1. Introduction
Weakly forced convection, thunderstorms forming without the support of synoptic-scale lift and shear, constitutes the majority of thunderstorms worldwide. As jet stream dynamics shift poleward during the summer months, weakly forced thunderstorms (WFTs) become increasingly common producers of midlatitude precipitation, especially in the southeastern United States. Without the support of large-scale forcing, WFTs rely upon mesoscale boundaries or preferential heating along orographic features for development. While some of these boundaries are regular and predictable, others are more difficult to detect. Consequently, several recent attempts have been made to identify heterogeneities in land cover (e.g., Haberlie et al. 2015), soil moisture (e.g., Frye and Mote 2010; Ford et al. 2015), terrain (e.g., Miller et al. 2015a), and humidity (e.g., Fabry 2006; Lee et al. 2016) that may influence convection under synoptically quiescent conditions in the southern and eastern United States.
Despite the prevalence of this storm mode in the southeast United States, attempts to explicitly identify WFTs have been limited and challenging (Miller and Mote 2017). Though forecasting textbooks emphasize the role of climatology in formulating a weather forecast (e.g., Lackmann 2011, p. 311), such a tool is unavailable for this already difficult-to-forecast thunderstorm type. In addition to forecasting implications, the absence of a WFT climatology impedes the establishment of broader connections between disorganized convection and the global climate system. Illustrating this potential, studies have discussed the important fraction of convective rainfall and latent heat release contributed by isolated, shallow convection in tropical regions (Schumacher and Houze 2003) as well as the role of lower-topped cumulus clouds in controlling radiative inputs and conditioning the atmosphere for subsequent deep convection (e.g., Johnson et al. 1999; de Szoeke et al. 2015). Though the precipitation climatology (Ingram et al. 2013) and regional lightning patterns (Murphy and Konrad 2005) in the southeast United States maintain a summertime signal consistent with WFTs, there has been no attempt to create an explicit WFT climatology for this, or any other, region of the world.
Further, the accurate prediction of pulse thunderstorms, defined as severe WFTs [see Miller and Mote (2017) for a detailed description of the disorganized convection nomenclature employed here], remains a challenge for contemporary forecasters (Guillot et al. 2008). For instance, a recent study found that National Weather Service (NWS) accuracy statistics during severe thunderstorm warning outbreaks, typically occurring in modest-instability, weak-shear environments, were poorer than the national average (Bruick and Karstens 2017). Climate models suggest that these pulse thunderstorm-supporting environments will become increasingly frequent in future climate regimes (Diffenbaugh et al. 2013) with recent studies already documenting a shift toward these conditions (Senkbeil et al. 2017; Ye et al. 2017). Within quiescent, WFT-favorable regimes, ever-growing urban areas are also known to promote convection (Shepherd 2005; Mote et al. 2007). As extreme examples of landscape change, urbanization poses a propensity to modify thunderstorm initiation patterns (Niyogi et al. 2011; Ashley et al. 2012). Given the disproportionate concentration of cultural and economic output in these areas, a more thorough understanding of their influence on regional thunderstorm climatologies, as well as the thunderstorms they produce, is essential. Though posing a weaker severe wind and hail threat than supercells and derechos on a per-storm basis, WFT-associated lightning kills more people (Ashley and Gilson 2009) and threatens to suspend sporting events and airport ground operations for many more days each year.
As the climatic conditions conducive to their formation and the human-induced landscape change aiding their initiation both advance, so should our understanding of WFTs and pulse thunderstorms. This study seeks to investigate foundational, yet underexamined, questions of WFT activity through 15 yr of radar observations in the southeastern United States (Fig. 1), including their frequency, favored areas of development, the synoptic conditions enhancing their formation, and their tendency to produce severe weather.
WSR-88D sites included in this study. Red lines indicate regions nearest to each radar. Upper-air soundings were also collected at three approximately collocated radiosonde and radar sampling sites along a diagonal transect through the center of the study area: Tampa, Peachtree City (KFFC) and Nashville (radar designation: KOHX; radiosonde designation: KBNA).
Citation: Journal of Applied Meteorology and Climatology 56, 11; 10.1175/JAMC-D-17-0005.1
2. Data and methods
The goal of the methodology is to develop a dataset of WFTs, defined here as storms forming in the absence of synoptic-scale lift and shear, which in turn yields generally small, short-lived, diurnally driven convection. This relationship between storm environment and storm morphology was demonstrated by Weisman and Klemp (1982) and has been incorporated into meteorological curricula during the intervening decades (e.g., Markowski and Richardson 2010; Bluestein 2013). However, storms in weakly forced environments can sometimes violate the morphology suggested by Weisman and Klemp (1982) by growing into groups of cells that do not propagate “in any systematic, predictable way” (Markowski and Richardson 2010, p. 203). Because these storms are nonetheless weakly forced, the methodology must also retain the flexibility to include such events when they are detected. Pulse thunderstorms, a key consideration of this work, often occur in the latter arrangement (Burgess and Lemon 1990; Miller and Mote 2017).
Based on the work referenced above, the following methodology operates on the hypothesis that storms with similar radar-inferred morphologies form, on average, in similar convective environments. Section 2a describes how the storms and their radar-inferred morphologies are detected, whereas section 2b details how the storm morphologies are related to the convective environment. Because thunderstorm morphology is observed at a much greater spatial and temporal resolution via radar than convective environments are observed via radiosondes, it is more feasible to designate WFTs based on their morphologies once their composite convective environment is known. Essentially, this approach establishes relationships between morphologically similar storms and their host environment, so that WFTs can be identified in the absence of proximate sounding data.
a. Deriving thunderstorms from radar imagery
Thunderstorms are identified using 15 yr (2001–15) of Weather Surveillance Radar-1988 Doppler (WSR-88D) level-III composite reflectivities for 30 sites in the southeastern United States (Fig. 1) during the warm season (May–September), totaling approximately 10 million radar scans. Consistent data were not available for two WSR-88Ds, KHTX (near Huntsville, Alabama) and KDGX (near Jackson, Mississippi), until 2002 and 2003, respectively. The composite reflectivity product, commonly used to identify areas of convection, is a 1-km-resolution gridded image extending 230 km from the radar site that indicates the strongest reflectivity detected at any altitude above each grid cell. Composite reflectivity is also advantageous in its ability to mitigate challenges associated with beam blockage in the southern Appalachian Mountains; however, it also introduces several limitations, which are discussed in section 2c. Between 2005 and 2006 all WSR-88Ds received a signal processing upgrade that allowed the radar to detect reflectivity with greater precision. The effect of this upgrade on any temporal trends will be discussed in section 3. Radar images for each site, generally available at ~5-min intervals when convection is present, were ordered chronologically and divided into consecutive 24-h periods from 1200 UTC 1 May to 1200 UTC 1 October of each year.
Connected-neighborhoods labeling was applied to grid cells with reflectivities of 40 dBZ or greater, a common reflectivity threshold for convection (e.g., Haberlie et al. 2015; Fabry et al. 2017), across all scans for each day. If gaps in radar coverage persisted for longer than 30 min, then the day was subdivided upon the coverage gap, and the labeling procedure was performed separately on each subdivision. In connected neighborhoods labeling, each convective grid cell (≥40 dBZ) is examined in three dimensions (x, y, t) to determine if any other cell in a 26-member neighborhood (a perfect 3 × 3 × 3 cube minus the center cell of interest) is also convective. The labeling technique extracts storms by expanding each group of spatiotemporally contiguous convective grid cells in all dimensions until no additional convective cells can be identified. Figures 2a–c illustrate a hypothetical example of the labeling process. For any redundant storms in regions of overlapping radar coverage, only the storm detected by the nearest radar, as shown by the polygons in Fig. 1, was retained.
Illustration of the connected neighborhoods labeling and Ward’s clustering workflow. (a) Beginning with a hypothetical five-image sequence, (b) reflectivities ≥ 40 dBZ are masked. (c) The images are ordered sequentially, contiguous regions of convective reflectivities are extracted as thunderstorms, and five morphological variables are calculated. (d) Storms are then clustered into statistically similar groups based on the five morphological characteristics, with each storm beginning as an individual cluster. (e) Storms are combined into groups by (f) iteratively merging the two clusters that will yield a new group with the smallest total within-cluster error of all possible mergers. Cluster centers are shown as black dots, and the total within-cluster error is proportional to the sum of all members’ separation from their parent center. In (f), Ward’s method combines the purple and green clusters from (e) because the new resulting cluster contains less total error than if, for instance, the green and blue clusters had been merged.
Citation: Journal of Applied Meteorology and Climatology 56, 11; 10.1175/JAMC-D-17-0005.1
For each storm, the time of first detection (measured in hours after 1200 UTC; TFD), duration (min; DUR), maximum reflectivity (dBZ; MAX REF), maximum size in a single image (number of grid cells; MAX SIZE), and solidity (SOL) were recorded. The solidity measure, a unitless ratio between 0 and 1, compares a storm’s average size in each image with its maximum size in any single image, MAX SIZE. This ratio helps distinguish storms that were morphologically uniform across their entire lifetime (large solidity) from those that exhibited considerable inter-image changes in their spatial footprint (small solidity). These variables were recorded given their role in traditional storm-mode definitions, and the use of similar measures in previous storm-mode classification studies (e.g., Lakshmanan and Smith 2009; Miller et al. 2015b). The latitude and longitude of the storm’s centroid in its first-detected radar scan were also recorded for use in constructing the spatial climatology.
A quality assurance and control procedure was implemented to improve the likelihood that each group of convective pixels represented a whole, legitimate convective event (i.e., remove ground clutter and partially detected events). “Storms” were removed from the dataset if one of the following five conditions applied: The storm 1) was ongoing at the beginning or end of a 24-h period (or an intra-24-h subdivision); 2) abutted the edge of a radar’s spatial coverage domain; 3) persisted fewer than 30 min, a routine duration requirement (e.g., Lock and Houston 2014; Burghardt et al. 2014); 4) was first detected within a 10 × 10 gridcell square centered on the radar site; or 5) was initially detected at a recurring first-detection location (first-detection centroid repeated ≥5 times) and never exceeded eight grid cells in MAX SIZE. Conditions 1 and 2 were designed to eliminate storms for which only partial data existed, whereas conditions 3, 4, and 5 were meant to reduce the amount of suspected ground clutter in the dataset. In eliminating partially detected storms, many large, organized convective systems that spanned multiple WSR-88D coverage domains were discarded by condition 2. However, because the goal of this study is to document WFTs, relatively stationary storms, the removal of spatially expansive convective systems was desirable.
Condition 5 was based on the idea that the landscape features responsible for ground clutter are fixed and will recur within the dataset. Of the unique first-detection initiation centroids, 89.5% were never repeated and 0.2% repeated five or more times. Given that first-detection centroids of any repetition were uncommon within the dataset, first-detection centroids recurring ≥5 times were viewed with even greater skepticism. The distribution of MAX SIZEs for storms originating from these highly active first-detection points (online supplementary Fig. 1) were disproportionately skewed toward small values, consistent with the appearance of ground clutter on radar. Condition 5 affected relatively few events, but was necessary to produce accurate graphic representations of the spatial climatology. The stipulations above yielded approximately 1.9 million viable storms.
Severe weather events (winds ≥ 26 m s−1, hail ≥ 0.56 cm in diameter, or a tornado) from NCEI’s Storm Data publication (https://www.ncdc.noaa.gov/stormevents/) were paired with storms if the report’s latitude and longitude coincided with any of a storm’s constituent grid cells and the report occurred after its TFD and before its time of last detection. Though several studies have documented data quality concerns within Storm Data’s severe weather reports (e.g., Weiss et al. 2002; Trapp et al. 2006; Miller et al. 2016), these discrepancies were judged unlikely to influence the broad distinction between hazardous and nonhazardous WFTs. For instance, by only requiring that the report occur after first detection and before last detection, the influence of the report’s uncertain timestamp (Williams et al. 1999) was minimized. However, spatial errors in Storm Data reports and/or unreported severe weather events may cause truly pulse thunderstorms to be classified as benign WFTs within the dataset. Overall, 91.0% (82 777 out of 90 955) of Storm Data reports were successfully paired with a convective reflectivity.
b. Designating weakly forced thunderstorms
Ward’s clustering (Ward 1963) was applied to identify natural groupings of the five storm morphology metrics listed above among all of the thunderstorms in the dataset, and then assess the composite convective environments of each morphological grouping. Storms belonging to clusters with small, short-lived, diurnally driven characteristics that also formed in weakly sheared, unstable environments will be considered WFTs. Ward’s clustering was selected because, unlike other hierarchical, agglomerative clustering techniques, it avoids the use of Euclidean distance to determine clusters. Instead, distance is expressed via the standardized total within-cluster error, which is proportional to the sum of the differences between each member of a cluster and the cluster’s mean. The process begins with all storms as separate clusters and iteratively combines the two groups whose merge will yield the new group with the smallest total within-cluster error of all the possible merges. Merging continues until a desired number of clusters is reached. Figures 2d–f depict a simplified 2D example (MAX SIZE and TFD only) using a small hypothetical dataset distributed similarly to the real dataset. Figure 2d shows each “storm” is plotted as its own cluster at the beginning of the procedure, and Fig. 2e depicts how Ward’s method would have grouped the storms after several hundred iterations. Figure 2f illustrates a single step in the hypothetical clustering process from five to four groups. Ward’s method merged the purple and green clusters from Fig. 2e because the new resulting cluster contains less total error than if, for instance, the green and blue clusters had been merged.
Given the dissimilar units of the spatial and temporal metrics, this Ward’s method was conceptually more appropriate, and it yielded the most interpretable results among other methods tested. See Gong and Richman (1995) for a comparison of various clustering procedures and their application in climate science. The cluster analysis was performed until 10 clusters remained. Using fewer than 10 clusters overgeneralized some clusters while using more than 10 clusters yielded no appreciable insight toward intracluster variability. Hereinafter, the 10 Ward’s clusters will be referenced as storm types and abbreviated T1 … T10 following the values shown in the Type column of Table 1.1
Counts of total and severe storms grouped into each of the 10 types and medians of the radar-derived metrics used to categorize them. When applicable, the interquartile range is shown in parentheses. WFT types are shown in boldface type. Given the limited precision of the composite reflectivity product, MAX REFs often congregated around discrete values yielding a narrow interquartile.
The results of the cluster analysis (Table 1) indicate that Ward’s clustering capably segregates the storms according to their spatial and temporal traits. However, because the definition of a WFT is intimately related to the storm environment, composite 1200 UTC soundings were generated at three approximately collocated radiosonde (http://esrl.noaa.gov/raobs/) and radar data collection sites (KBNA/KOHX, in Nashville, Tennessee; KFFC, in Peachtree City, Georgia; and KTBW in Tampa, Florida) along a diagonal transect through the center of the study area (Fig. 1) using the Sounding and Hodograph Analysis and Research Program in Python (SHARPpy) software package (Blumberg et al. 2017). Days at each site were stratified according to the thunderstorm type that contributed the largest total number of grid cells at that site on a given day (Table 2).
Number of 1200 UTC soundings categorized by prevalent storm type for KBNA, KFFC, and KTBW. WFT types are shown in boldface type.
Figure 3 depicts composite wind profiles (full skew T–logp diagrams are available as online supplementary Figs. 2–4), and Tables 3 and 4 present kinematic and thermodynamic variables at each site, respectively. The parameters in Tables 3 and 4 were either selected because of their traditional association with storm organization [0–6-km shear, mixed-layer CAPE (MLCAPE), forecast surface-based CAPE (SBCAPE), and low-level θe], or they were chosen to represent elements of the composite wind fields and storm morphology metrics that may not be captured by the other variables (0–8-km shear, 0–12-km max wind, and 0–12-km mean wind, TPW).
Composite KBNA, KFFC, and KTBW 1200 UTC vertical wind profiles for each storm type. Types 1–5 (labeled in boldface font) correspond to WFTs.
Citation: Journal of Applied Meteorology and Climatology 56, 11; 10.1175/JAMC-D-17-0005.1
Kinematic parameters (m s−1) of KBNA, KFFC, and KTBW 1200 UTC composite soundings for each storm type. WFT types are shown in boldface type. In several cases, an appreciable increase in the strength of the midlatitude westerlies occurred just above the 6-km level traditionally used to calculate bulk wind shear. Consequently, shear was calculated over the 0–8-km layer, which can also help infer the organization of deep moist convection (Markowski and Richardson 2010, p. 201).
Thermodynamic and moisture parameters of KBNA, KFFC, and KTBW 1200 UTC composite soundings for each storm type. Mean θe is calculated over the 1000–850-hPa layer, and MLCAPE was calculated using the lowest 100 hPa. WFT types are shown in boldface type.
When compared with the composite sounding metrics, seemingly minor differences between storm types correspond to appreciable variations in the convective environment. For instance, T7 is differentiated only by small decreases in MAX REF and MAX SIZE from T3, but formed in more stable, directionally sheared environments, and stronger flow environments at KBNA and KFFC. This comparison illustrates the proficiency of Ward’s clustering to separate morphologically similar storms whose small, short-lived characteristics were likely related to weaker instability on days with stronger forcing rather than the stronger instability on days with weaker forcing. We do not expect that each storm type corresponds to a specific forcing mechanism, only that types with/without synoptic lifting mechanisms can be differentiated through their shear environments.
Based on guidance from Tables 1–4 and Fig. 3, T3, T4, and T5 were judged to best holistically represent WFTs in the southeast United States because of their mostly short-lived, small, morphologically uniform, and diurnally driven storms in generally weak-shear, high-instability environments at the three transect points. Though T1 and T2 represent larger, stronger storms than the stereotypical WFT, upscale growth likely occurred in a relatively disorganized fashion without the aid of appreciable vertical wind shear. (See online supplementary Fig. 5 as an example.) These storms, referenced at the beginning of section 2, are consistent with the often multicell nature of WFTs (Miller and Mote 2017). For the sake of simplicity and by necessity, the WFT categorizations above are generalized across the entire southeast United States because only a handful of WSR-88Ds are collocated with radiosonde launch points. In total, 885 496 storms were classified as WFTs.
The synoptic patterns associated with WFT activity at KFFC, selected for its location in the center of the domain, were represented using the Earth Systems Research Laboratory’s North American Regional Reanalysis (Mesinger et al. 2006) daily compositing tool (http://www.esrl.noaa.gov/psd/cgi-bin/data/narr/plotday.pl/). For each storm type, synoptic composites of 500-hPa vector wind, 850-hPa geopotential height, and total precipitable water (TPW) were generated. The first two fields will help establish the presence (or absence) of any synoptic-scale forcing, whereas the third will indicate the moisture content of the atmosphere. These same variables have been previously used to discern disorganized thunderstorm environments (Miller et al. 2015b).
c. Limitations
In some cases, confident WFT/non-WFT classifications at one transect site might be less clear elsewhere in the domain; however, these cases represent a minority of the categorizations. For instance, T5 at KBNA (3.9% of all KBNA days) could be argued to represent a more organized kinematic environment, whereas T8 at KTBW is debatably representative of a weak-shear, high-CAPE regime (<1% of KTBW days). More importantly, the most frequent storm types at each site are among the most stereotypical WFT environments. All three composite soundings for T2, T3, and T4, collectively constituting 64.6% of all thunderstorm days along the transect, depict stereotypical weak-shear, high-instability regimes. Further, the relative frequencies of the storm-type days, also obey regional climatological expectations. The kinematically active T6 days occur most frequently at the northernmost site along the transect and gradually decrease with southward extent. Similarly, the higher-moisture T1 and T2 days are most frequent at the KTBW coastal site, and decrease with northward extent toward the interior of the southeast United States.
This method of WFT identification was also complicated by the presence of storms with 40-dBZ composite reflectivities that appeared small, short-lived, diurnally driven but were actually associated with a different process. For instance, bright banding within areas of non-weakly forced precipitation, especially at greater distances from the radar, may have presented a similar morphological signature to a WFT. This is one reason that all 30 WSR-88D sites in the Southeast were included and that storms were tracked using observations from the nearest radar. Another safeguard was the exclusion of storms that ventured too close to edge of the radar’s coverage domain. Figure 4a shows that nearly 100% of storms that qualified as WFTs possessed MAX REFs meeting or exceeding 50 dBZ. Even if bright banding did, in some cases, artificially bolster nonconvective reflectivities above the 40-dBZ limit, the storms appearing in the final WFT dataset almost certainly represented areas of convection according to most reflectivity-based definitions. [See Haberlie et al. (2015) for a thorough summary of such convection identification definitions.] However, when inspected via composite reflectivity, a WFT-like signature would have been noted. Because we did not individually inspect all >800 000 WFTs, nonconvective bright banding–related echoes, elevated convection, and stratiform echo maxima may have infiltrated the dataset in some cases.
(a) Percent of storms meeting or exceeding the MAX REF values on the x axis. (b) Distributions of 0–6-km wind shear values for each of the Ward’s clusters at KFFC. Examples of composite reflectivity imagery for (c) a case in which the WFT identification routine struggled and (d) a more representative WFT case are also shown.
Citation: Journal of Applied Meteorology and Climatology 56, 11; 10.1175/JAMC-D-17-0005.1
Similarly, areas of sporadic 40-dBZ reflectivity within a more organized and strongly forced mesoscale convective system might mimic the expected WFT morphology. Despite the measures taken to eliminate storm morphologies whose composite environments were too strongly forced (as inferred from vertical wind shear), some strongly forced convective structures may have circumvented the methodology. For instance, Fig. 4b shows the distribution of all 0–6-km wind shear values used to create the composite soundings at KFFC. Whereas the bulk of the distributions for the WFT clusters are similar to the composite wind fields, a few outliers are present. Figure 4c shows a radar image for one of the 0–6-km shear outliers from the T3 distribution (4 May 2002; 17.5 m s−1 0–6-km shear). In this case, small areas of convection in a non-weakly forced complex mimicked the morphology of a WFT, and deceived the methodology. However, this situation is an outlier (Fig. 4b). Figure 4d shows an example radar image from the middle of the same T3 distribution (12 June 2010; 4.1 m s−1 0–6-km shear) where the methodology capably captures WFTs. Future efforts might seek to improve upon these aspects of the classification methodology, particularly the false positives resulting from the use of composite reflectivity.
3. Results and discussion
a. Spatial and temporal distribution of WFT activity
Though the role of terrain in storm development is ultimately secondary to atmospheric processes, the following discussion leverages the large WFT sample size to identity underlying terrain influences on WFT first-detection patterns that become apparent when large-scale forcing is weak. (Section 3b will consider the synoptic-scale environments in more detail.) The general influence of landforms and terrain features in weakly forced regimes is well known, and the broad pattern of WFT activity shown in Fig. 5 is consistent with these features. Larger WFT first-detection densities sharply outline the southeast U.S. coast, especially south of KMHX (in Morehead City, North Carolina), reflecting the role of the heated land surface in building parcel instability and/or the efficiency of the sea breeze in initiating convection. WFTs are frequent over the Florida Peninsula, coastal areas, and southern Appalachian Mountains where the mesoscale sea breeze circulation (e.g., Pielke 1974) or preferential orographic heating (e.g., Hallenbeck 1922) can trigger convection in the absence of large-scale dynamical support. WFT frequency also diminishes in the northern tier of the domain where jet stream dynamics would support organized convection more frequently.
(a) WFT first-detection density in events per yr per 0.1° × 0.1° grid cell. (b) Notable features that are referenced in the text. Because the radar beam becomes less concentrated at increasing distance from the radar, small WFTs, detected by a more concentrated beam, are more common close to the radar, systematically driving the WFT climatology toward larger frequencies near radars and smaller concentrations farther from radars. To mitigate this effect, only WFTs with MAX SIZE ≥ 24 (the mode of the MAX SIZE distribution for all WFTs) were used to create the above images. The first-detection minima labeled in (b) were selected because they were evident in the absence of radar boundaries, but ranging effects may still have exacerbated their magnitude and shape.
Citation: Journal of Applied Meteorology and Climatology 56, 11; 10.1175/JAMC-D-17-0005.1
Beyond these general patterns, Fig. 5a shows that the regional WFT climatology is more nuanced than what might be assumed. For instance, there is a slight reduction in WFT frequency within a transition region between higher frequencies along the coast and likewise high frequencies along the dotted line in Fig. 5b (marked A). Previous work, noting similar signatures with a southeastern U.S. lightning climatology, attributes the increase in convection to preferential heating along the Atlantic fall line (Bentley and Stallins 2005), a geologic transition from the Piedmont region to the low-lying coastal plain. This southwest-to-northeast-oriented topographic relief would favor surface heating by creating a more orthogonal angle of incidence for incoming radiation beginning as soon as the sun rises above the horizon. Previous research has found this type of topography to favor convection in weakly sheared environments (Miller et al. 2015a). Similarly, Sims and Raman (2016) show that differential heating along accompanying soil-type boundaries in this zone support thermal circulations that also aid convection. The slight reduction of WFT activity on the fringe of the fall line transition region may also be related to surface divergence resulting from gradual upslope flow toward the fall line, a result that can be partially observed in the simulations of surface convergence by Kirshbaum et al. (2016). This recent modeling study showed that even the near-negligible relief surrounding the interior Mississippi Valley was still responsible for a nearly 50% decrease in the incidence of convection over the valley.
However, even within areas of generally increased frequency, the WFT climatology captures small, subtle minima. For instance, reduced WFT activity is noted over Lake Okeechobee, Florida, and Lake Pontchartrain, Louisiana, (marked B and C, respectively, in Fig. 5b) likely related to the stabilizing effect of the relatively cool water in comparison with the land surface temperature and/or surface divergence resulting from the lake breeze (Frank et al. 1967). On an even smaller scale, the French Broad River valley in western North Carolina (marked D in Fig. 5b), the Tennessee River valley in east Tennessee (marked E in Fig. 5b), and the Hiwassee and Notteley River valleys along the North Carolina–Georgia state line (marked F in Fig. 5b) are also coincident with a decrease in WFT activity. Such patterns were also noted in a high-resolution study of convective cloud formation in satellite imagery with their cause attributed to divergent flow at the surface resulting from the upslope component of the mountain–valley circulation (Gibson and Vonder Haar 1990). Other features of the WFT climatology support recent work that documents a decrease in radar echoes over the Mississippi Delta (marked G in Fig. 5b; Kirshbaum et al. 2016) and a relative maximum in unorganized precipitation offshore of the North Carolina coast (marked H in Fig. 5b; Rickenbach et al. 2015).
Two additional large relative minima are also apparent in the climatology that are not readily explained: east-central Mississippi and west-central Alabama (marked I in Fig. 5b) and the broad corridor from northeast Georgia (marked J in Fig. 5b) stretching northeast to central North Carolina. The first relative void is discernable in Gibson and Vonder Haar’s (1990) analysis of both shallow and deep satellite-derived convective cloud frequency. They assert that the feature is not a result of random vertical motions, but seeing no viable terrain features that might contribute to the signature, they only loosely speculate that this minimum is tied to land-use/land-cover patterns. The same minimum is also clearly noted in a similar, but more recent, study of convective cloud activity by Gambill and Mecikalski (2011). Though the purpose of their work was to investigate ties between convective clouds and land-cover type, the analysis is conducted on an aggregate level over the whole southeast United States, and does not consider this minimum specifically.
Similarly, the relative void roughly paralleling the southern Appalachians from northeast Georgia to central North Carolina possesses no obvious terrain or land-cover heterogeneities. Yet, the same minima can be observed in radar analyses of convective frequency by both Outlaw and Murphy (2000) and Fabry et al. (2017). Rickenbach et al. (2015) also document this feature; however, they note that convection tends to fill the void later in the day relative to its surroundings, rather than be altogether absent. The steep relief posed by the Appalachian Mountains, though likely suppressing convection in the immediately surrounding flat terrain through surface divergence associated with the valley–mountain circulation, would seem unlikely to suppress WFT activity >100 km to the south. One initial hypothesis is that the slightly larger proportion of cropland in this region, a land-cover type associated with decreases in convective cloud percentage by Gambill and Mecikalski (2011), may discourage WFT activity. Alternatively, stabilized air formed by convection within the active WFT regions on either side may lead to weak subsidence over the minimum. Further research is required to more directly establish which, if either, of these processes contributes to the first-detection minima.
Consistent with expectations, Fig. 6a shows that WFTs are most common in July and August with smaller frequencies in May and September. Straddling the core of the warm season, these months represent transition periods from spring to summer and summer to fall when midlatitude westerly flow strengthens into the cold season and instability is weaker. Clear interannual variability in the number of WFTs is also been observed over the past 15 yr (Fig. 6b) with annual deviations near 20% in some years (maximum negative anomaly, −19.8% in 2002; maximum positive anomaly, 20.7% in 2007). The visual trend in the number of WFTs suggests a possible shift toward more numerous WFTs since 2006. However, the WFT-frequency transition seemingly corresponds to a WSR-88D processing upgrade (Patel and Macemon 2004). The increased data precision available post-2006 would allow for more precise detection of 40-dBZ echoes, leading previously nonqualifying cells to perhaps satisfy the convective reflectivity threshold. The possible post-2006 shift in WFT activity should be treated with skepticism until future research can examine the effects of data quality improvements in greater detail.
Distribution of WFT frequency by (a) month and (b) year. The annual frequency is expressed in terms of the percent anomaly from the 15-yr Southeast average number of WFTs that occurred each year. For years 2002 and 2003, the 13- and 14-yr averages for KDGX and KHTX were added to the Southeast total as a rough correction for the absence of data from those two sites.
Citation: Journal of Applied Meteorology and Climatology 56, 11; 10.1175/JAMC-D-17-0005.1
b. Synoptic patterns associated with WFT activity
Figure 7 characterizes the 500-hPa winds, 850-hPa geopotential heights, and TPWs for the same dates used to compute the KFFC composite soundings in section 2b. (Online supplementary Figs. 6 and 7 depict the 25th and 75th percentiles for the two scalar fields, 850-hPa height and TPW.) Note that these days may not have been conducive for WFTs over the whole Southeast; rather, they were tied to WFT-favorable conditions in the polygon containing KFFC in Fig. 1. The composites reveal two modes of variability that favor WFT activity near the center of the study domain: 1) the expansion of anticyclonic flow related to the Bermuda high from the east and 2) the intrusion of high pressure over the southern Great Plains from the west. T1, T2, and T3 are representative of the first mode with an anticyclonic circulation over the Bahamas expanding across the southern tier of the Southeast. Meanwhile, T4 and T5 correspond to the second mode with midlevel ridging over the central United States placing KFFC in a region of northwesterly 500-hPa flow.
Synoptic composite maps of (left) 500-hPa mean wind, (center) 850-hPa height, and (right) TPW for the 5 WFT types. The 25th- and 75th-percentile maps for 850-hPa height and TPW are available as online supplementary Figs. 6 and 7, respectively.
Citation: Journal of Applied Meteorology and Climatology 56, 11; 10.1175/JAMC-D-17-0005.1
Closer to the surface, the influence of the Bermuda high is more apparent. The 850-hPa geopotential heights show that all five WFT clusters are characterized by the intrusion of the Bermuda high into the southeastern United States. Similar 850-hPa height patterns have been found on days characterized by lightning-inferred WFT activity in southwest Virginia (Miller et al. 2015b) as well as days associated with urban-initiated convection in Atlanta, Georgia (Bentley et al. 2012). On T3 days, the westward expansion of the Bermuda high is similar to the other WFT clusters (Fig. 7); however, its strength is weaker over the Southeast, with the 1564-m contour situated entirely over the western Atlantic Ocean. A similar regime to T3, the largest WFT cluster by number of storms (Table 1), was identified by Diem (2013) as being conducive to a disproportionate number of rainfall days in the Atlanta metropolitan area. On larger time scales, the position of the Bermuda high may exercise a control on the frequency of disorganized convection in the Southeast and contribute to the interannual variability seen in Fig. 6b. The 850-hPa-height composites (Fig. 7) may represent a middle ground whereby disorganized convection dominates. If the Bermuda high advances too far westward all convection is suppressed, and if it drifts too far eastward, meridional flow develops, and larger, more productive precipitation systems form (Stahle and Cleaveland 1992).
Within each synoptic mode of variability, moisture availability appears to exercise an additional influence on WFT frequency and size. When high pressure from the western Atlantic dominated (mode 1), storms grew larger and were longer-lived on the most moist subset of these days (T1 and T2). In contrast, when TPW values were smaller yet the circulation pattern remained unchanged (T3), storms were more numerous, but remained smaller and shorter-lived. The same is true when high pressure from the southern Great Plains encroached over the Southeast (mode 2). T4 days, with higher TPW than T5 days, were associated with fewer, but larger and longer-lived, storms.
Interactions between moisture and convection are complex, and readers interested in a more comprehensive account of the relationship are directed toward Sherwood et al. (2010) and James and Markowski (2010). However, there are some basic, intuitive relationships that may help explain the formation of the larger storms in the higher-TPW environments. When present at the surface, higher humidity promotes convection by increasing CAPE, and when moisture extends above the surface, it can mitigate the stabilizing effect of entrainment (e.g., Jorgensen and LeMone 1989). Table 4 shows evidence of the former in that KFFC’s composite sounding CAPE calculations mirror the relative increases–decreases of TPW. However, the differences in MAX SIZE and DUR between T2 and T3, both demonstrated to form in similar synoptic regimes, are very large, whereas the difference in forecast SBCAPE is only roughly 50 J kg−1. Such a relatively minor difference in instability is unlikely to affect such a dramatic shift in storm morphology.
In both the lower and midtroposphere, higher humidity has been tied to increased water loading in thunderstorms by reducing evaporation due to entrainment (Wissmeier and Goler 2009) though this relationship is also dependent on CAPE (James and Markowski 2010). Because radar reflectivity is proportional to the size and number of hydrometeors, larger or more abundant hydrometeors would favor thunderstorms reaching the 40-dBZ threshold used to define areas of convection in section 2a, driving increases in MAX SIZE and possibly DUR. Table 5 shows that this is indeed the case for the KFFC composite soundings. On T1 and T2 days, associated with the largest and longest-lived WFTs, mean mixing ratios between 1000 and 850 hPa are only roughly 3% larger than on T3, T4, and T5 days. However, in the low-to-midtroposphere (850–500 hPa) the difference in mean mixing ratio increases to 21%. This result is consistent with James and Markowski (2010) who found that in their 1500 J kg−1 CAPE simulations, the duration of convection was extremely sensitive to midlevel RH, with lower RHs suppressing convective intensity and duration.
Mean relative humidity (RH) and mixing ratio (MR) for both the lower (1000–850 hPa) and middle (850–500 hPa) troposphere from KFFC’s composite soundings.
c. Characteristics of pulse thunderstorms
Constituting just 0.6% of all WFTs, pulse thunderstorms represented a very small fraction of the dataset. However, this small subset is disproportionately concentrated among the two least frequent storm types. Though T1 and T2 only account for 2% of all WFTs, 66% all pulse thunderstorms are associated with these two groups (Table 1). In contrast, only 0.03% of pulse thunderstorms are associated with T3, the most frequent WFT type. Table 6 compares the relative spatial and temporal storm metrics of pulse thunderstorms and all WFTs. Pulse storms are considerably larger and longer-lived, suggesting that most WFT severe weather episodes occur with cells that are members of a larger disorganized group. This finding is consistent with previous accounts of pulse thunderstorm morphologies described at the beginning of section 2.
Comparison of spatial and temporal metrics for pulse thunderstorms to the all-inclusive set of WFTs. The medians for each storm attribute are shown, and when applicable the interquartile range is shown in parentheses.
Figure 8 shows the first-detection locations of all pulse thunderstorms during the 15-yr study period. The patterns shown in this image must be interpreted with caution given that previous research has tied the frequency of severe weather reports to population density and NWS severe weather warning issuance (Weiss et al. 2002). Nonetheless, when compared with the annual first-detection density of all WFTs (Fig. 5a), the spatial distribution of pulse thunderstorms shows clear departures from the broader set of all WFTs. The Florida Peninsula and Gulf Coast, maxima of WFT first-detection density, are relatively inactive areas of pulse thunderstorm activity. Alternatively, the regions of greatest pulse thunderstorm first-detection density are displaced farther north in the domain, namely, the western Carolinas and western Virginia along the Blue Ridge Mountains. This region was also identified by Harrison and Karstens (2017) as a local maximum of NWS severe weather products that also referenced slower storms speeds than their neighboring regions. The reason for the concentration of pulse thunderstorm first detections in this region is not immediately apparent. Variations in the thermodynamic storm environment, such as the height of the freezing level, may favor severe convection farther north in the domain.
First detection density of all 5378 pulse thunderstorms within the dataset. The number in the legend represents the total number of pulse thunderstorms during the entire 15-yr study period. WSR-88D sites are depicted as black dots.
Citation: Journal of Applied Meteorology and Climatology 56, 11; 10.1175/JAMC-D-17-0005.1
As discussed in section 3b, T1 and T2 days demonstrate similar in 850-hPa heights to the other WFT days, but are generally differentiated by greater forecast SBCAPE (Table 4) and increased moisture (Fig. 7). Though the concentration of pulse thunderstorms in these two clusters may be reflexively attributed to instability, the SBCAPE differences in the composite soundings do not appear large enough to account for the concentration of pulse thunderstorms in T1 and T2 versus T3, T4, and T5. Once again, the more pronounced variations in moisture, particularly in the midlevels, could offer an alternative explanation. Field observations during the Cirrus Regional Study of Tropical Anvils and Cirrus Layers–Florida-Area Cirrus Experiment (CRYSTAL-FACE) found that a 20% increase in mean mixing ratio between 750 and 500 hPa corresponded to a 1-km increase in cloud-top penetration even when CAPE remained unchanged (Sherwood et al. 2004). This increase is nearly identical to the 21% greater T1 and T2 midlevel mixing ratios found in section 3b. Cloud-top heights often serve as a proxy for convective intensity (e.g., Adler and Negri 1988; Bedka 2011). Thus, the association of pulse thunderstorms with higher-TPW clusters may be partially explained by greater midlevel moisture favoring higher cloud tops and, by extension, stronger updrafts.
4. Conclusions
We mined 15 years of radar observations between 2001 and 2015 in the southeast United States for instances of spatiotemporally contiguous convective echoes, of which 885 496 were deemed to represent WFTs. Pronounced spatial variations in WFT and pulse thunderstorm first-detection density were evident even in a region where WFTs are generally described as ubiquitous. However, the spatial focus of pulse thunderstorms, the Blue Ridge Mountains, was significantly displaced from the areas of greatest all-inclusive WFT activity: the Florida Peninsula and Gulf Coast.
WFT environments near the center of the domain formed via two distinct modes of variability. Both modes were characterized by two centers of anticyclonic flow in the midlevels but differed in the direction from which the dominant high-pressure encroached upon the Southeast. With the first mode, the Bermuda high circulation expands westward over the Florida Peninsula, whereas with the second, anticyclonic 500-hPa winds shift east from the southern Great Plains region placing the Southeast in a region of weak northwest flow. In both modes, greater moisture availability, rather than instability, was associated with larger, longer-lasting WFTs, which were also responsible for the majority of severe weather reports.
Combining the spatial pattern of WFT development with the links to larger-scale features, forecasts of both the location and severity of WFTs may be improved. With a 15-yr WFT dataset and its derived climatology now available, operational forecasters can begin to better recognize the moisture and circulation patterns most conducive to WFTs, particularly pulse thunderstorms. Forecasters may begin considering whether to tailor the probability of precipitation in weakly forced environments based on the local, presumably landscape-driven, effects depicted in Fig. 5. This is also a valuable result for city planners tasked with siting infrastructure that may be adversely impacted by thunderstorm activity (e.g., an airport or sports stadium), or alternatively, may desire preferential storm activity (e.g., a reservoir). Future research needs to examine the near-storm environment of WFTs, especially pulse thunderstorms, in a much more comprehensive and statistically robust manner. In this paper, the parameters presented were derived from a composite sounding rather than a distribution of many model-derived proxy soundings, an approach that future researchers might consider.
Acknowledgments
The authors thank Dr. Mace Bentley and three anonymous reviewers for their helpful comments on earlier versions of this manuscript.
REFERENCES
Adler, R. F., and A. J. Negri, 1988: A satellite infrared technique to estimate tropical convective and stratiform rainfall. J. Appl. Meteor., 27, 30–51, doi:10.1175/1520-0450(1988)027<0030:ASITTE>2.0.CO;2.
Ashley, W. S., and C. W. Gilson, 2009: A reassessment of U.S. lightning mortality. Bull. Amer. Meteor. Soc., 90, 1501–1518, doi:10.1175/2009BAMS2765.1.
Ashley, W. S., M. L. Bentley, and J. A. Stallins, 2012: Urban-induced thunderstorm modification in the southeast United States. Climatic Change, 113, 481–498, doi:10.1007/s10584-011-0324-1.
Bedka, K. M., 2011: Overshooting cloud top detections using MSG SEVIRI Infrared brightness temperatures and their relationship to severe weather over Europe. Atmos. Res., 99, 175–189, doi:10.1016/j.atmosres.2010.10.001.
Bentley, M. L., and J. A. Stallins, 2005: Climatology of cloud-to-ground lightning in Georgia, USA, 1992–2003. Int. J. Climatol., 25, 1979–1996, doi:10.1002/joc.1227.
Bentley, M. L., J. A. Stallins, and W. S. Ashley, 2012: Synoptic environments favourable for urban convection in Atlanta, Georgia. Int. J. Climatol., 32, 1287–1294, doi:10.1002/joc.2344.
Bluestein, H. B., 2013: Severe Convective Storms and Tornadoes: Observations and Dynamics. Springer, 456 pp.
Blumberg, W. G., K. T. Halbert, T. A. Supinie, P. T. Marsh, R. L. Thompson, and J. A. Hart, 2017: SHARPpy: An open source sounding analysis toolkit for the atmospheric sciences. Bull. Amer. Meteor. Soc., 98, 1625–1636, doi:10.1175/BAMS-D-15-00309.1.
Bruick, Z. S., and C. D. Karstens, 2017: An investigation of local and national NWS warning outbreaks for severe convective events. J. Oper. Meteor., 5, 14–25, doi:10.15191/nwajom.2017.0502.
Burgess, D. W., and L. R. Lemon, 1990: Severe thunderstorm detection by radar. Radar in Meteorology, D. Atlas, Ed., Springer, 619–647.
Burghardt, B. J., C. Evans, and P. J. Roebber, 2014: Assessing the predictability of convection initiation in the high plains using an object-based approach. Wea. Forecasting, 29, 403–418, doi:10.1175/WAF-D-13-00089.1.
de Szoeke, S. P., J. B. Edson, J. R. Marion, C. W. Fairall, and L. Bariteau, 2015: The MJO and air-sea interaction in TOGA COARE and DYNAMO. J. Climate, 28, 597–622, doi:10.1175/JCLI-D-14-00477.1.
Diem, J. E., 2013: Influences of the Bermuda high and atmospheric moistening on changes in summer rainfall in the Atlanta, Georgia region, USA. Int. J. Climatol., 33, 160–172, doi:10.1002/joc.3421.
Diffenbaugh, N. S., M. Scherer, and R. J. Trapp, 2013: Robust increases in severe thunderstorm environments in response to greenhouse forcing. Proc. Natl. Acad. Sci. USA, 110, 16 361–16 366, doi:10.1073/pnas.1307758110.
Fabry, F., 2006: The spatial variability of moisture in the boundary layer and its effect on convection initiation: Project-long characterization. Mon. Wea. Rev., 134, 79–91, doi:10.1175/MWR3055.1.
Fabry, F., V. Meunier, B. Puigdomènech Treserras, A. Cournoyer, and B. Nelson, 2017: On the climatological use of radar data mosaics: Possibilities and challenges. Bull. Amer. Meteor. Soc., 98, 2135–2148, doi:10.1175/BAMS-D-15-00256.1.
Ford, T. W., S. M. Quiring, O. W. Frauenfeld, and A. D. Rapp, 2015: Synoptic conditions related to soil moisture-atmosphere interactions and unorganized convection in Oklahoma. J. Geophys. Res. Atmos., 120, 11 519–11 535, doi:10.1002/2015JD023975.
Frank, N. L., P. L. Moore, and G. E. Fisher, 1967: Summer shower distribution over the Florida Peninsula as deduced from digitized radar data. J. Appl. Meteor., 6, 309–316, doi:10.1175/1520-0450(1967)006<0309:SSDOTF>2.0.CO;2.
Frye, J. D., and T. L. Mote, 2010: Convection initiation along soil moisture boundaries in the southern Great Plains. Mon. Wea. Rev., 138, 1140–1151, doi:10.1175/2009MWR2865.1.
Gambill, L. D., and J. R. Mecikalski, 2011: A satellite-based summer convective cloud frequency analysis over the southeastern United States. J. Appl. Meteor. Climatol., 50, 1756–1769, doi:10.1175/2010JAMC2559.1.
Gibson, H. M., and T. H. Vonder Haar, 1990: Cloud and convection frequencies over the southeast United States as related to small-scale geographic features. Mon. Wea. Rev., 118, 2215–2227, doi:10.1175/1520-0493(1990)118<2215:CACFOT>2.0.CO;2.
Gong, X., and M. B. Richman, 1995: On the application of cluster analysis to growing season precipitation data in North America east of the Rockies. J. Climate, 8, 897–931, doi:10.1175/1520-0442(1995)008<0897:OTAOCA>2.0.CO;2.
Guillot, E. M., T. M. Smith, V. Lakshmanan, K. L. Elmore, D. W. Burgess, and G. J. Stumpf, 2008: Tornado and severe thunderstorm warning forecast skill and its relationship to storm type. 24th Int. Conf. Interactive Information Processing Systems for Meteorology, Oceanography, and Hydrology, New Orleans, LA, Amer. Meteor. Soc., 4A.3, http://ams.confex.com/ams/pdfpapers/132244.pdf.
Haberlie, A. M., W. S. Ashley, and T. J. Pingel, 2015: The effect of urbanisation on the climatology of thunderstorm initiation. Quart. J. Roy. Meteor. Soc., 141, 663–675, doi:10.1002/qj.2499.
Hallenbeck, C., 1922: The topographic thunderstorm. Mon. Wea. Rev., 50, 284–287, doi:10.1175/1520-0493(1922)50<284:TTT>2.0.CO;2.
Harrison, D. R., and C. D. Karstens, 2017: A climatology of operational storm-based warnings: A geospatial analysis. Wea. Forecasting, 32, 47–60, doi:10.1175/WAF-D-15-0146.1.
Ingram, K. T., K. Dow, L. Carter, and J. Anderson, Eds., 2013: Climate of the Southeast United States: Variability, Change, Impacts, and Vulnerability. Island Press, 358 pp., https://www.sercc.com/ClimateoftheSoutheastUnitedStates.pdf.
James, R. P., and P. M. Markowski, 2010: A numerical investigation of the effects of dry air aloft on deep convection. Mon. Wea. Rev., 138, 140–161, doi:10.1175/2009MWR3018.1.
Johnson, R. H., T. M. Rickenbach, S. A. Rutledge, P. E. Ciesielski, and W. H. Schubert, 1999: Trimodal characteristics of tropical convection. J. Climate, 12, 2397–2418, doi:10.1175/1520-0442(1999)012<2397:TCOTC>2.0.CO;2.
Jorgensen, D. P., and M. A. LeMone, 1989: Vertical velocity characteristics of oceanic convection. J. Atmos. Sci., 46, 621–640, doi:10.1175/1520-0469(1989)046<0621:VVCOOC>2.0.CO;2.
Kirshbaum, D. J., F. Fabry, and Q. Cazenave, 2016: The Mississippi Valley convection minimum on summer afternoons: Observations and numerical simulations. Mon. Wea. Rev., 144, 263–272, doi:10.1175/MWR-D-15-0238.1.
Lackmann, G., 2011: Midlatitude Synoptic Meteorology. 1st ed. Amer. Meteor. Soc., 345 pp.
Lakshmanan, V., and T. Smith, 2009: Data mining storm attributes from spatial grids. J. Atmos. Oceanic Technol., 26, 2353–2365, doi:10.1175/2009JTECHA1257.1.
Lee, T., and Coauthors, 2016: An observational study of convection initiation under quiescent conditions in the southeastern United States. 18th Symp. on Meteorological Observation and Instrumentation, New Orleans, LA, Amer. Meteor. Soc., 6A.3, https://ams.confex.com/ams/96Annual/webprogram/Paper282669.html.
Lock, N. A., and A. L. Houston, 2014: Empirical examination of the factors regulating thunderstorm initiation. Mon. Wea. Rev., 142, 240–258, doi:10.1175/MWR-D-13-00082.1.
Markowski, P., and Y. Richardson, 2010: Mesoscale Meteorology in Midlatitudes. 1st ed. Wiley-Blackwell, 407 pp.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360, doi:10.1175/BAMS-87-3-343.
Miller, P. W., and T. L. Mote, 2017: Standardizing the definition of a “pulse” thunderstorm. Bull. Amer. Meteor. Soc., 98, 905–913, doi:10.1175/BAMS-D-16-0064.1.
Miller, P. W., A. Ellis, and S. Keighton, 2015a: Spatial distribution of lightning associated with low-shear thunderstorm environments in the central Appalachians region. Phys. Geogr., 36, 127–141, doi:10.1080/02723646.2015.1011257.
Miller, P. W., A. Ellis, and S. Keighton, 2015b: A preliminary assessment of using spatiotemporal lightning patterns for a binary classification of thunderstorm mode. Wea. Forecasting, 30, 38–56, doi:10.1175/WAF-D-14-00024.1.
Miller, P. W., A. W. Black, C. A. Williams, and J. A. Knox, 2016: Quantitative assessment of human wind speed overestimation. J. Appl. Meteor. Climatol., 55, 1009–1020, doi:10.1175/JAMC-D-15-0259.1.
Mote, T. L., M. C. Lacke, and J. M. Shepherd, 2007: Radar signatures of the urban effect on precipitation distribution: A case study for Atlanta, Georgia. Geophys. Res. Lett., 34, L20710, doi:10.1029/2007GL031903.
Murphy, M. S., and C. E. Konrad, 2005: Spatial and temporal patterns of thunderstorm events that produce cloud-to-ground lightning in the interior southeastern United States. Mon. Wea. Rev., 133, 1417–1430, doi:10.1175/MWR2924.1.
Niyogi, D., P. Pyle, M. Lei, S. Pal Arya, C. M. Kishtawal, M. Shepherd, F. Chen, and B. Wolfe, 2011: Urban modification of thunderstorms: An observational storm climatology and model case study for the Indianapolis urban region. J. Appl. Meteor. Climatol., 50, 1129–1144, doi:10.1175/2010JAMC1836.1.
Outlaw, D. E., and M. P. Murphy, 2000: A radar-based climatology of July convective initiation in Georgia and surrounding area. NWS Eastern Region Tech. Attachment 2000-04, 24 pp., http://www.weather.gov/media/erh/ta2000-04.pdf.
Patel, N., and R. W. Macemon, 2004: NEXRAD Open Radar Data Acquisition (ORDA) signal processing & signal path. 20th Int. Conf. on Interactive Information and Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology, Seattle, WA, Amer. Meteor. Soc., 5.4, https://ams.confex.com/ams/pdfpapers/70926.pdf.
Pielke, R. A., 1974: A three-dimensional numerical model of the sea breezes over south Florida. Mon. Wea. Rev., 102, 115–139, doi:10.1175/1520-0493(1974)102<0115:ATDNMO>2.0.CO;2.
Rickenbach, T. M., R. Nieto‐Ferreira, C. Zarzar, and B. Nelson, 2015: A seasonal and diurnal climatology of precipitation organization in the southeastern United States. Quart. J. Roy. Meteor. Soc., 141, 1938–1956, doi:10.1002/qj.2500.
Schumacher, C., and R. A. Houze, 2003: The TRMM precipitation radar’s view of shallow, isolated rain. J. Appl. Meteor., 42, 1519–1524, doi:10.1175/1520-0450(2003)042<1519:TTPRVO>2.0.CO;2.
Senkbeil, J. C., M. E. Saunders, and B. Taylor, 2017: Changes in summer weather type frequency in eastern North America. Ann. Assoc. Amer. Geogr., 107, 1–17, doi:10.1080/24694452.2017.1295839.
Shepherd, J. M., 2005: A review of current investigations of urban-induced rainfall and recommendations for the future. Earth Interact., 9, doi:10.1175/EI156.1.
Sherwood, S. C., P. Minnis, and M. McGill, 2004: Deep convective cloud-top heights and their thermodynamic control during CRYSTAL-FACE. J. Geophys. Res., 109, D20119, doi:10.1029/2004JD004811.
Sherwood, S. C., R. Roca, T. M. Weckwerth, and N. G. Andronova, 2010: Tropospheric water vapor, convection, and climate. Rev. Geophys., 48, RG2001, doi:10.1029/2009RG000301.
Sims, A. P., and S. Raman, 2016: Interaction between two distinct mesoscale circulations during summer in the coastal region of eastern USA. Bound.-Layer Meteor., 160, 113–132, doi:10.1007/s10546-015-0125-6.
Stahle, D. W., and M. K. Cleaveland, 1992: Reconstruction and analysis of spring rainfall over the southeastern U.S. for the past 1000 years. Bull. Amer. Meteor. Soc., 73, 1947–1961, doi:10.1175/1520-0477(1992)073<1947:RAAOSR>2.0.CO;2.
Trapp, R. J., D. M. Wheatley, N. T. Atkins, R. W. Przybylinski, and R. Wolf, 2006: Buyer beware: Some words of caution on the use of severe wind reports in postevent assessment and research. Wea. Forecasting, 21, 408–415, doi:10.1175/WAF925.1.
Ward, J. H., 1963: Hierarchical grouping to optimize an objective function. J. Amer. Stat. Assoc., 58, 236–244, doi:10.1080/01621459.1963.10500845.
Weisman, M. L., and J. B. Klemp, 1982: The dependence of numerically simulated convective storms on vertical wind shear and buoyancy. Mon. Wea. Rev., 110, 504–520, doi:10.1175/1520-0493(1982)110<0504:TDONSC>2.0.CO;2.
Weiss, S. J., J. Hart, and P. Janish, 2002: An examination of severe thunderstorm wind report climatology: 1970–1999. 21st Conf. Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., 11B.2, https://ams.confex.com/ams/pdfpapers/47494.pdf.
Williams, E., and Coauthors, 1999: The behavior of total lightning activity in severe Florida thunderstorms. Atmos. Res., 51, 245–265, doi:10.1016/S0169-8095(99)00011-3.
Wissmeier, U., and R. Goler, 2009: A comparison of tropical and midlatitude thunderstorm evolution in response to wind shear. J. Atmos. Sci., 66, 2385–2401, doi:10.1175/2009JAS2963.1.
Ye, H., E. J. Fetzer, S. Wong, and B. H. Lambrigtsen, 2017: Rapid decadal convective precipitation increase over Eurasia during the last three decades of the 20th century. Sci. Adv., 3, e1600944, doi:10.1126/sciadv.1600944.
The sequencing of the storm types was modified from its original output to improve the interpretability of the results.