This is conducive to more frequent episodes of baroclinic instability that favor development, growth, and energy levels of extratropical cyclonic systems as is evidenced by descriptions in section 3. An analysis of mean spatial patterns of the index of Eady baroclinic growth rate (not shown) also indicates a maximum zone coincident with fall and winter storms.

The role of diabatic heating is strong in the generation of storms off the east coast of Asia and North America, providing another mechanism for generating the strong thermal gradient mentioned above, which supports the Pacific/Atlantic maxima (Hoskins and Valdes 1990). Release of latent heat also plays an important role in the intensity of the low pressure systems and the rapidity of cyclogenesis (Chang et al. 1982; Posselt et al. 2008). In a sense, the NP (and North Atlantic) storm tracks may be considered as self-maintaining, since the diabatic heating maxima in the storm track region are caused by horizontal and vertical flow displacements, which, in turn, are in part a by-product of the individual storms (Hoskins and Valdes 1990). Storm tracks can be self-maintained also by the fact that they help drive the Kuroshio owing to the low-level mean flow induced by the eddy activity and, in turn, the Kuroshio is crucial for the existence of the NP maximum (Hoskins and Valdes 1990).

The lysis density analysis results obtained here reinforces the concept that the Gulf of Alaska is the “graveyard” of storms in the NP as pointed out, for example, by Simmonds and Keay (2002). The tdens results (Fig. 4) suggest that most of the storm systems that end up in the Gulf of Alaska do not form locally. A targeted assessment verifies this fact (Fig. 13), which indicates that most storms are not formed locally, with the exception of the summer season, for which a source region in the Aleutian Islands is indicated. Also, the midocean cyclogenesis is likely due to secondary cyclogenesis and downstream development, as discussed in Hoskins and Hodges (2002). This may also be roughly concluded by realizing that the mean track speed (12 m s⁻¹) and lifetime (5 days) of these storms (winter considered) results in a lifetime distance traveled of ∼5000 km. Another important aspect is the fact that there are indications of tropical cyclones undergoing extratropical transitions and reaching GOA, especially during summer and autumn (Fig. 14).

Once in the Gulf of Alaska, there are several mechanisms at work that contribute to the large observed lysis values. First, barotropic processes typically are the dominant dynamical mechanisms at work. Understanding the role played by barotropic processes is important since the life cycles of baroclinic waves in a zonally symmetric flow are characterized by positive conversions in the growth phase and negative barotropic conversions in the decay phase (Frisius et al. 1998; Simmons and Hoskins 1978). This way, barotropic processes play an important role in the lysis phase of the storm tracks. Second, and perhaps exerting greater importance, the role of regional topography must be considered. The topography of coastal Alaska and northwest North America—the coastal highlands and mountains—is manifested as an influencing agent because its physical presence could be great enough to help destroy shear.

The shear of the meridional wind and the zonal wind contribute equally to the total shear production (destruction) (Frisius et al. 1998). Negative shear production has been shown to be responsible for the location of the “downstream end” of the storm track (Frisius et al. 1998; Whitaker and Dole 1995). Shear destruction of the transient eddies (barotropic processes) associated with lysis reduces the turbulent kinetic energy (TKE) because via this process, eddies release kinetic energy to the mean flow [TKE = advection + baroclinic production + shear production (destruction) + dissipation] and the storms die out, in this case, in the Gulf of Alaska (Black and Dole 2000; Frisius et al. 1998).

Another factor at work in the NP context is the presence of a favored location for a downstream blocking ridge over the Aleutians/Alaska region, also known as the “Alaska block” pattern (Frisius et al. 1998; Lau 1988), which can act to retard the eastward progression of systems along the Pacific storm track. According to Simmonds and Keay (2002), the Gulf of Alaska is also a region of high cyclone frequency; however, this is not incompatible with high lysis values because GOA is also an area of strong baroclinicity, indicated by the Eady growth rate assessment, which can enhance activity, and because systems can become stalled against topography or blocking patterns, mentioned above, which could enhance residence time. Thus, both the barotropic deformation (D) and the tropospheric baroclinic growth (B) could play an important role in the dynamics over the Gulf of Alaska, as illustrated by the ratio of D/B over GOA, as shown by Black and Dole (2000). Such a result is consistent with a relatively high Eady growth rate, an indicator that is proportional to vertical shear.

As for the lifetime, summer storms last longer than any other season. This result expands upon two previous studies (Mesquita et al. 2008; Zhang et al. 2004) that confined comparisons of lifetime to summer and winter. Summer storms are also the least intense with winter, spring, and autumn intensities being greater and similar to one another. Simmonds and Keay (2002) also found that, in general, the mean intensity is significantly reduced in summer with the most intense summer systems found over and southwest of the Aleutian region, also in accordance with our results. Their intensity plots, using a different measure of intensity, of the winter and summer intensity resemble our intensity DJF and JJA plots as well. Finally, the mean speed of the storms is higher in winter. In summary, summer storms last longer, move more slowly, and are less intense. A common factor linking these traits together is the growth rate (not shown). Summer storms tend to have much lower growth rates. One import feature of the climatology analysis is that the transitional seasons, spring and autumn, tended to have similar values for most of the storm parameters considered.

Whitaker and Dole (1995) mention that storm tracks exhibit considerable variability in interannual and intraseasonal time scales. We agree that interannual variability is considerable and are able to shed more light on the interseasonal issue; however, we have found a different result in our study concerning the interseasonal aspect: spring and autumn present similar characteristics for all of the variables with the exception that autumn intensity is stronger than in spring. Small differences in lifetime and lysis were also noted (autumn was larger in both cases). The greatest contrasts are observed during winter and summer, which are the times of greatest contrasts in temperature gradients. Summer storms have the lowest values for genesis, track density, intensity, and lysis. Summer storms, however, last longer than winter ones. This is in accordance with Zhang et al. (2004) and Mesquita et al. (2008). This result is expanded here to show that they last longer than in any of the other seasons. The track density (tdens) results by decade (Fig. 9) underscore the strong interannual and interdecadal variability that work against isolation of a trend signal.

Maxima spatial patterns exhibited a north/south shift. The main factor responsible for the maintenance of this zonal variation of baroclinicity is the distribution of heating (Frisius et al. 1998; Hoskins and Valdes 1990). A rotation “point” in the central Aleutian Islands as, for example, in the gdens, tdens, and mspd cases, was also apparent. This dogleg in the storm tracks observed here is caused in part by the temperature gradient changes in the North Pacific, but is also suspected to be a by-product of a midtropospheric (700 mb) moisture pattern in which the Bering Sea region is drier than the Gulf of Alaska in all seasons except summer (not shown).

Of the three aggregate regions considered in this paper, AIY has higher annual genesis compared to the other regions, GOA has a higher number of storms (tdens) and higher lysis, and BeS shows storms with longer lifetime and greater intensity, and speed. When considering storms with intensities exceeding 12 × 10⁻⁵ s⁻¹, however, BeS was found to consistently possess more of these strong intensity storms than GOA (Fig. 10). The Eady growth rate analysis indicated a zone of high Eady potential south of Russia’s Kamchatka Peninsula, which can serve to reenergize older systems as they move into BeS. It is hypothesized here that, by the time systems enter GOA, the increasing shear regime coupled with greater downstream distance from the Kamchatka Eady max precludes stronger development in most cases.

In general, most of the storm parameters trends observed in this study were not very robust. Most of the statistically significant trends at the 99% significance level are related to the genesis for the AIY region and genesis and lysis for BeS. The fact that different trends are found for the periods before and after 1979 reflects an underlying factor in this case: the influence of large modes of climatic variability, which may play an important role in such trends. These issues are considered in a companion paper. The issue of fleeting statistical significance speaks to both a few degrees of freedom issue caused by the relatively short time series available to work with and the large interannual variability observed in these variables. A possible additional source of pre- and post-1979 variability could be due to the introduction of large quantities of satellite data into the reanalysis after 1979, which has progressively increased with time (Bengtsson et al. 2004; Bromwich et al. 2007). There are a variety of points to make about the trends results, however.

When considering the trends for the three regions and the three different periods of time—the entire 1948–2008 time series and pre- and post-1979—we observe that most of the trends that are statistically significant (at the 95% and 99% significance levels) occur in the GOA region, generally for genesis density and lysis density. An interesting point is that Latif and Barnett (1996) showed strong positive SST trends for GOA. Positive SST trends and SST gradients are also observed in the NCEP–NCAR reanalysis data (not shown). Despite this, strong positive trends are not noted for those storm parameters that would be influenced by SST forcing, namely, genesis density: gdens decreases wherever its trend is significant. Hence, SST does not appear to be a key storm-forcing agent in the Gulf of Alaska. Some studies have related the sea ice cover in the Sea of Okhotsk to the location and intensity of the Aleutian low, which could play an important role in the GOA trends. For example, Honda et al. (1999) suggest a significant influence of ice cover on the downstream pressure field. Few strong, significant trends were noted for the Alaska interior. As for the Bering Sea, not many trends were found there.

Two variables are of particular importance for Alaska when it concerns trends: the number of storms (tdens or enter) and their intensity (int). None of the three regions considered show significant trends for these variables. Other studies have suggested that “storms are increasing.” For example, McCabe et al. (2001), using Serreze’s (1995) tracking algorithm, found a considerable increase in cyclone frequency in high latitudes (60°–90°N) and a pronounced decrease in midlatitudes (30°–60°N). The results of McCabe et al. represented broadly integrated swaths. Our results, however, indicate that, for the regions considered here, this is not really the case: there is no particular trend in storm activity in the AIY, BeS, or GOA regions.

As for the intensity, this result also appears to contrast with the paper by McCabe et al. (2001), who found that the storm intensity has increased both in mid and high latitudes. Graham and Diaz (2001) found an intensification of NP winter cyclones, which resulted from the increasing upper-tropospheric zonal winds between 25° and 40°N associated with ENSO. It is not simple to compare trends in intensity with McCabe et al. (2001) and Graham and Diaz (2001) because of the different approaches used to define and calculate the intensity variable. This could also be due to the use of pressure, which can be influenced by the large-scale background. If intensity is measured as the pressure minima, as in Graham and Diaz (2001), it would be influenced by any large-scale changes. However, the lack of intensity trends found in our study is consistent with the fact that these regions are a little north of the main path/track of the Pacific maximum, except the GOA, which is the region with high lysis density.

An interesting result is a clear indication that SST does not appear to be a strong storm control for the marine regions explicitly considered: the Gulf of Alaska and Bering Sea. Instead, local shear and topography conditions appear to exert the greater control. This is important in light of potential future increasing trends in SST if sea ice continues to retreat, that is, that storm activity will not follow suit because of SST alone. It is important to note that, overall, Bering Sea trends are difficult to identify in part because of high interannual variability. The relative lack of influence of SST also shows up in the lysis values that remain relatively consistent all along the west coast, even in the presence of considerable SST gradients.