Abstract

Heavy precipitation and flash flooding have been extensively studied in the central United States, but less so in the Northeast. This study examines 187 warm-season flash flood events identified in Storm Data to better understand the structure of the precipitation systems that cause flash flooding in the Northeast. Based on the organization and movement of these systems on radar, the events are classified into one of four categories—back-building, linear, multiple, and other/size—and then further classified into subtypes for each category. Eight of these subtypes were not previously recognized in the literature. The back-building events were the most common, followed by the multiple, other/size, and linear types. The linear event types appear to produce flash flooding less commonly in the Northeast than in other regions. In general, the subtypes producing the highest precipitation estimates are those whose structures are most conducive to a long duration of sustained moderate to heavy rainfall. The event types were found to differ from those in the central United States in that the events were more often found to be more disorganized in the Northeast. One event type in particular, back-building with merging features, while not more disorganized than the previously recognized event types, offers promise for improved forecasting because its radar signature makes the duration of sustained heavy precipitation potentially easier to predict.

1. Introduction

Flash floods remain one of the greatest weather hazards, despite decades of research. A significant hindrance to improved flash flood forecasting is that quantitative precipitation forecasting remains perhaps the greatest challenge in short-term weather forecasting (Fritsch et al. 1998; Cuo et al. 2011). Heavy precipitation may arise from a wide variety of atmospheric conditions, and flash floods may occur even on days that do not appear to be particularly conducive to heavy rainfall (Doswell et al. 1996).

To improve the understanding of heavy precipitation and flash floods, many recent studies have focused on classifying and elucidating the organization and atmospheric conditions associated with these events, so as to improve the awareness of flash flood hazards in real-time forecasting. Following early work on the organization of mesoscale convective systems (MCSs) by Bluestein and Jain (1985), Bluestein et al. (1987), and Houze et al. (1990), Parker and Johnson (2000) examined 88 linear MCSs in the central United States during May 1996 and May 1997, classifying them based upon the relative position and motion of a convective line and its accompanying stratiform precipitation: trailing stratiform, parallel stratiform, and leading stratiform.

More recently, Schumacher and Johnson (2006, hereafter SJ2006) classified 184 heavy precipitation events based on the organization of the precipitation and described the prominent surface and upper-air features associated with each type of event, including new categories of precipitation organization: areas of quasi-stationary/back-building precipitation and MCSs containing a training line of convection adjoined by an area of stratiform precipitation. Gallus et al. (2008) and Lombardo and Colle (2010) classified events in the central and northeastern United States, respectively, into categories of cellular convection, nonlinear systems, and linear systems [including classes similar to the linear systems studied by Parker and Johnson (2000) and SJ2006].

This study seeks to classify the organization of precipitation and describe estimated precipitation amounts associated with flash flood events in the Northeast for the warm seasons of the years 2003–07. There is reason to suspect that the most common organizational modes of precipitation differ in the Northeast as compared to other regions of the country. Fritsch et al. (1986) examined the precipitation resulting from mesoscale convective complexes (MCCs) and found that these systems accounted for 30%–70% of the precipitation from April through September over a large area spanning from the Rocky Mountains to the Mississippi River. These systems, however, rarely travel east of the Appalachian Mountains (Tollerud et al. 1987). The MCCs studied by Fritsch et al. (1986) tend to be overnight phenomena, while ordinary convection is most common in the late afternoon hours, fed by diurnal heating (Davis 2001). In examining the timing of heavy rainfall, Winkler et al. (1988) found that the heaviest precipitation tends to occur in the late afternoon to early evening hours in the eastern United States, but in the overnight hours in the central United States.

In short, these findings suggest that heavy precipitation in the Northeast may more commonly result from localized, diurnally forced convection, and be less commonly associated with larger, more organized MCSs than in other regions of the country, especially the Midwest and southern plains. Recent work by Lombardo and Colle (2010) supports this assertion, as they found that the Northeast appears to produce more frequent nonlinear and cellular precipitation morphologies and less frequent linear morphologies (i.e., well-organized MCSs) than the Midwest.

The following research questions drive the examination of flash floods in the Northeast represented in this study: Are the predominant patterns of the organization of flash-flood-producing precipitation different in the Northeast than in the Midwest, southern plains, and other parts of the United States, as the work of Lombardo and Colle (2010) suggests? Are there morphologies of precipitation that have not been recognized as flash flood producers in the literature? And finally, which types of events tend to produce the most precipitation and cause the most significant impact in the Northeast?

Section 2 describes the data and methodology used in this study. Section 3 presents the classification system used in this study, describing the appearance on radar of the event types newly identified in this study. Section 4 analyzes the relative frequency and precipitation characteristics of each event type. Section 5 presents the environmental conditions associated with some types of events. Section 6 summarizes the key findings of this study.

2. Data and methodology

The overarching objective of this contribution is to study flash floods in the Northeast that result from warm-season midlatitude convection in order to understand how the organization and environmental conditions associated with the precipitation differs from the organization and environmental conditions of flash-flood-producing convection in other regions. For this study, the Northeast is defined as the states of Pennsylvania, New York, New Jersey, Connecticut, Rhode Island, Massachusetts, Vermont, New Hampshire, and Maine.

To assemble the climatology, all flash flood reports in the National Climatic Data Center’s Storm Data database (NCDC 2003, 2004, 2005, 2006, 2007) were recorded for the region of interest for the months of May–September in the years 2003–07. Flash floods attributed to dam failures on days with fair weather were excluded from the sample. Also excluded from the climatology were 16 single- and multiple-day flash flood events caused by precipitation associated with tropical cyclones and their remnants, as the present study is concerned exclusively with flash-flood-producing extratropical cyclones. All flash flood reports in the northeastern states on days during which tropical cyclones or their remnants were traversing the area were excluded. This procedure may also exclude predecessor rain events (PREs; Galarneau et al. 2010) of heavy precipitation ahead of the tropical cyclones.

The desired end result was a climatology consisting of only one event per diurnal cycle: one flash flood per event day. An “event day” was defined to extend from 0800 local time (LT) on one day to 0800 LT on the following day, with some allowance for events that spanned two event days: storm reports bridging the 0800 LT cutoff were classified as being part of the event day spanning the larger time interval relative to 0800 LT.

For multiple-report event days, 24-h precipitation data from cooperative observing stations archived by the Northeast Regional Climate Center were used to select the county with reported flash flooding containing the highest observed 24-h precipitation total. Multisensor precipitation estimate (MPE) data were considered at this step, but they were unavailable for the entire period. Other precipitation estimates computed with radar-estimated rainfall, such as Hydro-NEXRAD (Krajewski et al. 2011) and the Areal Mean Basin Estimated Rainfall (AMBER) program (Davis 1993) were not used at this step because the intent was to pick a significant rainfall event for each event day, not necessarily the most extreme event.

If more than one flash flood was reported in the selected county, the flash flood location selected to represent the event day was then determined to be the reported location in that county receiving the highest radar-estimated precipitation total. When the area of greatest precipitation extended into a neighboring county that also reported a flash flood, or when the density of rain gauges was insufficient to capture the area of maximum precipitation in adjacent counties reporting flash flooding (i.e., when the maximum radar-estimated precipitation at least doubled the maximum observed rain gauge observation), the reported flash flood location with the maximum estimated rainfall from radar was used.

Locations of flash flooding for each event were determined through the event descriptions in Storm Data. The flood’s location was often described as a road or a stream, or more broadly as occurring in the vicinity of a town or city. When the specific location of the flooding—a named stream or a road—was not cited in the report, the town or city named in the description was viewed with Geographical Information Systems (GIS) software, including ArcGIS and Google Earth, to identify potential locations where flooding could be easily detected—points where a stream intersected or ran closely alongside a road. From among the potential flood locations, a stream–road intersection in the basin with the highest radar-estimated precipitation total was identified as the most likely flash flood location. For vague urban flooding reports, the point at the center of the basin with the highest estimated precipitation total within the named town or city was identified as the most likely flash flood location.

The use of flash flood reports implies the inclusion of both major, life-threatening flash flooding of streams and minor ponding of water on roads. Flash flood reports are also subject to the vagaries of those doing the reporting and, thus, are more likely to be generated in more densely populated areas, and more likely during the day than overnight. Despite the inconsistencies in flash flood reporting, flash flood reports are a means of locating hazardous hydrometeorological events—particularly those at relatively small scales—that go undetected by rain and stream gauges. Section 4 will address some of these concerns about the reliability of flash flood reports.

Precipitation totals were estimated from data provided by the local radars in the Weather Surveillance Radar-1988 Doppler (WSR-88D) Next Generation Weather Radar (NEXRAD) network. Previous MCS classification studies (such as Parker and Johnson 2000, Schumacher and Johnson 2005, and SJ2006) used the NOWrad product from WSI, which contains a mosaic of individual radars. Because these products have different fields of view and spatial and temporal resolutions, there may be differences in the interpretation of the radar imagery between studies.

In this study, rainfall estimates were computed using AMBER, which is used operationally by the National Weather Service (NWS) to monitor potential flash flood situations in real time (Davis 1993). AMBER computes the mean estimated rainfall for a set of user-defined basins by converting the digital hybrid reflectivity (DHR)—a level III NEXRAD product—to an estimated rainfall rate using the reflectivity–rainfall rate (ZR) conversion.

The basins used in this study came from two sources. For the events occurring under the auspices of the Pittsburgh NWS forecast office, a set of basins that has been developed for local use by the NWS Pittsburgh forecast office was used (R. Davis 2008, personal communication). For all other WSR-88D sites, the basins derived by the National Basin Delineation Project were used (Arthur et al. 2005).

Rainfall amounts were first computed using the standard ZR relationship. These totals were then compared with the cooperative observer 24-h precipitation totals in the vicinity of the flash flood location and with any precipitation observations or estimates cited in the flash flood report. If the radar estimates substantially underestimated the observed rainfall totals and nearby atmospheric soundings detected a warm, nearly saturated lower troposphere, the rainfall estimates were recomputed in AMBER using a tropical ZR conversion. If these updated rainfall totals were found to be more accurate compared to the observations, these new estimates were used to represent the rainfall estimate for this case. Otherwise, the standard ZR relationship was used for the given case.

Storm total rainfall amounts were computed for the basin in which the flash flooding was reported, for any basins hydrologically upstream from this flooded basin, and for the basin in the vicinity of the flood—not necessarily in the same county as the flash flood report, but affected by the same precipitation system(s)—that received the most rainfall. The latter amounts are reported in section 4. Also reported in section 4 are the mean 5-min rainfall rates for each storm type over the duration of the event.

The environmental conditions associated with each case were examined via the North American Regional Reanalysis (NARR; Mesinger et al. 2006). The NARR is available at a grid spacing of approximately 32.5 km × 32.5 km every 3 h. Composite maps for each storm type were generated by interpolating the NARR data onto a 0.5° × 0.5° grid with the flood location transposed to a randomly selected point located at 42°N, 74°W. As such, all maps are flood relative, and geographical considerations (such as topography) do not necessarily apply to the composite maps in section 5.

3. Event classification

A survey of the flash flood reports in Storm Data yielded a total of 1251 flash flood reports on 201 event days for the Northeast from May through September 2003–07, excluding events directly associated with named tropical storms and fair-weather dam breaks. Of these 201 event days, 11 were missing radar data, leaving 190 events whose radar signatures could be classified. A further three events were eliminated because they were the continuation of precipitation from the previous event day. Thus, the total number of events analyzed in this study is 187.

To achieve the overarching objective of determining whether flash-flood-producing rainstorms in the Northeast are qualitatively different from those in other areas of the country, it is crucial to determine how well the archetypes of heavy precipitation in the literature describe the observed organization of heavy precipitation in the Northeast. The classification of events in this study began by following the classification scheme described by Parker and Johnson (2000) and SJ2006. These included the leading stratiform (LS), parallel stratiform (PS), trailing stratiform (TS), and training line/adjoining stratiform (TLAS) types, which will be referred to more generally as “linear” events, as well as the back-building/quasi-stationary (BB) class of events.

SJ2006 also classified events as “synoptic” (SYN) if the area of precipitation, usually large and possessing a radar signature that did not meet the classification criteria of the linear systems, persisted for longer than 24 h, and as “multiple” (MULT) if more than one MCS contributed to the daily precipitation total. Events not meeting the description of any of the above categories were classified as “other.” While the above classification scheme was robust for much of the country in the SJ2006 study, it appeared to be insufficient to describe heavy precipitation events in the Northeast, as four of the seven Northeast MCS cases examined by SJ2006 were classified as other.

After classifying the 187 events in the current study using the above classification system, it was modified and expanded to account for the morphologies of precipitation observed in the Northeast. Events were classified into four general event types, akin to those in SJ2006: Back-building, Linear, Multiple, and Other/Size. These categories were subdivided as described below. Throughout the paper, general event types are identified by an initial capital letter, while the specific event types are identified by an all-capital abbreviation.

The Linear group of events was based on those in SJ2006, classified based on whether the convective line traveled ahead of (LS), parallel with (PS), behind (TS), or alongside (TLAS) an area of more moderate, stratiform precipitation. Some MCSs contained linear structures, or had a linear shape as a whole, but did not fit the criteria listed in SJ2006. These events were classified in the Other/Size group, or if more than one MCS affected the flood area, in the Multiple group.

The Back-building group consisted of events whose primary source of precipitation was a quasi-stationary, back-building system. When a back-building system provided all of the precipitation contributing to the flash flood, the event was classified as a BB event. There was originally no intent to divide the back-building events into subgroups, as they had been adequately described by SJ2006 and had ostensibly been well known in the forecasting community for quite some time. After reexamining the back-building cases, it was determined that nearly half shared a characteristic sequence of a back-building MCS, which persisted until an approaching linear MCS, often of the TS type, merged with it. After the merger, the back-building convection dissipated or moved away from the flash flood location. These events were referred to as “back-building merging” (BBMERGE) events (Fig. 1).

Fig. 1.

Radar images from a sample BBMERGE event in central CT on 27 Jul 2005. The back-building convection (labeled BB) remains stationary while a linear (TS) MCS (labeled MCS) approaches from the northwest. By 2228 UTC [the time in (d)], the systems have merged. Images are at (a) 2058, (b) 2128, (c) 2158, and (d) 2228 UTC.

Fig. 1.

Radar images from a sample BBMERGE event in central CT on 27 Jul 2005. The back-building convection (labeled BB) remains stationary while a linear (TS) MCS (labeled MCS) approaches from the northwest. By 2228 UTC [the time in (d)], the systems have merged. Images are at (a) 2058, (b) 2128, (c) 2158, and (d) 2228 UTC.

For several of the events that had originally been classified as Multiple or Back-building events, the flood location was affected by both back-building cells and one or more MCSs. Unlike in the BBMERGE cases, the features occurred independently of one another, with a time interval of at least 15 min (and usually more than 30 min) between the back-building MCS and the other MCS(s). In the first type of multiple back-building cases, back-building convection forms and then dissipates, followed by the passage of one or more linear MCSs. In the second type, one or more linear MCSs traverse the flood area, and later back-building convection sets up over the flood area. These two types of events were combined in the “multiple back-building” (BBMULT) group. These events remained in the Back-building group, rather than in the Linear group, because the precipitation resulting from the back-building features contributed the majority of the rainfall in nearly every case. Figure 2 displays schematic diagrams for the three subclasses of events in the Back-building group.

Fig. 2.

Schematic diagram of the four back-building subclasses of events. Subclasses are determined by the relative motion of back-building cells and MCSs and their relative transits across the flood area. Gray areas indicate the areas where flash flooding is most likely. Figure adapted from Parker and Johnson (2000) and Schumacher and Johnson (2006).

Fig. 2.

Schematic diagram of the four back-building subclasses of events. Subclasses are determined by the relative motion of back-building cells and MCSs and their relative transits across the flood area. Gray areas indicate the areas where flash flooding is most likely. Figure adapted from Parker and Johnson (2000) and Schumacher and Johnson (2006).

The Multiple events were divided into subclasses, based not on the appearance of the features themselves, but on the relative movement of the individual features on radar. It was believed that this approach would best segregate the events based on the three-dimensional wind field, providing a physical basis for the subclasses. The Multiple events consisted of what could best be described as scattered convection: thunderstorms, multicellular storms, squall lines, and MCSs.

If the individual storm elements trained in the same direction at approximately the same speed, traversing the flood location via roughly the same path, these events were classified as “multiple training” (MULTTR) events (see Fig. 3 for sample radar imagery). These events have a different appearance from the Back-building class of events in that the individual cells and MCSs do not necessarily share the same genesis location and in that the individual storm elements are separated by a greater distance (usually >50 km) than the cells in a BB event. Like the BB events, the MULTTR events feature training convection, but unlike the BB events, they do not appear to be backward propagating and self-sustaining.

Fig. 3.

Radar images from a sample MULTTR event in eastern PA on 26 Aug 2007. The precipitation features, labeled 1, 2, and 3, are traveling from southwest to northeast. Images are from (a) 2230, (b) 2329, (c) 0028, and (d) 0131 UTC.

Fig. 3.

Radar images from a sample MULTTR event in eastern PA on 26 Aug 2007. The precipitation features, labeled 1, 2, and 3, are traveling from southwest to northeast. Images are from (a) 2230, (b) 2329, (c) 0028, and (d) 0131 UTC.

On some occasions, storm elements were moving in seemingly random directions and at different speeds; the flash flooding was the fortuitous result of several cells randomly traversing the same location. These events were classified as “multiple random” (MULTRAND) events. The slow movement typically exhibited by the cells suggests that light upper-level winds were present, and the nonuniform movement of the individual cells suggests the presence of directional wind shear on the horizontal plane, the influence of local low-level boundaries, or the anchoring effect of local topography.

The third class of Multiple events consists of events in which two or more broad MCSs with moderate to heavy precipitation rates, traveling toward the same location from different directions, ultimately merge to form one larger MCS (typically merging just before reaching the eventual flood location). These events differed from MULTRAND events both in the behavior of the individual features, in that the individual features in the MULTRAND events did not merge, and the size of the individual features, which was often (but not always) larger in these events. These events were classified as “multiple merging” (MULTMERGE) events. The most common scenario for MULTMERGE events was for one MCS traveling approximately south to north to merge with another MCS traveling approximately west to east, and for the resulting combined MCS to continue toward the northeast (see Fig. 4 for sample radar images). Figure 5 displays schematic diagrams of these three multiple-event types.

Fig. 4.

Radar images from a sample MULTMERGE event in northeastern PA on 27 Jun 2007. The area of precipitation initially located to the west is traveling west to east and merging with the area of precipitation that is initially in the center of the field of view and traveling south to north. Images are from (a) 2000, (b) 2100, (c) 2158, and (d) 2258 UTC.

Fig. 4.

Radar images from a sample MULTMERGE event in northeastern PA on 27 Jun 2007. The area of precipitation initially located to the west is traveling west to east and merging with the area of precipitation that is initially in the center of the field of view and traveling south to north. Images are from (a) 2000, (b) 2100, (c) 2158, and (d) 2258 UTC.

Fig. 5.

Schematic diagram of the four Multiple subclasses of events. Subclasses are determined by the relative motions of cells and MCSs, as illustrated. Gray boxes indicate the area of highest flooding potential for each type. Figure adapted from Parker and Johnson (2000).

Fig. 5.

Schematic diagram of the four Multiple subclasses of events. Subclasses are determined by the relative motions of cells and MCSs, as illustrated. Gray boxes indicate the area of highest flooding potential for each type. Figure adapted from Parker and Johnson (2000).

The other events originally functioned as one catch-all category for otherwise unclassifiable events. Because this group contained a large percentage of events, it was subdivided based on the spatial and temporal scales of the contributing storms and MCSs. To reflect this, the group’s name has been modified from SJ2006 to Other/Size. Those events having a length scale of 50 km or less were classified as “small mesoscale” (SM) events. In these cases, only one thunderstorm or multicellular storm, often traveling relatively slowly, was responsible for the flash flood.

MCSs with a length scale of 50–150 km that did not fit the descriptions of the linear events were classified as “medium mesoscale” (MM) events. These events were similar in size to most of the linear MCSs, and may have contained convective lines for part of their existence, but they lacked the particular linear structures defined for the linear events.

The SYN events, categorized by SJ2006 as broad areas of rainfall persisting for more than 24 h, were included in the Other/Size group. Features with a length scale of greater than 150 km that dissipated within 24 h of their initiation were classified as “large mesoscale” (LM) events. These events looked very similar on radar to the SYN events, and similarly produced flash flooding as a result of an extended duration of moderate to heavy rainfall, but the shorter lifetime of these precipitation systems placed them in a different subcategory. Like the MM events, LM and SYN events may have contained embedded linear structures or even have taken on a linear appearance as a whole, but they did not match the descriptions of the Linear event types.

These latter three categories—MM, LM, and SYN events—correspond to the type of morphology that Gallus et al. (2008) and Lombardo and Colle (2010) refer to as “nonlinear.” The Other/Size group of events essentially functioned as a group of “unclassifiable” categories, not at all dependent on the structure of the precipitation systems, but only on their size.

In summary, there are four categories of precipitation organization outlined in this study: Back-building, Linear, Multiple, and Other/Size, each of which has three or four subcategories describing their structure, size, and/or movement. The names, abbreviations, and descriptions of these event types are summarized in Table 1. The following section will describe how frequently each of these event types occurred and the precipitation characteristics associated with each type.

Table 1.

Descriptions and abbreviations (abbrev) of storm types in the classification system used in this study.

Descriptions and abbreviations (abbrev) of storm types in the classification system used in this study.
Descriptions and abbreviations (abbrev) of storm types in the classification system used in this study.

4. Event climatology

Using the classification system outlined in section 3, Fig. 6 displays the relative frequency of the occurrence of each event type. The most common event types were BB and BBMERGE, which accounted for approximately 17% and 14% of all events, respectively. Among the Other/Size events, the SYN and SM types were most common. The subcategory of Multiple events that occurred most frequently was the MULTTR group. Much like Parker and Johnson (2000) had found for the southern plains region, the TS events were the most common among the Linear events.

Fig. 6.

Frequency of northeast U.S. flash flood events, 2003–07, by event type, following the scheme of the present paper. All 187 events in this study are represented. Both the total frequency and the relative frequency (%) are given.

Fig. 6.

Frequency of northeast U.S. flash flood events, 2003–07, by event type, following the scheme of the present paper. All 187 events in this study are represented. Both the total frequency and the relative frequency (%) are given.

To make a meaningful comparison between the frequency of event types examined in this study and the results of SJ2006, only the events that met similar selection criteria are used. Figure 7 displays the results for the 23 events that produced a radar-estimated 24-h precipitation total meeting the precipitation threshold used by SJ2006. In comparison to SJ2006’s results for the Northeast (briefly described in section 3), a much higher percentage of events included back-building convection, a much lower percentage of events was of the SYN type, and, similarly, a small proportion of events was found to be Linear. More than half of the events in this study that met the precipitation threshold come from the Back-building group. The Other/Size events were represented by the largest events (the SYN and LM types) and the smallest (SM). Each of the three Multiple types of events produced one or two flash flood events meeting the precipitation threshold. The Linear group had the smallest representation among these extreme heavy precipitation events, with only the PS and TLAS types producing one event each meeting the 50-yr precipitation frequency threshold.

Fig. 7.

Frequency of northeast U.S. flash flood events, 2003–07, by event type, with radar-estimated rainfall meeting the 24-h precipitation threshold of Schumacher and Johnson (2006). Both the total frequency and the relative frequency (%) are given for these 23 events.

Fig. 7.

Frequency of northeast U.S. flash flood events, 2003–07, by event type, with radar-estimated rainfall meeting the 24-h precipitation threshold of Schumacher and Johnson (2006). Both the total frequency and the relative frequency (%) are given for these 23 events.

The different methods of event selection in the two studies, 24-h rain gauge measurements in SJ2006 and Storm Data flash flood reports in the present study, appear to be the primary cause of the divergent results. Specifically, the detection of smaller-scale and shorter-lived systems by the flash flood reports led to a much higher percentage of BB events and a much lower percentage of SYN events meeting the SJ2006 precipitation threshold in this study.

The remainder of this section will examine the precipitation data in more detail to seek trends in the characteristics of the events that produce larger or smaller rainfall totals. The median radar-estimated precipitation total for the basin with the highest precipitation estimate in the vicinity of the flash flood location was 90.44 mm (3.56 in.). The majority of events (89.3%) included at least one basin with a precipitation estimate exceeding 50 mm (1.97 in.), and a substantial minority (34.2%) included at least one basin with a precipitation estimate exceeding 100 mm (3.94 in.). Only one case had a maximum precipitation estimate of less than 25 mm (0.98 in.).

Forecasters could benefit from knowing the temporal characteristics of heavy precipitation and flash flood events, at both seasonal and diurnal time scales. Figure 8 shows the maximum basin-averaged precipitation estimate (size of circle) as a function of the month and time of day (abscissa and ordinate, respectively). While it is clear that flash floods occurred most often in the late afternoon to early evening during midsummer, corresponding with diurnal and seasonal climatologies such as those of Winkler et al. (1988) and Maddox et al. (1979), it is difficult to detect a clear preferred time of year or time of day for the heaviest flash-flood-producing rainfall.

Fig. 8.

Maximum basin-averaged precipitation estimates for each case (circles) as a function of month (abscissa) and time of day (ordinate, LT). The relative size of the circles indicates the magnitude of the precipitation. The boldface circle in the top-right corner—not a data point—represents a value of 4.0 in. (101.6 mm) for reference. The time of day is the time cited in the flash flood report.

Fig. 8.

Maximum basin-averaged precipitation estimates for each case (circles) as a function of month (abscissa) and time of day (ordinate, LT). The relative size of the circles indicates the magnitude of the precipitation. The boldface circle in the top-right corner—not a data point—represents a value of 4.0 in. (101.6 mm) for reference. The time of day is the time cited in the flash flood report.

Several trends, however, are present. The afternoon and early evening hours, from approximately 1200 to 2100 LT, contain the widest range of precipitation amounts, particularly from mid-May through August. These are the hours during which diurnally driven convection is most intense (accounting for the heaviest precipitation estimates), and they are also the hours when minor flooding is most likely to be reported (which may account for the lightest precipitation estimates).

Precipitation amounts tend to be moderate around midnight local time and show a slight increasing trend overnight until 0600 LT. This is most likely an artifact of the flash-flood-reporting process, which is more likely to detect more severe flooding in the overnight hours. While one would expect the majority of overnight floods to be reported in urban or populated areas, an analysis of impervious area data from the 2001 National Land Cover Dataset revealed that of the 19 events that were reported between midnight and 0800 LT, 13 events (68%) occurred in rural areas—basins with relatively low impervious area. Six of these 13 rural events were reported in towns or small cities rather than in true rural areas, but the impervious area of the contributing watershed was small—much of this was rural area—leading to the rural classification.

Once daylight breaks, the precipitation estimates decrease notably, to generally less than 50 mm (1.97 in.) from approximately 0730 to 1130 LT. Most morning flash floods in the Northeast (0600–1200 LT) are a result of either a large, long-lived system (LM or SYN) or an MCS that moves into western Pennsylvania or New York from Ohio.

Knowing whether each event type described in section 3 tends to be associated with relatively larger or smaller precipitation totals can help forecasters to anticipate how severe the potential flood threat might be from a given storm system early in its development. Figure 9 displays boxplots of maximum basin-averaged precipitation estimates for the basin in the vicinity of the flood report (though not necessarily within the flooded watershed) for each specific event type with at least five events.

Fig. 9.

Boxplots of maximum basin-averaged precipitation estimates (mm) near the flood location for each category of event types with at least five cases. The values plotted, from bottom to top, represent the minimum, 25th percentile, median, 75th percentile, and maximum. The numbers in parenthesis accompanying the labels represent the observed frequency of each type of event.

Fig. 9.

Boxplots of maximum basin-averaged precipitation estimates (mm) near the flood location for each category of event types with at least five cases. The values plotted, from bottom to top, represent the minimum, 25th percentile, median, 75th percentile, and maximum. The numbers in parenthesis accompanying the labels represent the observed frequency of each type of event.

The precipitation amounts produced by the different event types have visibly different distributions, as evidenced by the boxplots in Fig. 9, but the relatively small sample sizes associated with these groups limit the statistical significance of these differences. A two-tailed hypothesis test (Wilks 1995) with the null hypothesis that the mean precipitation produced by the BB group (the largest mean) differs from the mean of the precipitation produced by the MM group (the smallest mean) fails, as the null hypothesis cannot be rejected at the 10% level. (The test statistic, Z, was equal to 1.07.) That is, the samples of these two event types were statistically indistinguishable. No other pairing of two event types produces a more favorable test statistic.

The first five boxplots on the left-hand side of Fig. 9, corresponding to the Back-building events and to the larger Other/Size events (SYN and LM), are associated with relatively large median values, and these types account for all but one of the events exceeding 175 mm. These storm types are typically associated with long durations of moderate to heavy precipitation rates as a result of the sustained regeneration of convection, in the case of the Back-building events, or as a result of the passage of large areas of persistent rainfall, in the case of the SYN and LM events.

The five boxplots in the center of Fig. 9, corresponding to the MM and SM events, and the Linear events, tend to have a narrower distribution of smaller precipitation totals. The one exception to this trend is the TLAS events, which have a median precipitation total exceeding 100 mm and which produced the lone case aside from those above that exceeded 175 mm. Like the event types in the above paragraph, the TLAS type features a mechanism for sustained heavy rainfall rates: an elongated line of training convection.

The MM events tend to produce less precipitation. The 75th percentile for the MM events is lower than the 25th percentile for all but two other event types. That the MM events produce notably lower precipitation totals reflects the appearance of this class of events on radar. The most disorganized of the single-feature events, they were usually nondescript moderate-sized areas of moderate precipitation rates moving at a moderate speed: a formula for moderate precipitation totals. What is perhaps more remarkable is that these features, with their innocuous radar signatures, moderate precipitation rates, and relatively low precipitation totals, were able to generate several flash flood reports. The TS cases, though they had a larger median precipitation total than the MM cases, displayed the narrowest range of precipitation totals and the smallest maximum precipitation total. As stated above, these precipitation totals did not have a statistically significant difference from those of the event types producing the most precipitation.

The final three boxplots on the right of Fig. 9, representing the Multiple cases, feature nearly identical medians and narrow ranges of precipitation totals. Only three multiple cases of any type exceeded 125 mm (two MULTRAND and one MULTMERGE), while only three cases produced less than 50 mm of precipitation. It is likely that the lowest maximum precipitation totals would exceed those of other types of single-feature events such as the MM and TS events because, while those events included only one MCS that may produce only a limited duration of heavy rainfall over a given location, a multiple event by definition includes at least two or three such periods of heavy rainfall as a consequence of the passage of multiple features. At the other end of the scale, extremely high precipitation totals would conceivably be quite rare as a product of multiple events owing to their lack of a mechanism for producing persistent, heavy rainfall.

Many flash flood days produced at least several reports of flash flooding as the precipitation systems traversed a relatively large area; others, more localized or generating only mild to moderate flooding, produced few reports. Of the precipitation systems producing the greatest 10% of precipitation, 37% produced only one or two flash flood reports. About half of these one- or two-report events were SM, BB, BBMERGE, or BBMULT events, showing that localized storms can produce large precipitation totals and generate isolated reports of flash flooding in the Northeast.

While precipitation totals reveal a great deal about a given event’s potential impact, examining the evolution of the precipitation rate may better describe a storm type’s characteristics (Fig. 10). The events in the Back-building group tend to peak in intensity in the first third of their time over the selected basins, and decline in intensity over the last two-thirds of their passage. The final peak in the BBMULT rainfall rate is likely a result of the events in which the back-building forms late in the event, after the passage of an MCS.

Fig. 10.

Mean precipitation rate (mm h−1) as a function of storm-relative time for each event type. The scale on the abscissa represents the passage of time from the beginning (0) to the end (1.0) of the event. Event types with only one event are excluded.

Fig. 10.

Mean precipitation rate (mm h−1) as a function of storm-relative time for each event type. The scale on the abscissa represents the passage of time from the beginning (0) to the end (1.0) of the event. Event types with only one event are excluded.

In the Linear group, the PS events produce the greatest mean rainfall rate of any event type, a likely consequence of the relatively small sample size and consistent structure of PS events. Both TS and TLAS events display a similar profile to those of the Back-building events, with a peak in the first third of the event and a gradual decay. The differences among these groups in total precipitation are largely a consequence of the differences in their duration.

The members of the Other/Size group display rainfall rate profiles that reflect their spatial scales and structure. The SYN and LM events have a long duration of moderate (SYN) to heavy (LM) precipitation. The MM events, like the SYN events, have moderate rainfall rates, but a much shorter duration. The many small peaks of the SYN and MM events illustrate the random nature of the embedded strong convection within the broader area of moderate precipitation for these events. The SM events display a nearly parabolic pattern with a peak in the middle of the event. Although the SM events have a high mean rainfall rate, their short duration typically produces moderate precipitation totals.

The Multiple event types display similar profiles with mean rainfall rates of approximately 0.5 in. h−1. The MULTTR profile displays three peaks, likely corresponding to successive MCSs. The first peak is the strongest because this is likely the time at which the coherence in the storm-relative time of passage of an MCS is greatest. Similarly, the MULTRAND and MULTMERGE events display incoherence in the timing of the passage of individual cells and MCSs from case to case.

The locations of the events and the relationship, if any, between where flash-flood-producing storms occur and how much rainfall they produce could add another layer of information to inform the forecaster’s decision-making process. Figure 11 displays maps parsed by rainfall total. The Back-building events are much more evenly distributed across the Northeast than are the other three general event types. In contrast to the relative spatial uniformity of the Back-building events, the Linear events are largely clustered in western Pennsylvania. Many of these Linear events in western Pennsylvania are of the TS and PS types, which are not common elsewhere in the Northeast. SJ2006 found the TS type to be the second most common type of precipitation organization for extreme rainfall events in the region to the west of Pennsylvania; this study’s predominance of TS events in western Pennsylvania suggests that western Pennsylvania could be part of a larger region that is more favorable for this type of event than is the remainder of the Northeast. Many of the TS and PS events in this study originated in Ohio and traveled eastward into western Pennsylvania.

Fig. 11.

Map displaying the locations of the 187 flash flood events. The four panels represent the estimated precipitation totals: (a) 0–50.0, (b) 50.01–100.0, (c) 100.01–150.0, and (d) at ≥ 150.01 mm. The shape of the symbol represents the general event type, and the shading represents the specific event type.

Fig. 11.

Map displaying the locations of the 187 flash flood events. The four panels represent the estimated precipitation totals: (a) 0–50.0, (b) 50.01–100.0, (c) 100.01–150.0, and (d) at ≥ 150.01 mm. The shape of the symbol represents the general event type, and the shading represents the specific event type.

Like the Linear events, the type of Other/Size events that was most common in western Pennsylvania, the MM type, was notably rarer in other portions of the Northeast. Six of the eight MM events, which were described earlier in this section as producing relatively low precipitation estimates and as being unorganized, occurred in western Pennsylvania and produced less than 50 mm of rainfall. It is difficult to discern whether this regional dependence is a meteorological phenomenon, a product of a local environment that enhances runoff due to topography and land use, or a consequence of regional differences in the detection and reporting of flash floods.

Most of the SYN and LM events occurred in western Pennsylvania and in the megalopolis stretching from the Philadelphia area to Boston. This seems to be both a meteorological phenomenon, as large precipitation systems appear to travel either east or west of the Appalachian Mountains (as described below), and a hydrological phenomenon, as urban areas tend to generate greater runoff for a given amount of precipitation than rural areas. As a result of their predominant storm tracks, their more efficient runoff production, and their greater population density, the areas to the east and west of the Appalachian Mountains are more likely to produce a flash flood report for a given precipitation system.

Among the Multiple events, the MULTTR events occurred largely in western Pennsylvania and from eastern Pennsylvania through the New York metropolitan area, but they were scattered throughout the rest of the Northeast as well. The overall tendency for flash flood events to be reported in western and eastern Pennsylvania much more frequently than they are reported in central Pennsylvania appears to be partially a product of the higher terrain in central Pennsylvania. Murray and Colle (2011) found that convective storms tend to traverse western and eastern Pennsylvania more often than they traverse central Pennsylvania, and this geographical influence was attributed to the topography. Precipitation is thought to be favored both in its response to upslope effects in western Pennsylvania on the windward side of the Appalachian Mountains and in the lee of the mountains in eastern Pennsylvania.

At the same time, the larger population centers of eastern and western Pennsylvania assuredly contribute to the more frequent flash flood reports owing to a greater likelihood for flash flooding to be detected and reported. Additionally, the runoff response to a given amount of precipitation will be greater in a more developed area as a result of the larger percentage of impervious area. This explains not only the distribution of flash flood reports in Pennsylvania, but also the tendency for flash flood reports to cluster in and near large urban areas.

5. Environmental conditions

While a thorough analysis of the environmental conditions associated with each event type is beyond the scope of this paper, some results are briefly presented here. Figure 12 displays a map showing the mean flood-relative equivalent potential temperature (θe), geopotential height, and wind at 925 hPa for the time period prior to the report time of the flash flood for BBMERGE events. In general, there was little distinguishing the environment of these events from that of the BB events (not shown). Both event types were associated with relatively light winds (≤10 kt) and warm, moist environments featuring a θe ridge near the eventual flood location. The chief distinguishing feature was a weak low-level trough upwind of the back-building location for the BBMERGE events, which was most pronounced at 925 mb. This trough is associated with the MCS that is approaching the back-building convection. The southwesterly flow ahead of this trough, in the vicinity of the flash flood location, produces low-level θe advection, which helps to destabilize the lower troposphere and has been found to distinguish heavy rain cases (Junker et al. 1999). While the trough is present in the composite, it was readily apparent in only about half of the individual BBMERGE cases. Pending further examination, this trough appears to be the feature that offers the most promise for the forecast of a BBMERGE event before it materializes on radar.

Fig. 12.

Map displaying composite mean flood-relative equivalent potential temperature (shaded, K), geopotential height (contoured, m), and winds (barbs, kt) at 925 hPa at the time period prior to the time of the flood report for BBMERGE events. The black dot at 42°N, 74°W represents the flood location.

Fig. 12.

Map displaying composite mean flood-relative equivalent potential temperature (shaded, K), geopotential height (contoured, m), and winds (barbs, kt) at 925 hPa at the time period prior to the time of the flood report for BBMERGE events. The black dot at 42°N, 74°W represents the flood location.

It was mentioned in section 3 that the Multiple events were categorized based on the motion of the individual features on the radar, which was thought to represent the three-dimensional wind field. Figure 13 shows that this was the primary difference among the MULTTR, MULTRAND, and MULTMERGE types. While the temperature and dewpoint profiles were similar for these three event types, the MULTTR wind profile was nearly unidirectional with height. The MULTRAND wind profile features strong veering, suggesting that the differential motion of individual features on the radar may be related to the height of their base. Features with a more southerly motion may be rooted in the boundary layer, while those with a more westerly motion may be rooted above the boundary layer. Local outflow boundaries, while not evident in the NARR analysis, may also contribute to the differential motion in some cases. The MULTMERGE wind profile displays more moderate veering and the strongest winds of any multiple event type. Composite maps of MULTMERGE events (not shown) displayed a pronounced trough to the west (roughly upwind) of the flood location, suggesting that the merging features often consisted of one MCS ahead (or to the east) of the trough and another collocated with the trough. A more thorough analysis of the environmental conditions associated with each event type will be presented in a future paper.

Fig. 13.

Skew T plots displaying mean temperature (°C), dewpoint (°C), and winds (kt) for the grid point nearest the flood location at the time period prior to the time of the flood report for (a) MULTTR, (b) MULTRAND, and (c) MULTMERGE events. Hodographs and significant derived variables and indices are also presented.

Fig. 13.

Skew T plots displaying mean temperature (°C), dewpoint (°C), and winds (kt) for the grid point nearest the flood location at the time period prior to the time of the flood report for (a) MULTTR, (b) MULTRAND, and (c) MULTMERGE events. Hodographs and significant derived variables and indices are also presented.

6. Conclusions

This study examined 187 warm-season flash flood events in the Northeast for the years 2003–07 from the Storm Data database. Storms were classified based on their appearance on radar into 1 of 14 categories that fell into four groups: Back-building, Linear, Other/Size, and Multiple. The Back-building events were found to be most common, while the Linear events occurred least often. The present study sought to answer several questions about flash-flood-producing precipitation in the Northeast.

Are the predominant patterns of the organization of flash-flood-producing precipitation different in the Northeast than in the Midwest, southern plains, and other parts of the United States? By comparing the results in the present study with the findings of SJ2006, it appears that there are significant differences, particularly in that the linear modes of convection are appreciably less common and that the more scattered, cellular modes of convection are more common. Similarly, Lombardo and Colle (2010) examined warm-season convection in the Northeast using the same classification system that Gallus et al. (2008) had used for the central United States and found a similar result. This does not necessarily mean that storms with a linear organization are less of a flood threat in the Northeast, but that they are less common in comparison to other morphologies in the Northeast.

Are there morphologies of precipitation that have not been recognized as flash flood producers in the literature? Yes, there are several newly recognized morphologies. While back-building events have been recognized to be producers of prolonged heavy precipitation (e.g., Schumacher and Johnson 2005, SJ2006), subcategories such as BBMERGE were not recognized as distinct phenomena. A distinction was also made between large-scale events that transcended the diurnal cycle (SYN) and those whose lifetimes lasted less than 24 h (LM). While others (e.g., Gallus et al. 2008; Lombardo and Colle 2010) have recognized that scattered, cellular convection and thunderstorms compose a significant proportion of warm-season convection, particularly in the Northeast, these morphologies have not traditionally been associated with flash flooding. In this study, it is suggested that these scattered, cellular modes of precipitation be subdivided based upon the relative motion of the cells.

Which types of events tend to produce the most precipitation and cause the most significant impacts in the Northeast? Those with an internal mechanism for producing sustained rates of heavy precipitation: the BB and BBMERGE events, with their internal process for generating new convective cells; the TLAS events, with their training lines of heavy convection; and the SYN and LM events, with their large areas of moderate to heavy rainfall. While these storm types were also found to more frequently cause the most significant impacts in a survey of Storm Data flash flood reports, all storm types were found to be capable of producing damage or endangering lives.

This paper is but a first step in examining the impacts of flash-flood-producing precipitation systems in the Northeast. It is limited in that it does not examine the impacts of tropical cyclones and their remnants, and that it does not account for flash-flood producing precipitation systems in the cool season. Future papers will describe the atmospheric conditions associated with each type of event and will investigate the processes involved in the newly described morphologies. It is hoped that a more robust understanding of the underlying processes associated with these flash-flood-producing systems will lead to improved real-time forecaster awareness of potentially threatening situations and improved prevention of flash-flood-related rescues, injuries, and fatalities.

Acknowledgments

Financial support for this project was provided by the National Science Foundation, Grant EAR 0911076. We gratefully acknowledge the suggestions of Russ Schumacher and two anonymous reviewers, which helped to improve the paper.

REFERENCES

REFERENCES
Arthur
,
A. T.
,
G. M.
Cox
,
N. R.
Kuhnert
,
D. L.
Slayter
, and
K. W.
Howard
,
2005
:
The National Basin Delineation Project
.
Bull. Amer. Meteor. Soc.
,
86
,
1443
1452
.
Bluestein
,
H. B.
, and
M. H.
Jain
,
1985
:
Formation of mesoscale lines of precipitation: Severe squall lines in Oklahoma during the spring
.
J. Atmos. Sci.
,
42
,
1711
1732
.
Bluestein
,
H. B.
,
G. T.
Marx
, and
M. H.
Jain
,
1987
:
Formation of mesoscale lines of precipitation: Nonsevere squall lines in Oklahoma during the spring
.
Mon. Wea. Rev.
,
115
,
2719
2727
.
Cuo
,
L.
,
T. C.
Pagano
, and
Q. J.
Wang
,
2011
:
A review of quantitative precipitation forecasts and their use in short- to medium-range streamflow forecasting
.
J. Hydrometeor.
,
12
,
713
728
.
Davis
,
R. S.
,
1993
:
AMBER, a prototype flash flood warning system
.
Preprints, 13th Conf. on Weather Analysis and Forecasting, Vienna, VA, Amer. Meteor. Soc., 379–383
.
Davis
,
R. S.
,
2001
:
Flash flood forecast and detection methods
.
Severe Convective Storms, Meteor. Monogr., No. 50, Amer. Meteor. Soc., 481–526
.
Doswell
,
C. A.
, III
,
H. E.
Brooks
, and
R. A.
Maddox
,
1996
:
Flash flood forecasting: An ingredients-based methodology
.
Wea. Forecasting
,
11
,
560
581
.
Fritsch
,
J. M.
,
R. J.
Kane
, and
C. R.
Chelius
,
1986
:
The contribution of mesoscale convective weather systems to the warm-season precipitation in the United States
.
J. Climate Appl. Meteor.
,
25
,
1333
1345
.
Fritsch
,
J. M.
, and
Coauthors
,
1998
:
Quantitative precipitation forecasting: Report of the Eighth Prospectus Development Team, U.S. Weather Research Program
.
Bull. Amer. Meteor. Soc.
,
79
,
285
299
.
Galarneau
,
T. J.
,
L. F.
Bosart
, and
R. S.
Schumacher
,
2010
:
Predecessor rain events ahead of tropical cyclones
.
Mon. Wea. Rev.
,
138
,
3272
3297
.
Gallus
,
W. A.
, Jr.
,
N. A.
Snook
, and
E. V.
Johnson
,
2008
:
Spring and summer severe weather reports over the Midwest as a function of convective mode: A preliminary study
.
Wea. Forecasting
,
23
,
101
113
.
Houze
,
R. A.
, Jr.
,
B. F.
Smull
, and
P.
Dodge
,
1990
:
Mesoscale organization of springtime rainstorms in Oklahoma
.
Mon. Wea. Rev.
,
118
,
613
654
.
Junker
,
N. W.
,
R. S.
Schneider
, and
S. L.
Fauver
,
1999
:
A study of heavy rainfall events during the Great Midwest Flood of 1993
.
Wea. Forecasting
,
14
,
701
712
.
Krajewski
,
W. F.
, and
Coauthors
,
2011
:
Towards better utilization of NEXRAD data in hydrology: An overview of Hydro-NEXRAD
.
J. Hydroinf.
,
13
,
255
266
.
Lombardo
,
K. A.
, and
B. A.
Colle
,
2010
:
The spatial and temporal distribution of organized convective structures over the Northeast and their ambient conditions
.
Mon. Wea. Rev.
,
138
,
4456
4474
.
Maddox
,
R. A.
,
C. F.
Chappell
, and
L. R.
Hoxit
,
1979
:
Synoptic and meso-α scale aspects of flash flood events
.
Bull. Amer. Meteor. Soc.
,
60
,
115
123
.
Mesinger
,
F.
, and
Coauthors
,
2006
:
North American Regional Reanalysis
.
Bull. Amer. Meteor. Soc.
,
87
,
343
360
.
Murray
,
J. C.
, and
B. A.
Colle
,
2011
:
The spatial and temporal variability of convective storms over the Northeast United States during the warm season
.
Mon. Wea. Rev.
,
139
,
992
1012
.
NCDC
,
2003
:
Storm Data. Vol. 45, Nos. 5–9. [Available online at http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?wwEvent~Storms.]
NCDC
,
2004
:
Storm Data. Vol. 46, Nos. 5–9. [Available online at http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?wwEvent~Storms.]
NCDC
,
2005
:
Storm Data. Vol. 47, Nos. 5–9. [Available online at http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?wwEvent~Storms.
NCDC
,
2006
:
Storm Data. Vol. 48, Nos. 5–9. [Available online at http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?wwEvent~Storms.]
NCDC
,
2007
:
Storm Data. Vol. 49, Nos. 5–9. [Available online at http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?wwEvent~Storms.]
Parker
,
M. D.
, and
R. H.
Johnson
,
2000
:
Organizational modes of midlatitude mesoscale convective systems
.
Mon. Wea. Rev.
,
128
,
3413
3436
.
Schumacher
,
R. S.
, and
R. H.
Johnson
,
2005
:
Organization and environmental properties of extreme-rain-producing mesoscale convective systems
.
Mon. Wea. Rev.
,
133
,
961
976
.
Schumacher
,
R. S.
, and
R. H.
Johnson
,
2006
:
Characteristics of U.S. extreme rain events during 1999–2003
.
Wea. Forecasting
,
21
,
69
85
.
Tollerud
,
E. I.
,
D.
Rodgers
, and
K.
Brown
,
1987
:
Seasonal, diurnal, and geographic variations in the characteristics of heavy-rain-producing mesoscale convective complexes: A synthesis of eight years of MCC summaries
.
Preprints, 11th Conf. on Weather Modification, Edmonton, AB, Canada, Amer. Meteor. Soc., 143–146
.
Wilks
,
D. S.
,
1995
:
Statistical Methods in the Atmospheric Sciences: An Introduction
.
Academic Press, 467 pp
.
Winkler
,
J. A.
,
B. R.
Skeeter
, and
P. D.
Yamamoto
,
1988
:
Seasonal variations in the diurnal characteristics of heavy hourly precipitation across the United States
.
Mon. Wea. Rev.
,
116
,
1641
1658
.

Footnotes

*

Current affiliation: Dept. of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey.