characteristics of u.s. extreme rain events during 1999–2003

17
Characteristics of U.S. Extreme Rain Events during 1999–2003 RUSS S. SCHUMACHER AND RICHARD H. JOHNSON Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado (Manuscript received 3 January 2005, in final form 14 July 2005) ABSTRACT This study examines the characteristics of a large number of extreme rain events over the eastern two-thirds of the United States. Over a 5-yr period, 184 events are identified where the 24-h precipitation total at one or more stations exceeds the 50-yr recurrence amount for that location. Over the entire region of study, these events are most common in July. In the northern United States, extreme rain events are confined almost exclusively to the warm season; in the southern part of the country, these events are distributed more evenly throughout the year. National composite radar reflectivity data are used to classify each event as a mesoscale convective system (MCS), a synoptic system, or a tropical system, and then to classify the MCS and synoptic events into subclassifications based on their organizational structures. This analysis shows that 66% of all the events and 74% of the warm-season events are associated with MCSs; nearly all of the cool-season events are caused by storms with strong synoptic forcing. Similarly, nearly all of the extreme rain events in the northern part of the country are caused by MCSs; synoptic and tropical systems play a larger role in the South and East. MCS-related events are found to most commonly begin at around 1800 local standard time (LST), produce their peak rainfall between 2100 and 2300 LST, and dissipate or move out of the affected area by 0300 LST. 1. Introduction Extreme rainfall is responsible for a variety of soci- etal impacts, including flash flooding that can lead to damage, injury, and death. In an attempt to understand more about how these extreme-rain-producing weather systems are organized and the conditions in which they occur, Schumacher and Johnson (2005, hereafter SJ05) examined radar data and other observations for 116 extreme rain events in the eastern two-thirds of the United States over a 3-yr period. They found that ap- proximately 65% of the events were associated with mesoscale convective systems (MCSs), 27% were caused by synoptically forced weather systems, and 8% resulted from tropical cyclones and their remnants. In addition, although the types of MCSs capable of pro- ducing extreme rainfall were varied, they found that two patterns of MCS organization were most frequently observed. These types were dubbed “training line/ adjoining stratiform” (TL/AS; Fig. 1a), and “backbuild- ing/quasi-stationary” (BB; Fig. 1b). In the TL/AS pat- tern, convective cells typically move in a line-parallel direction, while there is very little motion in the line- perpendicular direction. This combination of motion characteristics leads to prolonged heavy rainfall at lo- cations along the convective line. Backbuilding/quasi- stationary MCSs occur when convective cells repeat- edly form upstream of their predecessors and pass over a particular area, leading to large local rainfall totals. This paper will provide more information about the climatological characteristics of these events, including their monthly and diurnal distributions, and informa- tion about the amount of rainfall that is typically ob- served with each type. The best example of a flash flood “climatology” in the scientific literature is the study by Maddox et al. (1979, hereafter MCH79). They examined the synoptic and mesoscale atmospheric conditions during 151 flash flood events in all parts of the United States and pin- pointed four surface and upper-air patterns in which the events typically formed. They also considered the spatial and temporal characteristics of these events. In total, MCH79 found that flash floods are possible dur- ing all seasons but are most common in the summer months (Fig. 2). However, this statement is highly de- pendent on the type of event being considered. “Syn- optic” events (those associated with significant large- Corresponding author address: Russ Schumacher, Dept. of At- mospheric Science, Colorado State University, Fort Collins, CO 80523. E-mail: [email protected] FEBRUARY 2006 SCHUMACHER AND JOHNSON 69 © 2006 American Meteorological Society

Upload: others

Post on 09-Feb-2022

2 views

Category:

Documents


0 download

TRANSCRIPT

Characteristics of U.S. Extreme Rain Events during 1999–2003

RUSS S. SCHUMACHER AND RICHARD H. JOHNSON

Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado

(Manuscript received 3 January 2005, in final form 14 July 2005)

ABSTRACT

This study examines the characteristics of a large number of extreme rain events over the easterntwo-thirds of the United States. Over a 5-yr period, 184 events are identified where the 24-h precipitationtotal at one or more stations exceeds the 50-yr recurrence amount for that location. Over the entire regionof study, these events are most common in July. In the northern United States, extreme rain events areconfined almost exclusively to the warm season; in the southern part of the country, these events aredistributed more evenly throughout the year. National composite radar reflectivity data are used to classifyeach event as a mesoscale convective system (MCS), a synoptic system, or a tropical system, and then toclassify the MCS and synoptic events into subclassifications based on their organizational structures. Thisanalysis shows that 66% of all the events and 74% of the warm-season events are associated with MCSs;nearly all of the cool-season events are caused by storms with strong synoptic forcing. Similarly, nearly allof the extreme rain events in the northern part of the country are caused by MCSs; synoptic and tropicalsystems play a larger role in the South and East. MCS-related events are found to most commonly begin ataround 1800 local standard time (LST), produce their peak rainfall between 2100 and 2300 LST, anddissipate or move out of the affected area by 0300 LST.

1. Introduction

Extreme rainfall is responsible for a variety of soci-etal impacts, including flash flooding that can lead todamage, injury, and death. In an attempt to understandmore about how these extreme-rain-producing weathersystems are organized and the conditions in which theyoccur, Schumacher and Johnson (2005, hereafter SJ05)examined radar data and other observations for 116extreme rain events in the eastern two-thirds of theUnited States over a 3-yr period. They found that ap-proximately 65% of the events were associated withmesoscale convective systems (MCSs), 27% werecaused by synoptically forced weather systems, and 8%resulted from tropical cyclones and their remnants. Inaddition, although the types of MCSs capable of pro-ducing extreme rainfall were varied, they found thattwo patterns of MCS organization were most frequentlyobserved. These types were dubbed “training line/adjoining stratiform” (TL/AS; Fig. 1a), and “backbuild-ing/quasi-stationary” (BB; Fig. 1b). In the TL/AS pat-

tern, convective cells typically move in a line-paralleldirection, while there is very little motion in the line-perpendicular direction. This combination of motioncharacteristics leads to prolonged heavy rainfall at lo-cations along the convective line. Backbuilding/quasi-stationary MCSs occur when convective cells repeat-edly form upstream of their predecessors and pass overa particular area, leading to large local rainfall totals.This paper will provide more information about theclimatological characteristics of these events, includingtheir monthly and diurnal distributions, and informa-tion about the amount of rainfall that is typically ob-served with each type.

The best example of a flash flood “climatology” inthe scientific literature is the study by Maddox et al.(1979, hereafter MCH79). They examined the synopticand mesoscale atmospheric conditions during 151 flashflood events in all parts of the United States and pin-pointed four surface and upper-air patterns in whichthe events typically formed. They also considered thespatial and temporal characteristics of these events. Intotal, MCH79 found that flash floods are possible dur-ing all seasons but are most common in the summermonths (Fig. 2). However, this statement is highly de-pendent on the type of event being considered. “Syn-optic” events (those associated with significant large-

Corresponding author address: Russ Schumacher, Dept. of At-mospheric Science, Colorado State University, Fort Collins, CO80523.E-mail: [email protected]

FEBRUARY 2006 S C H U M A C H E R A N D J O H N S O N 69

© 2006 American Meteorological Society

WAF900

scale weather systems) are distributed fairly evenlythroughout the year with slight maxima in the springand fall, while “frontal” and “mesohigh” (cool-air out-flow boundary) events occur almost exclusively duringthe warm season. This behavior is not surprising, sincethe baroclinic conditions necessary for synopticweather system (i.e., extratropical cyclone) develop-ment are most often in place in the spring and fall,whereas MCSs and other convective systems are moreprominent in summer. The research to be presentedherein will build upon these results by including infor-mation about the organization of the systems producingthe rainfall, rather than simply the prevailing environ-mental conditions.

Many others have worked to determine the climato-logical characteristics of flash flood and extreme rainevents in different regions of the United States, includ-ing Changnon and Vogel (1981) in Illinois, Winkler(1988) in Minnesota, Foufoula-Georgiou and Wilson(1990) in the Midwest, Houze et al. (1990) in Oklaho-ma, Giordano and Fritsch (1991) in the mid-Atlanticregion, Bradley and Smith (1994) in the southern GreatPlains, Konrad (1997) in the southeastern UnitedStates, Junker et al. (1999) in the Midwest, and Mooreet al. (2003) in the central United States. Althoughthese studies have provided important results region-ally, few studies have considered an area including theentire part of the country east of the Rockies.

Brooks and Stensrud (2000) objectively analyzedhourly precipitation data and found a July maximumfor rainfall rates greater than 25.4 mm (1 in.) h�1. Theiranalysis of the seasonal distribution of this heavy rain-fall shows that from October through March, rainfall ofthis magnitude is mainly confined to the Gulf coaststates; from April through September, it occurs rela-tively frequently in all parts of the country east of theRocky Mountains. One of the goals of their work was tomake forecasters aware of how likely heavy rainfallmight be during a particular time of year at a givenlocation, which is a goal of this research as well. Anunderstanding of the regional distribution of extremerainfall is one key aspect of attaining this goal, withBrooks and Stensrud’s work representing an importantfirst step. Though this study will not consider rainfallrate per se, it will hopefully provide additional infor-

FIG. 1. Schematic diagram of the radar-observed features of the(a) TL/AS and (b) BB patterns of extreme-rain-producing MCSs.Contours (and shading) represent approximate radar reflectivityvalues of 20, 40, and 50 dBZ. In (a), the low-level and midlevelshear arrows refer to the shear in the surface–925-hPa and 925–500-hPa layers, respectively. No consistent relationship was foundbetween the direction of the shear and the orientation of theconvection for BB MCSs; thus, no such vectors are shown in (b).The dash–dot line in (b) represents an outflow boundary; suchboundaries were observed in many of the BB MCS cases. Thelength scale at the bottom is approximate and can vary substan-tially, especially for BB systems, depending on the number ofmature convective cells present at a given time. From Schumacherand Johnson (2005).

FIG. 2. Monthly distribution of the flash flood events studied byMCH79.

70 W E A T H E R A N D F O R E C A S T I N G VOLUME 21

mation (such as diurnal distributions) about where andwhen extreme rain can be expected.

Also key to this discussion is the variation in extremerainfall frequency throughout the day. Wallace (1975)examined the diurnal variation of “heavy” precipitationacross the United States, and found a nocturnal maxi-mum for rain rates greater than 2.5 mm h�1 in theGreat Plains and Midwest, and an afternoon or eveningmaximum in the East and South. Winkler et al. (1988)examined the seasonal variability of the diurnal distri-bution for even heavier rainfall, and obtained resultssimilar to Wallace’s: in the warm season, heavy precipi-tation is maximized during the overnight hours in thecentral part of the country, but an afternoon/eveningmaximum is evident across the southern and easternstates. Similar results were presented in more recentstudies by Dai et al. (1999) and Knievel et al. (2004).Carbone et al. (2002) showed that these features of thediurnal cycle are caused by coherent convective “epi-sodes” that typically form over the Rocky Mountainsand propagate eastward.

Radar data can provide a further avenue for attaininginformation about the diurnal distribution of extremerainfall, as it is not limited by flood reports nor by thesparsity of the rainfall observing network. Now thatdata from the Weather Surveillance Radars-1988Doppler (WSR-88Ds) are readily available, such proj-ects can be undertaken relatively easily. In this study,radar data will be used to observe various temporalcharacteristics of extreme rainfall events, such as thetime of the heaviest rain and the times of precipitationonset and dissipation, aspects not always observable ata high resolution using traditional rain gauges.

In section 2, the data and methods used in the studyare presented. The overall and regional monthly fre-quency distributions of extreme rain events using botha fixed and a spatially varying case selection thresholdare presented in section 3. In section 4, these results arebroken down by storm type, as determined from radardata. Section 5 includes the diurnal distributions of theevents, and section 6 provides information about therainfall characteristics and the occurrence of flashflooding and severe weather in conjunction with theheavy rainfall in the events.

2. Data and methods

a. Selection of cases

As in SJ05, the National Weather Service (NWS)cooperative high-resolution 24-h rain gauge networkwill be used to select the “extreme rain events” used inthis study. For the month of May 2001, this network had

7923 active stations in the United States. Most of thestations in this network report 24-h rainfall in the morn-ing [0600 or 0700 local standard time (LST)], thoughsome stations report at other times of the day. Eventswere deemed extreme rain events when one or more ofthe gauges reported a 24-h rainfall total greater thanthe 50-yr recurrence amount (Hershfield 1961) for thatlocation (Fig. 3). Events exceeding this threshold arethose that have a 2% chance of occurrence in any givenyear, based on 30 yr of precipitation data. Since thecompletion of the analysis for the present study, theNWS has released updated precipitation recurrence in-terval data for parts of the eastern United States andthe Ohio Valley. However, since these new data onlycover part of the region of study, the older but complete(i.e., Hershfield 1961) thresholds will be used. The re-gion of study will be the same as in SJ05, namely thepart of the United States east of the Rocky Mountains,excluding Florida.

For this study, the 50-yr recurrence threshold wasapplied over a 5-yr period (1999–2003), which is 2 yrlonger than the period used in SJ05. After eliminatingbad rainfall data, it yielded 184 extreme rain events.Rainfall data were eliminated when there were no ra-dar echoes in the area during the 24-h reporting period,or when radar and rain gauge data did not seem tomatch and no other documentation could be found toconfirm that a large amount of rain actually fell in thatarea. For the purpose of this study, an “event” refers toa weather system that produces one or more rainfallobservations over the extreme rainfall threshold. Thistypically represents all or part of the 24-h period inwhich the rainfall was reported. However, a singleevent can include multiple 24-h periods if the sameweather system is responsible for the precipitation (e.g.,a tropical cyclone that produces heavy rainfall over sev-eral states in a 2–3-day period).

In addition to the 50-yr recurrence threshold, somedata will be presented in section 3 using a fixed case-selection criterion of 125 mm (24 h)�1 (125 mm is equalto approximately 4.92 in.). Though the spatially varyingthreshold described previously is most relevant for in-vestigating events that are truly extreme for their loca-tion, the spatially fixed threshold shows how frequentlya particular amount of rain occurs from one region tothe next. In the 1999–2003 time period, 382 such eventswere identified after eliminating bad rainfall reports asdescribed in the previous paragraph.

b. National composite radar reflectivity data

Each extreme rain event’s life cycle was then ob-served using composite radar reflectivity data from the

FEBRUARY 2006 S C H U M A C H E R A N D J O H N S O N 71

WSI Corporation NOWrad product. Data from theWSR-88Ds are used to generate this dataset, which haspixel resolution of 2 km � 2 km and temporal resolu-tion of 15 min. Each event was classified as either anMCS, a synoptic system, or a tropical system, based onthe radar observations. Convective systems with a re-gion of reflectivity �40 dBZ extending more than 100km in at least one direction and with durations between3 and 24 h were classified as MCSs, consistent with thecriteria of Orlanski (1975) and Parker and Johnson(2000). Events characterized by the strong large-scaleascent commonly associated with synoptic-scale fea-tures (i.e., extratropical cyclones) and/or lasting longerthan 24 h were classified as synoptic. Thus, elongated(longer than �1000 km) prefrontal squall lines andother convective systems that persisted for longer than24 h were classified as synoptic systems rather thanMCSs, though mesoscale aspects (and sometimes evenindividual MCSs) clearly played an important role inthe heavy rainfall. Events were classified as tropical ifthey were the direct result of a tropical cyclone or itsremnants. While mesoscale processes may be importantin all three of these classifications, the distinctions thathave been drawn are motivated in part by operationalforecasting concerns. Though the ingredients necessaryfor the development of extreme rainfall are the sameregardless of the strength of forcing (e.g., Doswell et al.

1996), there are some different challenges involvedwith forecasting the rainfall associated with a stronglyforced, elongated squall line compared with a landfall-ing tropical cyclone or an MCS occurring in an out-wardly benign large-scale setting (e.g., Maddox andDeitrich 1982).

Synoptic and MCS events were also arranged intosubclassifications based on their organizational struc-tures and system evolutions. These subclassificationsinclude distinguishing between convective and noncon-vective synoptic systems, as well as breaking the MCSevents into seven categories identified by Parker andJohnson (2000) and SJ05: TL/AS, BB, trailing strati-form (TS), parallel stratiform (PS), leading stratiform(LS), multiple MCSs, and other MCSs. The subclassifi-cations are based on the dominant pattern of organiza-tion at the time when the system was producing theextreme rainfall. The details and methods of classifyingevents are the same as described in SJ05.

Additionally, several temporal characteristics of eachevent (as inferred from the radar data) were recorded.These include the time of heavy rain onset (the time ofthe first echo �45 dBZ in the area where extreme rain-fall was reported), the time of peak rainfall (the hour inwhich high reflectivities were most persistent over thearea), and the time when all radar echoes ended ormoved out of the area. These times were cross-checked

FIG. 3. The 50-yr frequency for 24-h rainfall (in.) in the United States. Adapted fromHershfield (1961); figure courtesy of the Natural Resources Conservation Service of the U.S.Department of Agriculture. Also shown are regions that will be used, which are the same asthose used in Karl and Knight (1998). They will be referred to as (clockwise from upper left)plains, North, Northeast, Ohio–Mississippi valley, Southeast, and South.

72 W E A T H E R A N D F O R E C A S T I N G VOLUME 21

with hourly precipitation observations where availableand adjusted if necessary.

c. Flash flood and severe weather reports

Though rain gauge observations were used as an ob-jective way to select events for this study, it is alsohelpful to know whether the events selected actuallycaused flash floods, damage, or injury. In addition,since flash-flood-producing storms in the past havebeen associated with other types of severe weather(e.g., Smith et al. 2001; Rogash and Racy 2002), theproximity of severe wind, hail, and tornadoes to ex-treme rainfall could also be useful information. To thisend, two online databases were surveyed to determinewhether flash flooding and/or severe weather was re-ported in conjunction with the extreme rainfall. Thefirst is the National Climatic Data Center (NCDC)Storm Events database (information online at http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?wwEvent�Storms), which has written accounts of flash flooding,river flooding, severe weather, and several otherweather-related phenomena. The second source is thepreliminary storm reports archive from the Storm Pre-diction Center (SPC; information available online athttp://www.spc.noaa.gov/climo/). This database pro-vides a map of all tornadoes, severe wind, and severehail reports on a given day. Both databases include thetime of occurrence and a brief description of eachevent.

For each extreme rain event, the NCDC databasewas used to determine if flash flooding occurred at ornear the location where the extreme rainfall totals wereobserved. Similarly, the SPC archive was used to findthe locations and times of any nearby severe weatherreports. If tornadoes were reported, the NCDC data-base was then checked to find their Fujita scale ratings.If flash flooding, severe wind, severe hail, tornadoes, orsignificant tornadoes (F2 or greater) were reported inconjunction with the extreme-rain-producing storm sys-tem at approximately the same time as the heavy rain-fall was occurring, a “yes” was assigned for that phe-nomenon for that event. These determinations weremade as accurately as possible using the availabledatasets.

3. Monthly frequency distributions

a. Fixed threshold

Using the criteria already described, there were 382events in the years 1999–2003 where 125 mm of rainwas reported in 24 h at one or more gauge. As onemight expect, rainfall of this magnitude occurs mostfrequently in the southern part of the country (wheremoisture from the Gulf of Mexico is abundant) and isless frequent to the north (Fig. 4). [The regions used arethe same as those used in Karl and Knight (1998) andthe names that will be used are given in the caption forFig. 3.] Such events can occur in all seasons in the

FIG. 4. Monthly frequency distribution of 125 mm (24 h)�1 events in 1999–2003, by region(125 mm is equal to approximately 4.92 in.). The total distribution is shown at the bottom.Ordinates are scaled equally for the regional graphs and range from 0 to 30 events. The totalnumber of events in each region is shown below that region’s graph.

FEBRUARY 2006 S C H U M A C H E R A N D J O H N S O N 73

southern United States, but are mainly confined to thewarm season in the northern part of the country.

In total, such events are most common during Juneand July. In the South and Southeast regions, there arerelative minima in July and August, with relativemaxima before and after. The summer minima arelikely attributable to the onset of the Bermuda high andthe northward migration of the jet stream during thesummer. Bradley and Smith (1994), studying rainfall inOklahoma and northern Texas, found a similar bimodaldistribution in 125 mm (24 h)�1 events over a 43-yrperiod. In contrast, July is the peak month for 125 mm(24 h)�1 events in the North and Ohio–Mississippi val-ley region. Rainfall of this magnitude is relatively rarein the plains and Northeast regions. Nearly all suchevents in the plains region occur in the warm season,and those in the Northeast are most frequently ob-served in March, August, and September.

This analysis shows the frequency at which 125 mm(24 h)�1 rainfall events occur in the United States andin what months they are most likely to occur in differ-ent regions. As mentioned in the previous chapter, 125mm is a significant amount of precipitation for one dayin any area, and these results provide informationabout where and when this much rain might be ex-pected to fall. However, as these results show, it doesnot represent a truly extreme amount in all areas, and ithappens so rarely in others that any generalizationsmade about the events that cause it would not be verymeaningful. Therefore, the spatially varying extremerain threshold (Fig. 3) is applied, which yields the cases

(“extreme rain events”) that will be considered in therest of the study.

b. Spatially varying threshold

The application of this threshold provides a sample(totaling 184 events) that is much more suitable forfurther study, since the number of events is more ap-propriately balanced throughout the country. Theshapes of the monthly frequency distributions (Fig. 5)are mostly similar to those for the 125 mm (24 h)�1

events, and the consideration of more events in theplains and North regions leads to more meaningful re-sults with substantial warm-season maxima. Interest-ingly, in an area-weighted sense, there are more ex-treme rain events in the regions that are farther fromthe Gulf of Mexico and the Atlantic Ocean. However,as will be discussed later, a single event (such as a hur-ricane) occurring in these coastal areas has the poten-tial to produce a much greater amount of total precipi-tation than the events that typically occur in the centralpart of the country. The overall monthly frequency dis-tribution (bottom of Fig. 5) generally resembles thefindings of MCH79 (Fig. 2).

4. Results obtained using radar analysis

Since the monthly frequency distributions of extremerain events vary greatly throughout the United States, itis likely that the types of weather systems responsible

FIG. 5. As in Fig. 4, but for extreme rain events as defined in the text. Ordinates rangefrom 0 to 15 events on the regional graphs.

74 W E A T H E R A N D F O R E C A S T I N G VOLUME 21

for these events also vary both seasonally and fromregion to region. With this in mind, national compositeradar data were used to classify each event as synoptic,MCS, or tropical (as explained in SJ05 and in section 2).This analysis, which extends the results of SJ05 by 2 yr,shows that nearly two-thirds of all extreme rain eventsconsidered were associated with MCSs, and one-quarter were associated with synoptic weather systems(Table 1). Since almost all of the synoptic events alsoinvolved the repeated passage of deep convective cells,even more than two-thirds of the extreme rain eventscan be attributed to such convective processes.

MCSs were the most common extreme rain produc-ers in every region except the Southeast, where tropicalsystems caused the most extreme rain events. In theplains and North regions, MCSs accounted for a largemajority of the events, and the distribution in the Ohio–Mississippi valley region was very similar to the overalldistribution. In the Northeast and South regions, MCSswere still the most common extreme rain producers,but synoptic and tropical systems played a larger role inthose regions than in the region of study as a whole.

Though the overall percentage of tropical events wasrelatively low, the three events in the period of studythat caused the most widespread and destructive flood-ing were all tropical cyclones: Hurricane Floyd (1999)along the East Coast, Tropical Storm Allison (2001) inthe Gulf coast states, and Hurricane Isidore (2002) inthe Gulf coast states and extending into the Ohio Val-ley. So, while there may be relatively few extreme rainevents associated with tropical cyclones, the potentialfor damage and injury is much greater when they dooccur. (The likelihood that a given landfalling tropicalcyclone will produce extreme rain is much higher thanthat for a given MCS or synoptic system as well.)

The MCS and synoptic events were divided into sub-classifications, as described in section 2 and in SJ05.The number of events grouped into each subclassifica-tion is given in Table 2. (This table is identical to Table2 of SJ05, except that the 5-yr period 1999–2003 is con-sidered instead of the 3-yr period 1999–2001.) Synopticevents that were dominated by convective storms were

responsible for many more cases of extreme precipita-tion than those where nonconvective processes pro-duced most of the rainfall. The TL/AS pattern was themost common MCS type, followed by the BB and TStypes. This result is slightly different from that pre-sented in SJ05; in that paper (covering the 1999–2001time period), there were two more BB events than TSevents. In this period of study, there were an equalnumber of BB and TS events.

The approximate location and storm type of eachextreme rain event from 1999 to 2003 is shown in Fig. 6.These panels allow for comparison between yearswithin the sample and help to illustrate where certainstorm types are most prevalent. In 1999, the extremerain events were heavily concentrated in the Dakotas,Iowa, and Illinois. There were also several tropical cy-clones that affected the East Coast, including Hurri-canes Dennis, Floyd, and Irene. In 2000, the eventswere more evenly distributed throughout the area ofstudy, though there were several events in both Wis-consin and Illinois. Several major synoptic events af-fected the South in 2001, as did Tropical Storm Allison.

TABLE 1. Weather systems associated with extreme rain events, by region. Numbers in parentheses represent the percentage of events(rounded to the nearest whole percent) in a given region that are caused by the respective storm type. The sum of the regional valuesfor each storm type does not necessarily equal the total in the right-hand column, because events spanning more than one region arecounted in both regions but only once in the overall total.

System Plains North OH–MS NE South SE Total

MCS 22 (76) 43 (90) 34 (67) 7 (44) 20 (54) 7 (29) 122 (66.3)Synoptic 7 (24) 5 (10) 15 (29) 6 (37) 13 (35) 6 (25) 46 (25.0)Tropical 0 0 2 (4) 3 (19) 4 (11) 11 (46) 16 (8.7)Total 29 48 51 16 37 24 184

TABLE 2. Number of extreme rain events associated with thesubclassifications of synoptic systems and MCSs.

Synoptic systems Events% of

synoptic% of allevents

Convective 43 93.5% 23.4%Nonconvective 3 6.5% 1.6%Total 46 100% 25.0%

MCSs Events% ofMCSs

% of allevents

Training line/adjoining stratiform(TL/AS) 41 33.6% 22.3%

Backbuilding/quasi-stationary (BB) 24 19.7% 13.0%Trailing stratiform (TS) 24 19.7% 13.0%Other MCSs 17 13.9% 9.2%Parallel stratiform (PS) 11 9.0% 6.0%Multiple MCSs 3 2.5% 1.6%Leading stratiform (LS) 2 1.6% 1.1%Total 122 100% 66.3%

FEBRUARY 2006 S C H U M A C H E R A N D J O H N S O N 75

In 2002, Minnesota was impacted by a large number ofextreme rain events that caused major flooding. Hurri-cane Isidore also caused significant flooding after itproduced widespread heavy rain from Louisiana north-ward to Indiana. Finally, in 2003, the center of activitymoved eastward, with several extreme rain events inTennessee, Virginia, and North Carolina, and relativelyfew in the plains.

Figure 6 also illustrates that the distribution of theMCS subclassifications varied from region to region.This distribution is quantified in Table 3. In the North-

east, other MCSs, which tend to be less organized thanthe other MCS types, played a large role. In the plainsand the North regions, there were fewer MCS eventsclassified as other than in the overall distribution.These contrasts likely come about because low-leveljets, boundaries, and other mesoscale forcings requiredfor organized MCSs are fairly common in the centralUnited States, while they are less common in the east-ern part of the country (e.g., Laing and Fritsch 1997). Itis also important to note, however, that there were onlyseven MCS events in the Northeast region, so most of

FIG. 6. Approximate locations and storm types of all extreme rain events in (a) 1999, (b) 2000, (c) 2001, (d) 2002,and (e) 2003. Lettering system is as follows: AS, TL/AS; MM, multiple MCSs; O, other MCSs; SC, synoptic/convective; SN, synoptic/nonconvective; and T, tropical. The BB, TS, LS, and PS systems are lettered accordingly.Cases where widespread extreme rainfall was reported are outlined.

76 W E A T H E R A N D F O R E C A S T I N G VOLUME 21

the results from this region should be interpreted withcaution.

The monthly frequency distribution of extreme rainevents, separated by storm type, shows that most of theevents occurring in the summer months are caused byMCSs, but nearly all of the cool-season events are as-sociated with synoptic systems (Fig. 7 and Table 4).This result is not surprising, considering that thestrongly baroclinic conditions that are prevalent in thecool season are supportive of strong synoptic systems,and MCSs typically occur in the weaker forcing andgreater moisture of the warm season in the UnitedStates. However, it expands upon Fritsch et al.’s (1986)finding that 30%–70% of the total warm season pre-cipitation in the Midwest results from MCSs by showingthat 74% of extreme warm season (April–September)precipitation events in the entire eastern United Statesresult from MCSs (Table 4). Although this does notspeak to the proportion of the total amount of precipi-tation that results from extreme-rain-producing MCSs,it further emphasizes the important role that MCSs playin determining the warm-season precipitation charac-teristics of the United States.

It was shown above that nearly all extreme rain

events in the plains and the North regions occur duringthe warm season. In addition, nearly all of them areassociated with MCSs. As such, the monthly frequencydistributions for these regions (Figs. 8a and 8b) show anexpected result: MCS-caused events that occur betweenMay and September are the dominant extreme-rain-producing phenomena in these locations. Each of theseregions did receive a few synoptic events (mainly in thespring and early summer). At locations this far north,temperatures are typically too cold and the necessarymoisture is unavailable for extreme precipitation to oc-cur outside of the warm season. Though extratropicalcyclones certainly affect these regions during the win-ter, they usually bring snow instead of rain.

Slightly farther south and east in the Ohio–Mississippi valley region, synoptic systems play agreater role. Here, MCSs still produce the majority ofthe extreme rain events and are maximized in July, butcool-season synoptic events are also important (Fig.8c). Because the temperatures and dewpoints in thisregion, on average, are higher than those farther north,some extratropical cyclones that traverse this regionduring the autumn and winter bring long-lived convec-tive systems rather than heavy snow.

In the northeastern United States, extreme rainfallappears to occur on a different annual cycle, with nearlyall of the events occurring in August and September(Fig. 8d). (Note once again, however, that there arerelatively few Northeast events in the sample fromwhich to draw these conclusions.) August had the maxi-mum for MCS-related extreme rain events in theNortheast. Only 1 month later, however, MCSs, synop-tic systems, and tropical systems all contributed equally.Though MCSs do play an important role in the North-east, the number of extreme rain events in that regionis also strongly dependent on tropical cyclones, an ef-fect that will be discussed below.

The monthly frequency distribution in the Southeastregion (Fig. 8e) is unique because it is strongly modu-

TABLE 3. Distribution of extreme-rain-producing MCS types by region. Numbers in parentheses represent the percentage of MCSevents in that region associated with the respective storm type. For example, the entry in the upper left indicates that TL/AS systemsaccounted for 36.4% of plains MCS cases.

System Plains North OH–MS NE South SE Total

TL/AS 8 (36.4) 20 (46.5) 10 (29.4) 0 (0) 8 (40.0) 1 (14.3) 41 (33.6)BB 5 (22.7) 9 (20.9) 4 (11.8) 1 (14.3) 7 (35.0) 1 (14.3) 24 (19.7)TS 4 (18.2) 8 (18.6) 9 (26.5) 1 (14.3) 2 (10.0) 2 (28.6) 24 (19.7)LS 2 (9.1) 0 (0) 0 (0) 0 (0) 0 (3.4) 0 (7.1) 2 (1.6)PS 1 (4.5) 4 (9.3) 5 (14.7) 0 (0) 0 (0) 1 (14.3) 11 (9.0)Multiple MCSs 0 (0) 1 (2.3) 1 (2.9) 1 (14.3) 0 (0) 0 (7.1) 3 (2.5)Other MCSs 2 (9.1) 1 (2.3) 5 (14.7) 4 (57.1) 3 (15.0) 2 (28.6) 17 (13.9)Total 22 43 34 7 20 7 122

FIG. 7. Monthly frequency distribution of all extreme rainevents, separated by storm type.

FEBRUARY 2006 S C H U M A C H E R A N D J O H N S O N 77

lated by tropical cyclone activity. September is the mostactive month in this region, mainly because severaltropical systems affected this region in September1999–2003. Since tropical cyclone activity is extremelyvariable from year to year, it is possible that the resultsfound here would be completely different if another5-yr period were chosen for study. Thus, generaliza-tions made about the monthly distribution in the South-east region (and to some extent, the Northeast region)may not be as robust as those made about other re-gions.

Finally, in the South region, extreme rain events oc-cur most often in the late spring, with a secondary maxi-mum in autumn. Most of the springtime events resultfrom MCSs, while all three types play an important rolein the fall (Fig. 8f). Extreme rain events can occur in allmonths of the year in the South, as evidenced by thewintertime synoptic events.

The monthly frequency distributions of the indi-vidual MCS subclassifications are generally similar tothe overall distribution for MCSs, with a few notabledifferences (Table 4). For instance, the BB MCSs arealmost completely confined to the months of June andJuly, whereas TL/AS systems are more spreadoutthrough the year. This result likely comes about be-cause BB systems only occur when there is weak upper-level forcing (SJ05), and such conditions are usuallyobserved only in the summer in the United States. Onthe other hand, the boundaries that typically force TL/AS MCSs are observed throughout more of the year.

5. Diurnal distributions

To understand the driving forces behind the convec-tion, knowing the time of day at which it is most likelyto occur is crucial. The results of previous work regard-ing the diurnal characteristics of convection in generaland of extreme rainfall in particular were discussed in

the introduction, and the results to be presented hereshould provide further insights into the diurnal cycle ofprecipitation. As briefly mentioned in section 2, threetimes in the life cycle of each extreme rain event will bediscussed in the following. The onset time used here isthe hour of heavy rain onset as determined using theradar data (the time of the first echo �45 dBZ at theextreme rainfall location for most systems), the peaktime is the hour when the heaviest rain was falling at thestation(s) where extreme rain was reported (inferredfrom the radar data and cross-checked with hourly pre-cipitation data when available), and the end time is thehour when all radar echoes dissipated or moved out ofthe area. Synoptic, tropical, and multiple MCS caseshave been eliminated from the diurnal distributions be-cause their long durations make pinpointing onset andpeak rainfall times difficult, and possibly not evenmeaningful. Thus, the discussion to follow pertains onlyto extreme-rain-producing MCSs. The diurnal charac-teristics will be presented in an overall sense, and thenboth by region and by storm type, since each of theseperspectives is helpful in its own way.

In general agreement with past studies regardingheavy rainfall and flash floods, the storms producingextreme rainfall in this sample most often developed inthe late afternoon and evening, peaked after dark, anddissipated or moved out of the area in the early morn-ing hours (Fig. 9). The onset time for the “average”extreme rain event was around 1800 LST, with a peakbetween 2100 and 2300 LST and an end time around0300 LST. This supports previous findings that mostflash flood events are nocturnal (e.g., MCH79); al-though a peak rainfall time of 2100 LST may not belong after sunset during the summer, the flooding canoccur up to 6 h after the causative rainfall—well afterdark. Events began at all times of the day, but relativelyfew developed between 0200 and 1000 LST, with anassociated minimum in rainfall peaks between 0500 and

TABLE 4. Monthly frequency distribution of extreme rain events, including MCS subclassifications.

Month TL/AS BB TS PS LS Multiple MCSs Other All MCSs Synoptic Tropical Total

Jan 0 0 0 0 0 0 0 0 4 0 4Feb 0 0 0 0 0 0 0 0 2 0 2Mar 2 0 1 0 0 0 0 3 3 0 6Apr 4 1 2 0 0 0 0 7 4 0 11May 7 1 3 1 0 0 2 14 5 0 19Jun 10 6 3 2 0 0 3 24 6 3 33Jul 7 12 4 4 1 1 7 36 3 1 40Aug 5 2 6 3 1 1 3 21 4 2 27Sep 1 2 5 1 0 1 2 12 7 8 27Oct 3 0 0 0 0 0 0 3 2 2 7Nov 2 0 0 0 0 0 0 2 5 0 7Dec 0 0 0 0 0 0 0 0 1 0 1

78 W E A T H E R A N D F O R E C A S T I N G VOLUME 21

1300 LST. As a whole, the diurnal cycle presented inFig. 9 is very similar to that found by Jirak et al. (2003)for a large population of MCSs. These findings alsoagree with a conceptual model where convective stormsform as a result of daytime heating, increase in cover-age and intensity (and take on the structures that allowthem to produce large local rainfall totals) as the low-level jet intensifies after dark, and either dissipate ormove with greater speed in the early morning hours.Such processes have been described in part by Wallace(1975), MCH79, SJ05, and others. It should be notedthat the end time presented in these results does notnecessarily indicate the complete dissipation of the con-vective system, only that precipitation has ended at the

location where extreme rainfall was reported. TheMCSs responsible for extreme rainfall totals often per-sisted for several additional hours after leaving the areawhere they initially produced the heavy rains.

The diurnal frequency distributions in the plains andNorth regions are very similar to the overall distribu-tion, with heavy rain beginning in the late afternoon/evening, peaking in the nighttime hours, and ending inthe early morning (Figs. 10a and 10b). In the Ohio–Mississippi valley region (Fig. 10c), the most commononset, peak, and end times were similar to those in thesample as a whole, but the distributions of these timesare broader than in the plains and North regions. Thisindicates that extreme-rain-producing convection in the

FIG. 8. Monthly frequency distribution of all extreme rain events, separated by storm type, for the (a) plains,(b) North, (c) Ohio–Mississippi valley, (d) Northeast, (e) Southeast, and (f) South regions.

FEBRUARY 2006 S C H U M A C H E R A N D J O H N S O N 79

Ohio–Mississippi valley may occur at a wider variety oftimes than in the regions farther north and west. TheSouth region (Fig. 10d), along with a similarly widediurnal distribution, had an average peak time at 2100

UTC but a somewhat later end time, suggesting thatextreme rain events in the South have, on average, alonger duration than those in other regions. (Recall,however, that the amount of rain required for an event

FIG. 9. Diurnal frequency distribution for heavy rain onset, peak rainfall, and rainfall end forall 122 MCS-related extreme rain events, excluding the three classified as multiple MCSevents. Number of events is smoothed with a 3-h running mean.

FIG. 10. Diurnal frequency distribution for heavy rain onset, peak rainfall, and rainfall end for MCS-relatedextreme rain events, for the (a) plains, (b) North, (c) Ohio–Mississippi valley, and (d) South regions. Number ofevents is smoothed with a 3-h running mean.

80 W E A T H E R A N D F O R E C A S T I N G VOLUME 21

in the South to be “extreme” is greater than in otherregions, so there may be an inherent bias towardlonger-duration events in the South.) There were toofew MCS events in the Northeast and Southeast regionsto make meaningful conclusions about their diurnal dis-tributions.

A possible explanation for the broader diurnal dis-tributions observed in the South and Ohio–Mississippivalley regions is a change from a nocturnal rainfall re-gime to an afternoon regime as one moves from west toeast across these regions (e.g., Wallace 1975). To testthis idea, the diurnal distributions were calculated sepa-rately for the events occurring in the western parts ofthese regions and those occurring on the eastern side.Interestingly, these distributions were nearly indistin-guishable. This similarity may be because the diurnalcycle for very heavy rainfall in these areas is differentfrom that for the lighter rainfall rates studied by Wal-lace (1975). Crysler et al. (1982) showed that in Ten-nessee and West Virginia, intense rainfalls of short du-rations occurred most frequently during the afternoonand evening, but those lasting longer (with greateroverall rainfall totals) typically occurred during thenight and early morning hours.

The diurnal frequency distributions by storm typegenerally follow the overall distribution, though somestorm types have wider distributions than others (Fig.11). The TL/AS (Fig. 11a) and PS (Fig. 11d) distribu-tions have typical onset, peak, and end times that aresimilar to the overall MCS distributions, while the otherMCS (Fig. 11e) category has a distribution that is lessstraightforward. The most meaningful comparison maybe between the distributions for BB (Fig. 11b) and TS(Fig. 11c) MCSs, since they have exactly the same num-ber of events in the sample. The TS distributions foronset, peak, and end times have a slightly higher am-plitude and are somewhat narrower than those for theBB systems, indicating that perhaps TS MCSs aremodulated by the diurnal cycle to a greater extent thanare BB MCSs. The reason for this is not entirely clear,but it may be that MCSs that take on the TS structureare more likely to be forced by diurnal processes overthe western United States, such as in the “rain streaks”described by Carbone et al. (2002), while BB MCSs aremore tied to mesoscale processes such as outflowboundaries. This would still lead to the nocturnal maxi-mum for BB MCSs that is observed, but one that is lessdistinct than for some of the other MCS types.

The diurnal frequency distributions of extreme rainevents suggest that some storm types (e.g., TL/AS andTS MCSs) and the convective systems in some regions(mainly in the northern part of the country) operate on

a fairly consistent diurnal cycle, and other types andother regions have less predictable diurnal characteris-tics (e.g., BB MCSs and those in the South region).These results imply that in the northern United States,where there is also a strong seasonal cycle for extremerainfall, the ingredients for extreme precipitation onlycome into place at very specific times of year and timesof day. Or, in other words, a relatively rare set of cir-cumstances (e.g., sufficient moisture, the presence of alow-level jet, or a boundary with a particular orienta-tion), which only occurs in certain months and at cer-tain times, is needed for extreme rainfall to occur in thenorthern part of the country. In other regions, a widervariety of conditions (less dependent on the diurnal andseasonal cycles) may support such events. The samemay apparently be said for a few of the storm types.

6. Rainfall, flash flooding, and severe weatherinformation

a. Rainfall characteristics

In this section, the maximum rainfall total reportedduring each event is considered. Every event selectedfor this study has exceeded a certain threshold to en-sure that it is extreme for its area, however, there is stilla question of which events tend to cause the most rain-fall and have the most potential to cause serious dam-age. Certainly, these values are highly dependent on thedensity of the rain gauge network in the vicinity of theevent—a high-resolution gauge network is more likelyto capture the most extreme part of the storm than alower-resolution network—but the objective statisticsto be shown here agree generally with our own subjec-tive ideas about which events were the most severe.

Of the MCS-related extreme rain events, BB systemstended to produce higher maximum 24-h rainfall totalsthan the other MCS types (Table 5, far-right column),with an average maximum report of 199.6 mm (7.9 in.).The reason that BB systems, on average, are capable ofproducing higher rainfall totals is that they appear tohave no inherent temporal limit on their backbuildingand echo training effects. One can imagine such behav-ior lasting for many hours, as long as the moisturesource for the convection is not removed. In contrast,forward-propagating linear MCSs usually have a con-vective line with certain motion characteristics, whichcan also result in extreme rainfall but only for the du-ration that the length of the line allows. (However,these statistics do not give any information about theoverall rainfall coverage, which may be greater in theforward-propagating linear MCSs even though the ex-treme rainfall coverage is less.)

FEBRUARY 2006 S C H U M A C H E R A N D J O H N S O N 81

For purposes of comparison, the rainfall statistics forsynoptic and tropical systems have been included inTable 5. Tropical cyclone–related extreme rain eventswere relatively rare but were by far the most devastat-ing. The rainfall statistics show that the average maxi-mum rainfall report for tropical events is over 70 mmgreater than that for the BB MCSs.

Information about the storm types that caused thehighest rainfall totals in each region is also shown inTable 5. The important data in this table are in thecolumns; intercomparing between regions is not helpfulbecause the spatially varying threshold makes the av-erage maximum rainfall in the South inherently higher

than that in the North. The BB systems produced thehighest rainfall totals (of the MCS types) everywherebut the Southeast region. In the Southeast, however,there were only one or two examples of each MCSsubclassification, and therefore these results are likelyinsignificant.

b. Flash flooding

The main reason that rain gauge observations, as op-posed to flash flood reports, were chosen as the criteriafor selecting cases for this study is that flash flood re-ports are dependent on so many factors other than rain-fall itself. First, using flash flooding adds the hydrologi-

FIG. 11. Diurnal frequency distribution of heavy rain onset, peak rainfall, and rainfall end for (a) TL/AS,(b) BB, (c) TS, (d) PS, and (e) other MCS extreme rain events. Number of events is smoothed with a 3-h runningmean.

82 W E A T H E R A N D F O R E C A S T I N G VOLUME 21

cal component to the problem, which introduces a greatdeal of complexity. Second, if flash flooding occurs,whether or not it actually gets reported depends onwhere it occurs and whether anyone was there to ob-serve it. The rainfall observations are not perfect either,and the observing network has large gaps, but they pro-vide a much more objective (and meteorological) ap-proach to selecting cases.

With this in mind, one way to determine whether thethresholds developed for selecting extreme rain eventsare appropriate is to see if the cases selected were ac-tually associated with flash flooding. If the extreme rainevents were not responsible for flash flooding, thenstudying them would not get us any closer to solving the“flash flood problem.” Since only 24-h rainfall is beingused here, and since rain rate also plays a role, it ispossible that many of the events meeting the extremerain criteria could be longer-duration events that pro-duce copious amounts of rainfall but occur over longenough periods of time that they do not cause flooding.However, almost all of the cases used in this study didin fact cause flash flooding.

In total, 90.2% of MCS cases and 89.1% of all ex-treme rain events had corresponding flash flood reports(Table 6). The high percentage of cases with flashflooding was fairly consistent across the subclassifica-tions as well, as the only types with significantly lowerpercentages were those with only a few cases to con-sider.

c. Severe weather

Although severe weather is not as directly related tothe events studied in this project as flash flooding, op-erational forecasters have to issue warnings for bothwhile the events are occurring. A full treatment of howand why each of these types of systems produce severeweather is beyond the scope of this study, but dataabout how often each storm type produces hail, severewinds, and tornadoes may yield some informationabout the convective processes that are at work in eachof them. More detailed information regarding the tim-

ing of severe weather in relation to heavy rainfall canbe found in Wallace (1975), Maddox et al. (1986), andRogash and Racy (2002).

In total, just under half of the extreme rain eventsalso produced at least one severe wind report (58 mphor greater, as defined by the NWS; 1 mph � 0.447m s�1), 42% produced severe hail (0.75 in. or greater;1 in. � 25.4 mm), almost one-quarter spawned a tor-nado, and less than 5% produced a significant (rated F2or higher on the Fujita scale) tornado (Table 7). Thefact that so many extreme-rain-producing systems alsocaused severe winds seems to be somewhat of a con-tradiction, since severe windstorms typically require adry layer aloft but heavy rain events need a deep moistlayer. However, since organized MCSs and synopticsystems are responsible for so many of these events, itis possible that they attain a state whereby they canproduce both extreme rainfall and strong straight-linewinds, such as the wet microbursts described by Atkinsand Wakimoto (1991) and others. This topic is sug-gested for future investigation. Tornadoes (and signifi-cant tornadoes) have been previously documented inconjunction with flash flood events (e.g., Rogash and

TABLE 6. Number and percentage of extreme rain event typesassociated with flash flood (FF) reports.

SystemNo. with

FF reportsTotalNo. Percentage

TL/AS 39 41 95.1BB 22 24 91.7TS 20 24 83.3LS 1 2 50.0PS 11 11 100.0Multiple MCS 2 3 66.7Other MCSs 15 17 88.2All MCSs 110 122 90.2Synop–convective 38 43 88.4Synoptic–nonconvective 1 3 33.3All synoptic 39 46 84.8Tropical 15 16 93.8All events 164 184 89.1

TABLE 5. Average maximum 24-h rainfall total by region and storm type for several types of extreme rain events. For example, theupper-left entry indicates that the maximum reported rainfall averaged over all plains TL/AS events is 135.5 mm. The top entry in the“all regions” column, for instance, shows that the average maximum reported rainfall for all TL/AS events is 165.0 mm.

System Plains North OH/MS NE South SE All regions

TL/AS 135.5 146.0 178.8 N/A 202.7 284.5 165.0BB 173.6 166.5 179.6 363.5 247.7 203.2 199.6TS 140.7 134.1 143.0 134.4 170.2 184.2 147.2Other MCSs 108.2 96.8 136.6 147.5 211.8 199.9 154.2PS 164.6 154.8 172.5 N/A N/A 180.3 166.1Synoptic 126.5 115.9 173.2 148.9 224.1 192.2 175.8Tropical N/A N/A 216.9 197.0 404.6 240.4 271.9

FEBRUARY 2006 S C H U M A C H E R A N D J O H N S O N 83

Racy 2002), and a number of events of this type arefound in this sample. However, the spatial and tempo-ral resolutions of the radar data used herein do notallow for a more detailed analysis of any supercellstructures that may have been embedded in the ex-treme-rain-producing MCSs.

7. Summary and conclusions

Using rain gauge observations from the part of theUnited States east of the Rocky Mountains (excludingFlorida) in 1999–2003, it was found that 184 events ex-ceeded the 50-yr recurrence interval for 24-h precipita-tion accumulation. These cases were deemed extremerain events. The overall characteristics of these eventsare generally consistent with past studies of heavy pre-cipitation and flash floods:

• Extreme rain events in the complete area of studyoccurred most frequently in July.

• In the northern part of the country, extreme rainevents were confined almost exclusively to the warmseason; in the South, such events were generally lessdependent on season.

As in SJ05, national composite radar reflectivity datawere then used to make a quantitative determinationabout what types of weather systems are most oftenresponsible for producing extreme rainfall totals. Theseradar data were observed over the life cycle of eachextreme rain event to classify it as a synoptic system, amesoscale convective system (MCS), or a tropical sys-tem. The synoptic and MCS events were also put intopreviously established subclassifications based on theirradar-indicated organizational structures. Several re-sults have emerged from this work:

• The radar analysis showed that nearly two-thirds ofall extreme rain events were associated with MCSs,and 25% resulted from synoptic weather systems.

• The three most common patterns of MCS organiza-tion leading to extreme rainfall were TL/AS, BB, andTS.

• In the plains and North regions, an even larger pro-portion of the events was associated with MCSs, andin the Southeast, tropical systems played a large role.

• In almost the entire area of study, MCSs were thedominant summertime extreme rain producers. Sev-enty-four percent of summertime extreme rain eventsin the entire domain were caused by MCSs. In con-trast, synoptically forced systems produced the mostextreme rain events outside the warm season.

• Extreme rainstorms typically have an onset timearound 1800 LST, peak between 2100 and 2300 LST,and dissipate or move out of the affected area by0300 LST. These results are consistent with past workshowing that flash floods are most commonly noctur-nal.

Other information about the extreme rain events inthe sample, such as their rainfall production and wheth-er they were accompanied by flash flooding or severeweather was also noted:

• Tropical systems produced by far the most precipita-tion of any of the storm types. Of the MCS-relatedextreme rain events, BB MCSs tended to produce thehighest maximum 24-h rainfall totals.

• Nearly 90% of the extreme rain events had corre-sponding flash flood reports.

• Just under half of the extreme-rain-producing stormsalso produced at least one severe wind report, 42%produced severe hail, almost one-quarter spawned a

TABLE 7. As in Table 6 except for severe hail, winds, tornadoes, and significant tornadoes.

System

Severe hail Severe wind Tornadoes Significant tornadoes

No. of events % No. % No. % No. %

TL/AS 23 56.1 23 56.1 8 19.5 2 4.9BB 12 50.0 13 54.2 3 12.5 1 4.2TS 11 45.8 11 45.8 4 16.7 1 4.2LS 1 50.0 2 100.0 0 0 0 0PS 7 63.6 7 63.6 4 36.4 0 0Multiple MCSs 1 33.3 2 66.7 0 0 0 0Other MCSs 4 23.5 4 23.5 0 0 0 0All MCSs 59 48.4 62 50.8 19 15.6 4 3.3Synoptic–convective 17 39.5 23 53.5 15 34.9 4 9.3Synoptic–nonconvective 0 0 0 0 0 0 0 0All synoptic 17 37.0 23 50.0 15 32.6 4 8.7Tropical 2 12.5 6 37.5 9 56.3 0 0All events 78 42.4 91 49.5 43 23.4 8 4.3

84 W E A T H E R A N D F O R E C A S T I N G VOLUME 21

tornado, and less than 5% produced a significant tor-nado.

Acknowledgments. Precipitation data were pro-vided by the National Climatic Data Center. The WSINOWrad data were obtained from the Global Hydrol-ogy Resource Center at the Global Hydrology and Cli-mate Center, Huntsville, Alabama. The authors thankDavid Ahijevych for assistance in obtaining additionalradar data. The authors also thank three anonymousreviewers for their helpful suggestions. This researchwas supported by National Science Foundation GrantATM-0071371, and the first author was partially sup-ported by a one-year American Meteorological SocietyGraduate Fellowship.

REFERENCES

Atkins, N. T., and R. M. Wakimoto, 1991: Wet microburst activityover the southeastern United States: Implications for fore-casting. Wea. Forecasting, 6, 470–482.

Bradley, A. A., and J. A. Smith, 1994: The hydrometeorologicalenvironment of extreme rainstorms in the southern plains ofthe United States. J. Appl. Meteor., 33, 1418–1431.

Brooks, H. E., and D. J. Stensrud, 2000: Climatology of heavy rainevents in the United States from hourly precipitation obser-vations. Mon. Wea. Rev., 128, 1194–1201.

Carbone, R. E., J. D. Tuttle, D. A. Ahijevych, and S. B. Trier,2002: Inferences of predictability associated with warm sea-son precipitation episodes. J. Atmos. Sci., 59, 2033–2056.

Changnon, S. A., Jr., and J. L. Vogel, 1981: Hydroclimatologicalcharacteristics of isolated severe rainstorms. Water Resour.Res., 17, 1694–1700.

Crysler, K. A., R. A. Maddox, L. R. Hoxit, and B. M. Muller,1982: Diurnal distribution of very heavy precipitation overthe central and eastern United States. Natl. Wea. Dig., 7 (1),33–37.

Dai, A., F. Giorgi, and K. E. Trenberth, 1999: Observed andmodel-simulated diurnal cycles of precipitation over the con-tiguous United States. J. Geophys. Res., 104, 6377–6402.

Doswell, C. A., III, H. E. Brooks, and R. A. Maddox, 1996: Flashflood forecasting: An ingredients-based methodology. Wea.Forecasting, 11, 560–581.

Foufoula-Georgiou, E., and L. L. Wilson, 1990: In search of regu-larities in extreme rainstorms. J. Geophys. Res., 95, 2061–2072.

Fritsch, J. M., R. J. Kane, and C. R. Chelius, 1986: The contribu-tion of mesoscale convective weather systems to the warm-season precipitation in the United States. J. Appl. Meteor., 25,1333–1345.

Giordano, L. A., and J. M. Fritsch, 1991: Strong tornadoes andflash-flood-producing rainstorms during the warm season inthe mid-Atlantic region. Wea. Forecasting, 6, 437–455.

Hershfield, D. M., 1961: Rainfall frequency atlas of the UnitedStates for durations from 30 minutes to 24 hours and returnperiods from 10 to 100 years. U.S. Weather Bureau Tech.Paper 40, 61 pp.

Houze, R. A., Jr., B. F. Smull, and P. Dodge, 1990: Mesoscale

organization of springtime rainstorms in Oklahoma. Mon.Wea. Rev., 118, 613–654.

Jirak, I. L., W. R. Cotton, and R. L. McAnelly, 2003: Satellite andradar survey of mesoscale convective system development.Mon. Wea. Rev., 131, 2428–2449.

Junker, N. W., R. S. Schneider, and S. L. Fauver, 1999: A study ofheavy rainfall events during the great Midwest flood of 1993.Wea. Forecasting, 14, 701–712.

Karl, T. R., and R. W. Knight, 1998: Secular trends of precipita-tion amount, frequency, and intensity in the United States.Bull. Amer. Meteor. Soc., 79, 231–241.

Knievel, J. C., D. A. Ahijevych, and K. W. Manning, 2004: Usingtemporal modes of rainfall to evaluate the performance of anumerical weather prediction model. Mon. Wea. Rev., 132,2995–3009.

Konrad, C. E., II, 1997: Synoptic-scale features associated withwarm season heavy rainfall over the interior southeasternUnited States. Wea. Forecasting, 12, 557–571.

Laing, A. G., and J. M. Fritsch, 1997: The global population ofmesoscale convective complexes. Quart. J. Roy. Meteor. Soc.,123, 389–405.

Maddox, R. A., and W. Deitrich, 1982: Meteorological settingsassociated with simultaneous occurrences of severe thunder-storms and flash floods. Natl. Wea. Dig., 7 (3), 36–43.

——, C. F. Chappell, and L. R. Hoxit, 1979: Synoptic and meso-�scale aspects of flash flood events. Bull. Amer. Meteor. Soc.,60, 115–123.

——, K. W. Howard, D. L. Bartels, and D. M. Rodgers, 1986: Me-soscale convective complexes in the middle latitudes. Meso-scale Meteorology and Forecasting, P. S. Ray, Ed., Amer.Meteor. Soc., 390–413.

Moore, J. T., F. H. Glass, C. E. Graves, S. M. Rochette, and M. J.Singer, 2003: The environment of warm-season elevatedthunderstorms associated with heavy rainfall over the centralUnited States. Wea. Forecasting, 18, 861–878.

Orlanski, I., 1975: A rational subdivision of scales for atmosphericprocesses. Bull. Amer. Meteor. Soc., 56, 527–530.

Parker, M. D., and R. H. Johnson, 2000: Organizational modes ofmidlatitude mesoscale convective systems. Mon. Wea. Rev.,128, 3413–3436.

Rogash, J. A., and J. Racy, 2002: Some meteorological character-istics of significant tornado events occurring in proximity toflash flooding. Wea. Forecasting, 17, 155–159.

Schumacher, R. S., and R. H. Johnson, 2005: Organization andenvironmental properties of extreme-rain-producing meso-scale convective systems. Mon. Wea. Rev., 133, 961–976.

Smith, J. A., M. L. Baeck, Y. Zhang, and C. A. Doswell, 2001:Extreme rainfall and flooding from supercell thunderstorms.J. Hydrometeor., 2, 469–489.

Wallace, J. M., 1975: Diurnal variations in precipitation and thun-derstorm frequency over the conterminous United States.Mon. Wea. Rev., 103, 406–419.

Winkler, J. A., 1988: Climatological characteristics of summertimeextreme rainstorms in Minnesota. Ann. Assoc. Amer. Geogr.,78, 57–73.

——, B. R. Skeeter, and P. D. Yamamoto, 1988: Seasonal varia-tions in the diurnal characteristics of heavy hourly precipita-tion across the United States. Mon. Wea. Rev., 116, 1641–1658.

FEBRUARY 2006 S C H U M A C H E R A N D J O H N S O N 85