a. Warm season distribution

Figure 3 shows the percentage distribution of the nine convective structures across the northeastern United States for the 1071 convective elements included in this analysis. Of the three main organizational structures (i.e., cellular, linear, and nonlinear), cellular convection is the most common type, composing nearly half of all warm season convection. Within the cellular group, it is more common for cellular convection to organize into CC compared to IC. Supercells are categorized as ICs, suggesting that supercells are not common over the Northeast. NL systems make up one-third of all convective storms, while linear systems, the least common organizational type, develop only one-fifth of the time. Over the Northeast, half of the linear systems that develop have no associated stratiform precipitation, while about 30% are trailing stratiform systems.

The monthly distribution of each structure was also examined (Fig. 4). Both BL and IC develop more often during the early warm season months of May and June, respectively. The remaining organizational structures preferentially develop later in the warm season from mid- to late summer. Clusters of cells are equally as common from June through August. Nonlinear and two types of linear convection (NS and BE) develop more frequently during the late summer months, especially August. The remaining linear systems (TS, PS, and LS) do not show a preference for development during any part of the warm season, perhaps because of the more limited sample size.

b. Spatial variations and initiation

A Hovmöller plot shows the frequency of occurrence for isolated cells, cluster of cells, and linear and nonlinear types as a function of time of day across the four regions of the Northeast (Fig. 5). For the results presented herein, occurrence or development of a convective type is defined as the time it initiates, organizes into that type from a different convective structure, or enters into a region in Fig. 1 from outside the Northeast domain. All linear systems are grouped into one category (including BL) in order to increase sample size.

Two-thirds of all ICs occur from the morning into the late afternoon (1200–2100 UTC) over the land areas, with maximum development over the high terrain (Fig. 5a). The formation of CCs is maximized during approximately the same time frame (1200–0000 UTC) as ICs and occurs mainly over the upslope and high terrain regions (48% of all CC initiation; Fig. 5b). Unlike ICs, there is a small secondary maximum in CC development over the coastal plain in the early morning hours (0600–1200 UTC). Over the upslope and high terrain regions, the CCs may form during any time of the day, with their development decreasing from the high terrain region eastward to the coastal ocean.

Approximately two-thirds of all linear systems initiate between the late afternoon and early evening (1800–0000 UTC) over the land areas (Fig. 5c), about 6 h later than cellular development (IC and CC). There is almost no linear development during the early morning hours (0600–1200 UTC). Nonlinear systems primarily (61%) organize during the afternoon into the overnight hours (1800–0600 UTC) over all domains (Fig. 5d), which is ∼3–6 h later than the linear systems. There is a secondary peak in nonlinear organization at 1200 UTC near the coastline. Similar to linear systems and cellular clusters, NL initiation is maximized over the Appalachian upslope and declines eastward toward the coastal ocean.

Figure 6 compares the number of developing convective structures within each of the four domains throughout the diurnal cycle. About 41% of all Northeast convective structures (596 of 1451 events) develop over the upslope domain (Fig. 6a). This percentage decreases over the high terrain (27%) and coastal plain (24%), decreasing dramatically over the coastal ocean (7%). The peak in the frequency of occurrence over the upslope region is partially a reflection of upstream convection moving into the domain from the west. Manual inspection of all convective events entering the boundaries of the four domains in June 2007 suggests that approximately the same number of convective systems move into the high terrain domain as the upslope domain. Thus, the upslope domain is the most active of the four domains for convection developing in situ.

Over the upslope region (Fig. 6a), during the morning and early afternoon hours (1200–1800 UTC), CCs and NL are the most common convective structures. By the late afternoon and early evening (1800–0000 UTC), linear and nonlinear systems develop most often. The frequency of initiation declines slightly into the early nighttime (0000–0600 UTC), with linear and nonlinear systems remaining the most common. The total storm initiation declines into the nighttime and early morning hours (0600–1200 UTC), and nonlinear convection becomes the dominant organizational structure (41% of the total structures). Throughout the day over this region, there is a progression of the dominant initiating type from cellular and nonlinear in the morning, to linear and nonlinear in the afternoon/evening, to primarily nonlinear during the nighttime. This evolution from cellular clusters to more linear and nonlinear systems is also apparent over the high terrain and coastal plain regions (Figs. 6b,c). However, in the morning, there is less developing nonlinear convection over these two regions, with ICs over the high terrain domain rather than CCs. The high terrain and coastal plain are also dominated by nonlinear systems in the evening (0000–0600 UTC).

The diurnal evolution of convection is much different over the coastal ocean. Nonlinear convection is the most common organizational structure during the day, with the exception of the late afternoon and early evening hours (1800–0000 UTC). During this time, all convective structures develop at approximately the same frequency.

a. Cellular convective events

Figure 7 shows a series of spatial composites centered on the developing convection, illustrating the synoptic evolution for cellular (IC and CC) events, from 24 h before a cellular structure develops (t − 24 h) to the time of cell development (t − 0 h). At t − 24 h (Fig. 7a), a westward extension of the Bermuda anticyclone is located ∼1200 km to the southwest of the cellular point, with a weak surface trough immediately east of this location. Southwesterly flow around the anticyclone advects warm and moist air (θ_e ∼ 336 K) air toward where the cell(s) first forms, with MUCAPE between 600 and 1000 J kg⁻¹. The core of the 20 m s⁻¹ 300-hPa jet is ∼1200 km poleward of the initiation point (Fig. 7b). A weak 500-hPa trough is located ∼300 km to the west of the initiation point and the midlevel flow is nearly westerly. Inspection of many cases revealed that this nearly zonal flow in the composite is partially the result of the cancellation of the exact position of the short-wave troughs, as well as the variety of flow regimes (northwesterly, southwesterly, and westerly) during these events. About 34% of cellular events develop in association with a midlevel short-wave trough.

By t − 12 h (Fig. 7c), there is little change in the surface pressure field (Fig. 7c), while MUCAPE decreases to <500 J kg⁻¹ over the initiation point and the θ_e decreases to ∼334 K. This composite time occurs during a diurnal minimum for many events, since 92% of cellular events develop between 1500 and 2100 UTC (Table 1). The 500-hPa trough is still ∼300 km upstream of the initiation point, although it remains broad and weak (Fig. 7d).

At t − 0 h, the instability increases at the initiation point through daytime heating (MUCAPE > 1000 J kg⁻¹), and the θ_e increases to 340 K (Fig. 7e). The precipitable water (32 mm) has increased little (∼2 mm) over the previous 24 h (not shown). The surface trough to the east of the developing convection is more robust at this time (Fig. 7e). There is a weak 500-hPa trough axis centered where the cells first develop, so there is no vorticity advection, and the 300-hPa jet is displaced well to the north (Fig. 7f). As a result, there is little or no midlevel quasigeostrophic (Q vector) forcing (not shown). There is also little to no 850-hPa temperature advection where the cellular convection develops (Fig. 8a), and little frontogenetical forcing at 900 hPa (Fig. 8b).

Because there is limited synoptic forcing for ascent for these cellular events, it was hypothesized that upslope terrain forcing may provide a lifting mechanism. Figure 9 illustrates where the 47 cellular (IC and CC) convective events occurred over the Northeast with respect to four flow regimes at 900 hPa (045°–135°, 135°–225°, 225°–315°, and 315°–045°) obtained from the NARR dataset. From this analysis as well as the inspection of the individual events, ∼60% of these cellular events develop along the terrain slopes in association with an upslope flow component. For example, westerly flow at 900 hPa (triangles) yields cellular initiation along western slopes and crest of the Pennsylvania Appalachians. Southerly flow (diamonds) favors cellular initiation on the southern and eastern slopes of the Catskill Mountains in south-central New York. Thus, orographic lift is an important lifting mechanism during a majority of the cellular events. The remainder of the cellular storms forms along or near a surface baroclinic zone (∼25% of all cellular events, not shown), as well as along sea-breeze convergence zones over coastal New Jersey (∼3 events).

b. Linear convective events

Figure 10 illustrates the composite synoptic evolution of all linear events from t − 24 h to the time of the most intense linear convection (t − 0 h) over the Northeast. At t − 24 h (Fig. 10a), a surface cyclone and trailing cold front is located ∼1500 km to the northwest of where the linear system develops. There is enhanced MUCAPE (600–1000 J kg⁻¹) to the east of the surface front, which is associated with the northward advection of a surface θ_e ridge (∼338 K) by southwesterly flow around an anticyclone off the southeastern U.S. coast (Fig. 10a). The 500-hPa flow over the linear convection point is west-northwesterly at 10 m s⁻¹ as a result of a broad 500-hPa ridge ∼600 km to the west (Fig. 10b). The 30 m s⁻¹ 300-hPa jet core is centered ∼1000 km poleward of the convection.

At t − 12 h (Fig. 10c), the surface cyclone and associated upper-level trough advance eastward to within ∼1000 km of the linear convection point, while the 300-hPa jet gains a larger meridional flow component over the Great Lakes (Fig. 10d). Since 86% of the linear events included in the composites occur during the day (1500–0000 UTC; Table 1), the instability within the warm sector east of the surface front has decreased to ∼600 J kg⁻¹ (Fig. 10c) through nocturnal cooling. Warm advection shifts the θ_e gradient poleward of the linear convection point, while a surface trough begins to develop with a north–south-oriented axis over the convection location. This newly developing surface trough forms ∼1000 km east of the surface cold front.

At t − 0 h (Fig. 10e), the trough to the east of the surface cold front has amplified (Fig. 10e), with linear convection organizing near the surface trough axis. It appears that linear events develop in association with a prefrontal surface trough ahead of a cold front. Examination of the individual events revealed that these prefrontal troughs are almost always located in the lee of the Appalachians. Manual inspection of these linear events suggests that the prefrontal troughs in the Appalachian lee develop from 1) adiabatic compressional (downslope) warming, 2) synoptic-scale ascent as the midlevel trough advances ahead of the surface cold front, or 3) a forward-tilting frontal structure, with the warmest mean tropospheric air and associated surface pressure trough located ahead of the surface front (Schultz and Steenburgh 1999; Schultz 2005).

Although linear systems can develop along prefrontal troughs, it is not as common as the composites might suggest, given the event averaging. During 40% of all linear events, there is a prefrontal/lee trough present, although the convection does not always organize within this trough. Manual inspection of the surface pressure maps revealed that ∼22% of linear systems form within the axis of a prefrontal surface trough, while ∼39% develop ahead of and within 300 km of a surface cold front, not within a surface trough. Furthermore, ∼11% of the linear events form along a surface cold front within the associated surface trough, while ∼17% form within the warm sector in the absence of any surface baroclinic zones in close proximity.

At t − 0 h (Fig. 10e), moist, unstable air advects to the northeast creating a wedge of relatively large MUCAPE (1000–1400 J kg⁻¹) and a θ_e ridge near the linear development. Meanwhile, the precipitable water values over the organizing convection have increased from 33 to 38 mm during the previous 24 h (not shown). A broad 500-hPa trough axis northwest of the convection trails the surface cyclone by ∼600 km, while the amplifying 300-hPa jet remains ∼1000 km poleward of the linear convection (Fig. 10f). Given the locations of the 500-hPa trough and 300-hPa jet (Fig. 10f), there is weak Q-vector convergence (large-scale lift) from 700 to 500 hPa associated with the linear convection (not shown).

At 850 hPa for t − 0 h (Fig. 8c), the temperature advection is near zero over the developing convection, while there are relatively large warm and cold advections (±5.4 × 10⁻¹ °C h⁻¹) to the northeast and northwest of this point, respectively. Meanwhile, there is 900-hPa frontogenesis [>10 × 10⁻² K (100 km)⁻¹ (3 h)⁻¹; Fig. 8d] located immediately northwest of the linear convection (Fig. 8d), suggesting that linear systems organize on the warm side of a 900-hPa frontogenesis maximum.

c. Nonlinear convective events

At t − 24 h prior to nonlinear convective events (Fig. 11a), a surface pressure ridge is located over the nonlinear convection development point, with a broad surface trough ∼1500 km to the west. The lack of a well-defined surface trough is the product of compositing a variety of surface cyclone locations relative to the convection. The environment over the development location is weakly unstable (MUCAPE ∼ 200 J kg⁻¹), with most of the instability >400 km equatorward. A meridionally orientated θ_e gradient extends from the Midwest to the East Coast. There is ∼10 m s⁻¹ of 500-hPa west-northwesterly flow and a 20 m s⁻¹ 300-hPa jet over the nonlinear development location (Fig. 11b).

By t − 12 h, the broad surface trough advances eastward and is located over the Great Lakes, although there is little change in the ambient instability and thermal gradient as compared to 12 h earlier (Fig. 11c). The 500-hPa geopotential height field amplifies, as a midlevel trough deepens advancing toward the nonlinear composite point and the 25 m s⁻¹ 300-hPa jet core broadens (Fig. 11d).

By t − 0 h, a more defined surface trough is located immediately west of the developing nonlinear convection, and the surface winds have veered to southerly ahead of this trough (Fig. 11e). Regardless, the instability remains equatorward, with MUCAPE of ∼200 J kg⁻¹ over the developing convection. The destabilizing of the atmosphere through diurnal heating is not required for nonlinear convection, since 57% of all nonlinear cases included in the composite develop between 0600 and 1200 UTC and the other 43% occur evenly between 1200 and 0600 UTC (Table 1). The precipitable water values have increased from 27 to 34 mm over the previous 12 h (not shown). The nonlinear system organizes along the surface baroclinic zone, with ∼60% of the individual events included in this composite developing along a surface warm front (not shown). The 500-hPa trough continues to amplify as it approaches the developing convection (Fig. 11f). The 25 m s⁻¹ 300-hPa jet shifts equatorward, placing the nonlinear convection in the right entrance region of the jet core, an area favorable for ascent and Q-vector forcing (not shown). Furthermore, 0–6-km shear over the developing NL convection has increased over the previous 12 h, with 1000-hPa winds rotating to from westerly to southerly as the surface trough encroaches on the developing convection in conjunction with a strengthening 500-hPa westerly flow (Figs. 11c–f).

At t − 0 h, 850-hPa temperature advection maximizes over the nonlinear convection, with values >45 × 10⁻⁶ °C s⁻¹ (>1.6°C h⁻¹; Fig. 8e). Furthermore, the convection develops within a region of 900-hPa frontogenesis [6 × 10⁻² K (100 km)⁻¹ (3 h)⁻¹] associated with a strengthening warm front to the east of the surface trough (Fig. 8f). Overall, nonlinear systems develop in association with strong forcing by an active baroclinic system and limited instability. The large areal coverage of nonlinear systems developing under strong dynamical forcing is consistent with Schumann and Roebber (2010). Examining the relationship between convective morphology and forcing by potential vorticity features on the dynamic tropopause, they found that the stronger the forcing, the more widespread the convection. Recall that the development of nonlinear systems is more evenly distributed over all four domains (UP, HT, CP, and CO) compared to the other organizational structures (cellular and linear; Fig. 5), indicating that these systems are the least influenced by the underlying topography.

a. Comparison between organizational structures

To more quantitatively compare the physical processes among the convective structures (i.e., cellular, linear, and nonlinear), 1.5° latitude–longitude area-averaged values are calculated for a variety of variables centered on the compositing point, for each of the individual events included in the composites (Table 1). A bootstrap method (Zwiers 1990) was applied to the area-averaged variables to test for statistical significance between the three groups (i.e., cellular, linear, and nonlinear). For each variable, a new sample of the same size was generated 1000 times by randomly selecting from the original sample. The 95% confidence intervals around the mean were determined by finding the 2.5th and 97.5th percentile of the means of the 1000 resamples. The thermodynamic and dynamic quantities that support Northeast convective structures are also compared to Midwest and central Plains convective environments (Weisman and Klemp 1982; Parker and Johnson 2000; Coniglio et al. 2007).

Figure 12a illustrates the distribution of 900-hPa frontogenesis for the cellular, linear, and nonlinear events. The mean 900-hPa frontogenesis values for linear [11.6 × 10⁻² K (100 km)⁻¹ (3 h)⁻¹] and nonlinear [13.1 × 10⁻² K (100 km)⁻¹ (3 h)⁻¹] systems are significantly greater than cellular events [1.6 × 10⁻² K (100 km)⁻¹ (3 h)⁻¹] at the 95% level (not shown). A majority (∼77%) of the cellular events that develop under frontolytical situations are forced by orographic lift over the Appalachian terrain, while the remaining events were supported by Q-vector forcing at midlevels (not shown). The tail in the distribution for nonlinear events is skewed toward stronger frontogenetical forcing, with values as high as 63.1 × 10⁻² K (100 km)⁻¹ (3 h)⁻¹. Also, the mean 850-hPa warm air advection for nonlinear convective events (3.3 × 10⁻⁵ °C s⁻¹ or 0.2°C h⁻¹) is significantly (95% level) larger than the near-neutral temperature advection during cellular (−0.5 × 10⁻⁵ °C s⁻¹) and linear (−0.6 × 10⁻⁵ °C s⁻¹) events (Fig. 12b).

The mean precipitable water value for linear convection (37.9 mm) is significantly greater than cellular (32.5 mm) and nonlinear (33.4 mm) convection at the 95% level (Fig. 13a). Furthermore, the spread of values is smaller for linear events, with values ranging from 27.4 to 48.7 mm, compared with cellular (9.3–45.7 mm) and nonlinear (14.19–46.43 mm) events. The middle 50% of precipitable water values (35.2–40.7 mm) for Northeast linear convection is greater than the average values for linear systems in the Great Plains (24.3–33.5 mm; Parker and Johnson 2000).

Both cellular and linear convection develop in moderate MUCAPE environments (Fig. 13b). The interquartile range for cellular convection is between 516–1124 J kg⁻¹ and 754–1322 J kg⁻¹ for linear convection (Fig. 13b). The MUCAPE values are similar to the central U.S. Plains during May, with mean values ranging from ∼800 J kg⁻¹ for the PS type to ∼1600 J kg⁻¹ for the TS system (Parker and Johnson 2000). However, there is a large spread in MUCAPE that support Northeast linear convection (196–2737 J kg⁻¹; Fig. 13b). In contrast, the interquartile CAPE range for nonlinear convection is 21–301 J kg⁻¹, while the mean (207 J kg⁻¹) is significantly (95% level) less than both cellular (973 J kg⁻¹) and linear (1112 J kg⁻¹) values. Furthermore, the maximum MUCAPE value associated with nonlinear events is only 1706 J kg⁻¹, which is substantially less that the maximum MUCAPE for linear (2737 J kg⁻¹) and cellular convection (3413 J kg⁻¹; Fig. 13b). Two events included in the nonlinear composites had an MUCAPE value of 0 J kg⁻¹, one of which was strongly and continuously forced by 850-hPa warm air advection and 900-hPa frontogenesis. This is consistent with the derecho work of Evans and Doswell (2001), who found that over the eastern two-thirds of the United States, strongly synoptically forced derechos associated with a deep midlevel trough and surface cold front can occur in environments with little to no CAPE. The other no MUCAPE event developed in association with relatively weak dynamical forcing (not shown), although there was potential instability present (1000–800 hPa).

It has been shown that the relationship between CAPE and shear is important for determining the structure of convection (Weisman and Klemp 1982, 1984; Rotunno et al. 1988; Weisman et al. 1988; Parker and Johnson 2000). Ambient shear over several depths was compared between the three convective types, with shear calculated as the magnitude of the vector difference over the layer. For 0–1-km shear, cellular convection forms in the smallest shear environments, with the middle 50% of cellular events forming in 1–3 m s⁻¹ of ambient shear (Fig. 14a). The interquartile range increases for linear convection (2.5–5.2 m s⁻¹) and more so for nonlinear convection (3.1–6.8 m s⁻¹). The mean values of 0–1-km shear for linear (3.9 m s⁻¹) and nonlinear systems (5.3 m s⁻¹) are significantly (95% level) greater than for the cellular convection (2.1 m s⁻¹). Furthermore, the mean 0–1-km shear for nonlinear systems is greater than for linear systems at the 90% significance level. The 0–3-km shear for linear and nonlinear convection is significantly (95% level) greater than cellular convection as well (not shown).

Mean values of 0–6-km shear for all three convective types are significantly different from each other at the 90% level, with a mean 0–6-km shear for cellular storms of 9.5 m s⁻¹, 12.3 m s⁻¹ for linear systems, and 16.9 m s⁻¹ for nonlinear events (Fig. 14b). The relatively weak deep layer shear during cellular events, with an interquartile range of 5.6–13.1 m s⁻¹, suggests that most northeastern U.S. cellular events are not supercellular (Fig. 14b). Through the use of proximity soundings, Thompson et al. (2003) found that the probability of supercells over the Midwest increases as the 0–6-km shear increases from 15 to 20 m s⁻¹. Though numerical simulations, Weisman and Klemp (1982) showed that for splitting cells to develop, the ambient 0–6-km shear must be a minimum of 20 m s⁻¹.

The middle 50% of linear events have 0–6-km shear values ranging from 8.6 to 14.8 m s⁻¹. There is little difference between the 0–3- and 0–6-km shear values for linear systems (not shown), indicating that the greatest contribution to the ambient shear comes from winds below 3 km, consistent with Rotunno et al. (1988). About three-fourths of Northeast linear events included in this dataset have smaller 0–6-km shear values compared to all types of linear convective structures over the eastern half of the United States (Evans and Doswell 2001; Cohen et al. 2007; Coniglio et al. 2007), which ranged from 12 to 20 m s⁻¹.

In contrast, Northeast nonlinear systems have the greatest deep layer shear, with an interquartile range of 12.7–21.1 m s⁻¹, consistent with a well-defined baroclinic trough from the surface to midlevels (Figs. 11e,f). The 0–6-km shear vector during Northeast nonlinear events more closely resembles the shear values that support central U.S. quasi-linear systems (Evans and Doswell 2001; Cohen et al. 2007).

b. Comparison between Northeast and Midwest convective types

There are large differences in terrain between the Northeast and the central Plains, and the Atlantic coastal boundary adds another degree of complexity for the Northeast. These surface variations and associated ambient conditions (i.e., moisture, instability, and lift) will influence convective organizational structures over each region uniquely. To evaluate these differences, the structural distribution of convection over the Northeast (Fig. 3) is compared to the distribution over the Midwest highlighted by GSJ08.

Although the Northeast convective structures were classified according to the nine morphologies of GSJ08, the results cannot be compared directly since the analysis was completed manually by two separate research groups. Therefore, to compare our Northeast results with the GSJ08 analysis, we first classified the convective structures over the GSJ08 region of interest (i.e., Midwest) from 1 to 30 June 2002 using the same reflectivity data and approach described in section 2. We identified nearly twice as many convective structures across the Midwest compared to the GSJ08 analysis for that month. There are two reasons for this discrepancy (W. Gallus 2010, personal communication): 1) we classified all convective elements across the domain, while GSJ08 identified the dominant morphology over the Midwest during a particular time; and 2) we identified convective structures every 15-min, while GSJ08 identified them every hour. Since a goal is to obtain a qualitative comparison of the different structures between the Midwest and Northeast, the GSJ08 percentages of the various convective types were adjusted to account for these differences in classification. Two convective distributions for June 2002 were calculated, one using the classification of GSJ08 and the other using our approach. The percentage difference between these two June 2002 distributions was calculated (adjustment factor) and added to the total warm season structural distribution of GSJ08. This approach only provides a rough adjustment, so any differences noted between the two regions must be relatively large to be discussed.

Figure 15 illustrates the full warm season (April–August 2002) GSJ08 distribution modified by the adjustment factor noted above. The greatest variations between the original GSJ08 distribution (GSJ08, their Fig. 6) and Fig. 15 are that the adjusted distribution has 8% fewer NLs and 12% more CCs. Thus, the difference between the original GSJ08 distribution and the adjusted GSJ08 distribution are relatively small, giving us additional confidence in comparing the adjusted GSJ08 distribution with our Northeast results.

The main structural differences between Northeast (Fig. 3) and Midwest (Fig. 15) convection exist in the percentage of NL and IC convection. During the warm season, there is a greater percentage of NL systems over the Northeast (33%) than the Midwest (20%). There is likely a larger number of baroclinic systems over the Northeast during the warm season compared to the central Plains. A smaller percentage of convection organizes into ICs over the Northeast (14%) compared to the Midwest (24%). This difference may reflect the more frequent supercell development over the Midwest compared to the Northeast.