Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

VOLUME 23 JOURNAL OF HYDROMETEOROLOGY NOVEMBER 2022

Quantifying Flood Frequency Associated with Clustered Mesoscale Convective


Systems in the United States

HUANCUI HU,a ZHE FENG,a AND L. RUBY LEUNGa


a
Atmospheric Sciences and Global Change Division, Pacific Northwest National Laboratory, Richland, Washington

(Manuscript received 1 March 2022, in final form 1 August 2022)

ABSTRACT: Mesoscale convective systems (MCSs) that are clustered in time and space can have a broader impact on
flooding because they have larger area coverage than that of individual MCSs. The goal of this study is to understand the
flood likelihood associated with MCS clusters. To achieve this, floods in the Storm Events Database in April–August of
2007–17 are matched with clustered MCSs identified from a high-resolution MCS dataset and terrestrial conditions in a
land surface dataset over the central-eastern United States. Our analysis indicates that clustered MCSs preferentially oc-
curring in April–June are more effective at producing floods, which also last longer due to the greater rainfall per area and
wetter initial soil conditions and, hence, produce greater runoff per area than nonclustered MCSs. Similar increases of
flood occurrence with cluster-total rainfall size and wetter soils are also observed for each MCS cluster, especially for the
overlapping rainfall areas within each cluster. These areas receive rainfall from multiple MCSs that progressively wet the
soils and are therefore associated with higher flood likelihood. This study underscores the importance to understand clus-
tered MCSs to better understand flood risks and their future changes.
KEYWORDS: Flood events; Convective storms; Soil moisture

1. Introduction based quantification of floods in association with MCSs has


not been available until a recent work. By combining a newly
Mesoscale convective systems (MCSs) play an important
developed MCS tracking database with a storm event data-
hydrological role in the central United States because they
base, Hu et al. (2021a) found that MCSs account for most of
are responsible for 30%–70% of the warm season rainfall in
the floods during the warm season in the central and eastern
the region and they have been associated with severe floods
United States, particularly for the slow-rising floods and hy-
(Fritsch et al. 1986; Houze et al. 1990; Haberlie and Ashley
brid floods (flood episodes consisting of both flash and slow-
2019). These organized convections are often initiated near
rise flood events).
the foothills of the Rocky Mountains and grow while propa-
Notably, some MCSs can develop closely in time and space
gating eastward to the Great Plains (Laing and Fritsch 1997;
due to favorable environments (e.g., abundant moisture, per-
Houze 2004; Ashley et al. 2003). Most regions in the Great
sistent ascending mechanisms by frontal processes, mesoscale
Plains experience 4–5 MCSs each month in the warm season
vortices, cold pools, and moderate wind shear). One extreme
(Feng et al. 2019). Compared with other non-MCS rainfall
case of clustered MCSs occurred in May 2015, when more
produced by isolated convections and stratiform clouds, MCS
than 20 MCSs swept across the Texas–Oklahoma area (Fig. 1a),
rainfall is ∼7 times more intense (Hu et al. 2020a). Therefore,
producing total rainfall that can “cover the entire state of
MCS storms are one of the key meteorological phenomena
Texas 8 inches deep” (Martinez and Brunfeld 2015). Conse-
that can produce extreme precipitation, particularly in the
quently, this MCS cluster caused ∼45 million U.S. dollars of
central Great Plains (Kunkel et al. 2012; Stevenson and
flood damage in Houston alone (AP 2015) and substantial
Schumacher 2014). Consequently, higher surface runoff is a
losses in life and properties in surrounding areas including
common terrestrial response to the intense MCS rainfall and
Oklahoma and Louisiana due to flash and extensive floods
is intimately linked with flooding (Hu et al. 2020b). Case stud-
and even tornados associated with the clustered MCSs. This
ies of flooding associated with MCSs have been well docu-
case has been linked with the persistent and strengthened El
mented (Maddox et al. 1978; Bosart and Sanders 1981; Houze
Niño sea surface temperature pattern that deepened the sta-
et al. 1990; Schumacher and Johnson 2005, 2008), and flood
tionary trough west of Texas and enhanced the Great Plains
occurrence in the central United States that peaks in the
low-level jet and further attributed to the warming climate
warm season has been attributed to MCSs (Villarini 2016;
(Wang et al. 2015). Sharing similar favorable large-scale at-
Dougherty and Rasmussen 2019). However, the climatology-
mospheric dynamic and thermodynamic conditions, multiple
MCSs may develop and interact with each other through
modulation of their common large-scale environment to form
Supplemental information related to this paper is available at an MCS cluster. Such clustering of MCS events can collec-
the Journals Online website: https://doi.org/10.1175/JHM-D-22- tively affect areas well beyond that of each individual MCS,
0038.s1. leading to significant hydrologic impacts at larger scales than
the more localized effect of an individual MCS. Therefore,
Corresponding authors: Huancui Hu, huancui.hu@pnnl.gov; MCS clusters may contribute importantly to floods for two
L. Ruby Leung, ruby.leung@pnnl.gov reasons. First, local water budget is sensitive to precipitation

DOI: 10.1175/JHM-D-22-0038.1
Ó 2022 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright
Policy (www.ametsoc.org/PUBSReuseLicenses).
1685
1686 JOURNAL OF HYDROMETEOROLOGY VOLUME 23

FIG. 1. (a) Hovmöller diagram for MCS rainfall in May 2015 averaged over 288–408N and 1108–848W, with MCS clusters indicated by
blue shading and missing rainfall data indicated by gray shading. (b)–(d) Rainfall (mm) accumulated over the lifetime of MCSs A, B, and C.
(e)–(g) Surface runoff generated by MCSs A, B, and C. (h)–(j) Flood episodes associated with MCSs A, B, and C, with rainfall coverage area
indicated by gray shading. Flood episodes are categorized as slow-rising flood (blue circle), flash flood (orange star), and hybrid flood (green
triangle). Each row in (b)–(j) shares the color bar on the right.

input, and rainfall intensity can have a major influence on effectively moisten the soils to precondition for saturation–
surface runoff production (Vischel et al. 2009; Boone et al. excess flow and contribute to flooding when subsequent
2009; Best et al. 2015). Second, the occurrence of clustered MCSs from the same cluster pass over the same area. Such
MCSs exposes substantial areas under the influence of in- preconditioning of land surface has been known to play a cen-
tense rainfall. Because rainfall area is positively correlated tral role in flood frequency and magnitude (Berghuijs et al.
with MCS-related flooding occurrence (Hu et al. 2021a), the 2016; Ye et al. 2017). With a wet initial soil condition, flood
intense rainfall of MCSs with or without significant overlap- discharges can be very sensitive to even short-lived rainfall
ping area can possibly initiate floods over a more extended events, while in a dry condition substantial amounts of rainfall
area covered by clustered MCSs. However, studies of clus- can infiltrate into the soils and contribute to slower subsurface
tered MCSs were limited to cases or one season (Trier et al. processes instead of a rapid flood response (Zhu et al. 2018).
2006, 2014), while the occurrence and seasonality of clustered To better understand flood risk associated with the causative
MCSs and their associated flood risk are largely unknown. meteorological events and other flood generation processes
In addition to the wide coverage area characterized by in- over an extended area well beyond traditional catchment scale
tense rainfall, another interesting aspect associated with clus- has been one of the emerging perspectives to advance flood re-
tered MCSs is the overlapping of rainfall areas among MCS search (Merz et al. 2014; Kundzewicz et al. 2014). Floods and
events that can repeatedly affect the same area within a short weather systems are closely connected because weather sys-
time period. This is particularly the case for clustered MCSs tems cannot only substantially determine precipitation charac-
associated with long-lived mesoscale convective vortex gener- teristics such as intensity and area coverage, but also affect soil
ated by MCSs, which can induce organized secondary convec- moisture states through precipitation wetting and evaporative
tions within or in the proximity of the mesoscale convective drying by modulating cloud and precipitation processes and
vortex (Fritsch et al. 1994; Bartels and Maddox 1991; Trier surface energy balances, respectively. Both precipitation char-
et al. 2000; Trier and Davis 2002). The resultant successive acteristics and soil conditions are critical drivers of flooding
time–space coherent episodes of heavy rainfall occurring over and thus different types of weather systems are intimately re-
approximately the same location have been described as lated with the regional flood characteristics (Villarini 2016;
heavy rainfall “corridors” (Tuttle and Davis 2006; Trier et al. Saharia et al. 2017). This effort to associate flood risk with
2014) and may contribute more readily to flooding. This is causative weather events has been possible by linking flood
likely the case because earlier MCSs within the cluster can statistics with continental-to-global-scale precipitation datasets
NOVEMBER 2022 HU ET AL. 1687

and/or terrestrial features (Dougherty and Rasmussen 2019; 2. Materials and methods
Saharia et al. 2017). For MCS-related flooding, Hu et al.
a. MCS dataset
(2021a) examined the flood likelihood by linking the warm-
season flood occurrence documented by the National Centers Similar to Hu et al. (2021a), we use the high-resolution
for Environment Information (NCEI) Storm Events Database (4 km, hourly) MCS dataset (Feng 2019) developed by Feng
with a long-term MCS tracking dataset (Feng 2019). While the et al. (2019), which provides the track of each MCS occurring
work in Hu et al. (2021a) set the stage for our current study by between January 2004 and December 2017 over the United
demonstrating the strong connection between floods and MCS States. This MCS database is obtained by applying the Flexible
rainfall, our emphasis for this study is the flood risk associated Object Tracker (FLEXTRKR) algorithm (Feng et al. 2018)
with clustered MCSs. Our focus on clustered MCS events is to track MCSs defined by large cold cloud shields containing
largely motivated by their much broader area of impact beyond a radar-defined precipitation feature (PF, a contiguous area
individual MCSs, which can expose more regions to flooding with rain rate . 1 mm h21) with a major axis length . 100 km
solely due to their strong rainfall intensity. We are also and a convective feature containing radar reflectivity . 45 dBZ
interested in the role of land surface in affecting flooding at any vertical level that persists for longer than 6 h (Feng et al.
likelihood, particularly for regions experiencing repeated 2018, 2019). The cold cloud, precipitation, and radar informa-
wetting due to successive MCS passages. In addition to their tion are obtained from the NASA Global MergedIR satellite
high impact, a better understanding of clustered MCSs may dataset (Janowiak et al. 2001), Stage IV multisensory precipi-
also help reduce uncertainties in flood forecasting because tation dataset (Lin 2011) and 3D radar reflectivity mosaic
the favorable atmospheric environments that sustain clus- from the GridRad NEXRAD radar dataset (Bowman and
tered MCSs may be more predictable than those of individ- Homeyer 2017), respectively [see Feng et al. (2019) for more
ual MCSs (Trier et al. 2014), and the flood probability of technical details]. By tracking each MCS, this MCS dataset
provides important quantifications of MCS PF characteristics
precipitation conditioned on wet soils may be higher (Sharma
(e.g., lifetime, propagation, and convective and stratiform rain-
et al. 2018; Zhu et al. 2018; Wright et al. 2020), which is likely
fall separation) that are used in this study to explore the link-
the case for MCS clusters.
ages with flood frequency.
Therefore, in this study, we aim to extend the MCS–flood
We note that the above algorithm and thresholds for PF
linkage identified by Hu et al. (2021a) by quantifying the flood
size, radar reflectivity and durations used by FLEXTRKR are
frequency associated with MCS clusters. Our goal is to ex-
rooted in the current definition of MCS events with the goal
plore the relationship between flood occurrence and rainfall
to detect the largest form of deep convective storms. The
characteristics associated with MCS clusters, as well as the
MCS dataset in our study generally yields consistent climatol-
role of initial soil wetness on modulating flood responses. In
ogy of MCS in the United States from other studies (Geerts
this study, flood frequency quantification is mainly examined
1998; Pinto et al. 2015; Prein 2017; Haberlie and Ashley
in the context of MCS-related floods rather than all floods, as
2019), with an emphasis on the long-lived MCSs by requiring
we emphasize the differences between clustered and nonclus-
a minimum of 6-h lifetime. It is possible that a detected MCS
tered MCSs, while the flood frequency differences between
is embedded in a synoptic system (e.g., extratropical cyclone)
MCS and non-MCS events are discussed in Hu et al. (2021a). or demonstrates frontal features (Kunkel et al. 2012; Ralph
Our analysis is designed to address three questions: 1) What et al. 2011; Trier et al. 2014). Therefore, it is important to
is the flood likelihood associated with MCS clusters? 2) What note that the MCS dataset does not exclude events that might
are the differences of MCS characteristics and terrestrial con- be driven by synoptic systems other than mesoscale processes,
ditions for clustered and nonclustered MCSs, and how are but rather it aims to capture all the rainy events that match
flood occurrences modulated by these characteristics? 3) For the MCS characteristics.
each MCS cluster, what are the dominant factors for flood oc- It is also important to note that the rain rate threshold
currence and durations? value we use to define PFs (1 mm h21) seems moderate, but
To develop a joint view of MCS-related rainfall and land using a low-to-moderate threshold value is critical to capture
surface conditions on flood responses, we link flood occur- the complete life cycle of each MCS. Defining PF areas using
rence reported by the NCEI Storm Events Database with a a higher threshold value would certainly lead to smaller PF
high-resolution observation-based MCS dataset (Hu et al. areas at each time step, but it does not reduce the detected
2021a) and land surface conditions and processes provided number of MCS events substantially (see an example in
by a land surface simulation. Section 2 describes the three Fig. S1 in the online supplemental material for 2015). In other
datasets and the algorithm we used to identify MCS clusters, words, most of the MCSs are characterized by active precipi-
as well as how we attribute floods to the rainfall and soil tation areas with rain rate exceeding 5 mm h21, which lasts
conditions associated with MCS clusters. Section 3 presents for at least 6 h. Compared to using 1 mm h21 to define PFs,
the extreme case in May 2015 to illustrate the interactions using a higher threshold value could arbitrarily truncate the
between rainfall and terrestrial responses, while section 4 lifetime of an MCS by excluding time steps after its initiation
explores the climatology relationships between flood occur- or before its decay, causing an underestimation of the MCS
rence and MCS clusters. We test the sensitivity of the pa- duration and coverage area (Fig. S1). For this study, it is criti-
rameters determining MCS cluster in section 5 and conclude cal to capture the complete life cycle of each MCS to correctly
in section 6. identify the initial soil conditions at the time of MCS initiation.
1688 JOURNAL OF HYDROMETEOROLOGY VOLUME 23

Truncation of the MCS lifetime due to the use of a higher thickness from top to bottom) is used. The model is initialized
rain rate threshold can result in initial soil moisture state at 0000 UTC 1 March of each year using the soil states from
that has been preconditioned by earlier stages of the same the Noah simulation in the NLDAS archive for the same year
MCS with lighter rain rate instead of the pre-event soil and run for 6 months until 31 August, driven by the NLDAS
conditions. atmospheric forcing including precipitation, radiation, near-
surface air temperature, wind and humidity, and surface pres-
b. NCEI Storm Events Database
sure. The model reaches an equilibrium within the first month
The flood dataset we link with MCSs are the floods docu- (see an example in Fig. S2), so data from April to August in
mented by the NCEI Storm Events Database. The Storm 2007–17 are used in our analysis.
Events Database records the time and locations for significant This dataset has been validated by comparing observational-
meteorological events and related phenomena of great socie- based evapotranspiration estimation and other NLDAS model
tal impacts since 1950, compiled from National Weather Ser- outputs (Hu et al. 2020b). For evapotranspiration, our simula-
vice (NWS) and other sources including media and law tion reproduces the large east–west gradients in the central
enforcement (NWS 2018). Due to its emphasis on societal im- United States but overestimates the evapotranspiration, espe-
pacts, this dataset tends to weigh more heavily on events af- cially over the northern Great Plains relative to the MODIS
fecting urban areas or areas with dense population, while and FLUXNET estimated evapotranspiration. However, our
small events or large events with less socioeconomic impacts land surface data produce very similar patterns as seen in
might be underreported (Ashley and Ashley 2008; Barthold other land surface simulations archived by NLDAS-2 pro-
et al. 2015). However, comparisons with other streamflow duced by other land surface models, including evapotranspira-
based hydrologic flood datasets support the sufficient cover- tion, surface runoff, and soil moisture (see more details in Hu
age of floods both spatially and temporally by this Storm et al. 2020b). In this study, we use the soil moisture and surface
Events Database (Dougherty and Rasmussen 2019; Gourley runoff from our Noah-MP simulations at 4-km grid spacing to
et al. 2013). In addition, this database distinguishes “flash analyze the terrestrial conditions and responses to individual
floods” from other floods because of their rapid surge of MCSs and MCS clusters. Note that like most of the current
water and the shortage of time to mitigate that can cause land surface models, the Noah-MP model dataset does not ex-
significant fatality and property damages (NWS 2018). Other plicitly represent the terrain slopes and related lateral water
floods with high flow, overflow, or inundation that cause movement, which likely affect soil moisture and runoff. How-
damage are grouped as slow-rising floods (Dougherty and ever, datasets accounting for terrain effects with similar spatial
Rasmussen 2019). and temporal coverages are not currently available.
Our analysis of flood frequency and durations is based on d. Identify MCS clusters
flood episodes from the Storm Events Database, which can
encompass multiple flood events associated with the same We use the locations and sizes of MCS PFs throughout
synoptic meteorological system (NWS 2018). Therefore, in their lifetimes to identify clustered MCSs. We first identify
addition to flash and slow-rising flood episodes that embody MCS pairs that are close in time and space if their PF distance
only flash or slow-rising flood events, flood episodes involving is within a small time range. We then group MCS pairs having
a combination of flash and slow-rising flood events are catego- overlapping MCSs, and each group with at least three distinc-
tive MCSs is identified as an MCS cluster. The PF distances
rized as hybrid flood episodes (following Dougherty and
are calculated by Eq. (1) similar to the method used by Feng
Rasmussen 2019). From April to August in 2007–17, a total of
et al. (2015):
7106 flash flood episodes, 2842 slow-rising flood episodes, and
1238 hybrid flood episodes are collected east of 1108W in the D12 5 Dc12 2 L1 /2 2 L2 /2, (1)
United States, which corresponds to the spatial coverage of
the MCS dataset. Note that we exclude the data before 2007 where Dc12 is the distance between the centers of two MCSs
because flood occurrence is substantially lower than that after with PF1 and PF2, and L1 and L2 are the major axis lengths of
2007, presumably due to coverage biases (Hu et al. 2021a). PF1 and PF2, respectively.
These flood episodes are attributed to MCSs and MCS clus- If the MCS PFs have circular shapes, D12 represents the dis-
ters (explained in section 2e). tance between the edges of the MCS PFs. However, MCSs in
the United States are frequently organized to some form of
c. Land surface data
quasi-linear modes (Parker and Johnson 2000; Cui et al. 2021)
To account for the land surface processes that can be criti- and noncircular MCSs that occur close to each other in space
cal for flood generation, we use a set of high-resolution model and time may result in a negative distance [Eq. (1)].
output from the Noah-MP land surface model (Niu et al. To ensure the MCSs in a cluster are close in time and space,
2011). The simulation is configured to cover a similar spatial we use the thresholds for distance (dx) and time (dt) to be
area as the North American Land Data Assimilation System 0 and 4 h. Two circular PFs with D12 , 0 would suggest an
(NLDAS; 1/88 grids) but with a finer 4-km grid spacing to overlapping of PF areas while noncircular PFs, which may not
better account for the land surface heterogeneity (1/88-grid overlap, are at least in the vicinity of each other and may con-
forcing data bilinearly interpolated to the finer land surface model tribute to the same flooding at a basin outlet if they produce
grid). The default soil layer configurations (0.1, 0.3, 0.6, 1.0 m in rainfall in the same drainage area. And a 4-h time window
NOVEMBER 2022 HU ET AL. 1689

allows some time lag between MCSs to capture events that indicates repeated wetting of the same region within a short
have similar pathways but a few hours apart. Using dx 5 0 time period before substantial adjustments of soil moisture
and dt 5 4 h, a total of 291 MCS clusters are identified during such as evapotranspiration and subsurface drainage so the
April–August 2007–17, with a total of 1262 MCSs considered terrestrial conditions are more favorable for flooding. On
in clustered form out of a total of 3447 MCSs (data before one hand, we use the overlapping areas as one of the PF
2007 are not used due to some issues with the flood data; see characteristics to examine its relationship with the overall
section 2b). These 291 MCS clusters detected with dx 5 0 and flood frequency associated with each cluster. On the other
dt 5 4 h are used for the majority of analysis of this study. hand, we quantify the likelihood of flood occurrence as a
Despite using this simple method to identify MCS clusters function of the overlapping areas by further attributing the
based on major axis distances, we confirm that pixel-level cluster-related floods to their overlapping areas if the floods
overlapping between PFs occurs among the MCSs within occur within the 20-km range of the PF overlap areas,
every cluster using this set of parameters. However, we note following the same general methodology. Although storms
that the parameters dx and dt are somewhat arbitrarily de- can co-contribute to basin saturation and potentially flood-
fined to constrain the spatial and temporal distances between ing due to lateral movement of water even without spatial
the PFs of the MCS pairs that define the MCS clusters. As ex- overlapping of storms, our ability to analyze such effect is
pected, the number of MCS clusters are subject to the differ- limited because such lateral runoff responses are generally
ent values of dx and dt. To address this, we test the sensitivity not represented in most current land surface models (Fan
of MCS clusters and their related floods to these parameters et al. 2019) or land surface datasets (section 2c). More ad-
in section 5. vanced hydrologic models that account for lateral routing
In addition, merging and splitting of storms are taken into would be necessary for such purposes (Getirana et al. 2014;
account in the MCS tracking (Feng et al. 2019). While it is Gochis et al. 2018), which are beyond the scope of this
possible that two MCSs merging into a larger MCS (or a large study. Currently, we focus on the land surface responses to
MCS splitting into smaller MCSs) may contribute to the same storms with overlapping PF coverage areas, which happen
MCS cluster, each of them must meet the criteria of MCS fea- in all MCS clusters based on our definition.
tures defined by the FLEXTRKR algorithm before merging
f. Rainfall and terrestrial characteristics of MCSs and
(or after splitting). For example, we visually checked the
MCS clusters
MCS life cycle characteristics for 2015 and found that out of
149 clustered MCSs, 36 split from their “parent” MCSs while To explore the factors controlling or modulating the flood
12 merged into preexisting MCSs. Each of these child and occurrence and potentially durations associated with each
parent systems maintain their sizes and lifetimes as proper MCS or MCS cluster, we combine the PF characteristics from
MCSs. They often occur within clusters consisting of more the MCS dataset (section 2a) and the terrestrial responses
than 5 MCSs and thus using a different approach to treat split- from the land surface dataset (section 2c). In addition to the
ting/merging is unlikely to affect the cluster occurrence. instantaneous PF characteristics, event-total PF characteris-
Therefore, merging or splitting MCSs are considered as part tics are particularly important for flood frequency (Hu et al.
of an MCS cluster as they can collectively have significant hy- 2021a) and therefore are of our major focus. Accordingly, the
drologic impacts regardless of their origin or evolution. soil moisture conditions and runoff generation over the event-
total PF coverage areas are extracted and used for our analy-
e. Attribute floods to MCSs and MCS clusters
sis (section 4). The soil moisture conditions are quantified in
The flood episodes of different types in April–August two ways: soil moisture anomaly is calculated by comparing
(2007–17) are then attributed to MCS events and MCS clus- the soil moisture with the climatological value of the same
ters according to their time and location of occurrence day in the seasonal cycle, and the percentile of soil moisture
(Hu et al. 2021a). Flood episodes are attributed to an MCS if with respect to the historical samples on the same day of
floods occur within its total PF coverage area or less than the seasonal cycle over the same coverage area. To contrast the
20 km apart during the lifetime of the MCS. We note that the differences of the overlapping areas with the rest (section 4c),
20-km threshold is tested by Hu et al. (2021a) and selected the PF characteristics and terrestrial variables are also ex-
due to its optimal effect to distinguish floods associated with tracted over the overlapping areas. For most of our analysis,
MCS and non-MCS storms but also allow for the downstream we use the initial soil moisture states immediately before the
effects of rainstorms on flooding. The same is applied to MCS initiation of an MCS or a MCS cluster to study their relation-
clusters by attributing flood episodes to an MCS cluster if ship with flood characteristics. As the initial soil moisture
they occur within or less than 20 km from the total PF cover- anomalies capture the soil states before they are modulated
age area over the lifetime of the MCS cluster. by the MCS or MCS clusters, they reflect the net effect of
Due to their potential importance to flooding, we pay par- water gaining (e.g., snowmelt, non-MCS precipitation, earlier-
ticular attention to the overlapping PF areas among MCS season MCS events) and losing processes (e.g., evapotranspi-
clusters, defined as PF areas with at least 2 MCS passages in ration, runoffs) unrelated to the MCS or MCS clusters. To
the same cluster. As mentioned earlier, all the MCS clusters illustrate the role of successive wetting by clustered MCSs, we
have overlapping PF areas by the method described in section 2d. also compare the soil moisture conditions after the passage of
Such overlapping of PF areas among MCSs in the same cluster MCSs over the overlapping areas in section 4c.
1690 JOURNAL OF HYDROMETEOROLOGY VOLUME 23

FIG. 2. (a)–(f) MCS tracks within each MCS cluster identified in May 2015 and
(g)–(l) flood episodes associated with each MCS cluster. Each MCS is tracked with a
unique ID in the MCS dataset and color shaded in (a)–(f). Gray shadings in (g)–(l) in-
dicate the total PF coverage area of each MCS cluster.
NOVEMBER 2022 HU ET AL. 1691

TABLE 1. List of flood episodes associated with MCSs A, B, and C in section 3.

Episode Flood type Starting time End time State Related MCS
1 95612 Slow-rising 1800 UTC 5 May 2015 0500 UTC 1 Jun 2015 Oklahoma A, B, C
2 95615 Slow-rising 0600 UTC 6 May 2015 0500 UTC 1 Jun 2015 Texas A, B, C
3 94168 Slow-rising 0300 UTC 7 May 2015 0000 UTC 11 May 2015 Nebraska B
4 96627 Hybrid 1900 UTC 8 May 2015 1100 UTC 11 May 2015 Arkansas A, B, C
5 94239 Flash 0700 UTC 9 May 2015 2200 UTC 9 May 2015 Oklahoma A
6 95504 Hybrid 1400 UTC 9 May 2015 2100 UTC 11 May 2015 Texas A, B, C
7 94240 Flash 2000 UTC 9 May 2015 2300 UTC 9 May 2015 Arkansas A
8 97778 Slow-rising 2100 UTC 10 May 2015 0500 UTC 1 Jun 2015 Arkansas B, C
9 96013 Hybrid 2000 UTC 10 May 2015 0000 UTC 11 May 2015 South Dakota B
10 94244 Flash 0600 UTC 11 May 2015 0900 UTC 11 May 2015 Arkansas C
11 94243 Flash 0600 UTC 11 May 2015 1200 UTC 11 May 2015 Texas C
12 94972 Slow-rising 0700 UTC 11 May 2015 1500 UTC 11 May 2015 Missouri C
13 95618 Flash 1100 UTC 11 May 2015 1600 UTC 11 May 2015 Texas C
14 97954 Slow-rising 1300 UTC 11 May 2015 2200 UTC 12 May 2015 Illinois C

3. MCS clusters in May 2015 that the Red River flooding observed in Texas and Oklahoma
and other rivers in Arkansas remained above the flood stage
We use the extreme case of May 2015 as an example to
at the end of May (episodes 95612, 95615, 97778; all ended
demonstrate the detection of MCS clusters and the impact of
abruptly on 1 June, probably due to the flood entry segre-
repeated wetting due to overlapping PF. Using the method
gated by month), associated with the excessive rainfall over
described in section 2d, 6 MCS clusters are identified in May
similar area by the subsequent MCSs and MCS clusters.
2015 (shaded in blue in Fig. 1a). The first MCS cluster occur-
For the entire cluster that includes 9 MCSs (9–13 May), the
ring during 4–6 May mainly produced rainfall over northern
overlapping PF areas is 1.678 3 106 km2 while the total PF
Texas, while the following 5 MCS clusters were shifted a bit
coverage area is 3.759 3 106 km2, resulting in an overlapping
eastward and passed over eastern Texas and Oklahoma
fraction of 44.6% (Fig. 2b). Twenty-three of the 26 flood epi-
(MCS tracks for each cluster are shown in Fig. 2). The second sodes (4 slow-rising, 13 flash, and 6 hybrid floods) are found
MCS cluster during 9–13 May is particularly interesting, not in the proximity of the overlapping areas (Figs. 2b,h). Consis-
only because this MCS cluster consisted of 9 MCSs, but 3 of tent with the increase of runoff generation ratio associated
them followed very similar paths (referred to as MCSs A, B, with MCSs A, B, and C, the initial soil moisture conditions
C) over the Texas–Oklahoma region within 50 h (Figs. 1b–d). over the overlapping areas also become wetter following the
MCS A (9–10 May) strongly affected Oklahoma, but areas passage of the sequential of MCSs, changing from 48.8% to
with accumulated rainfall exceeding 50 mm also extended to 91.7%. Due to uncertainties in the land surface data, the run-
Arkansas and northeast Texas (Fig. 1b). Analysis of the land off generation ratios and soil moisture percentiles should only
surface data indicates that ∼13% of rainfall from this MCS is be used as indications of terrestrial dynamics rather than ex-
estimated to contribute to runoff (Fig. 1e), while the rest en- act quantifications of the surface water budgets. The continu-
ters the soils and can later contribute to evapotranspiration ous moistening of the soils illustrates that MCSs sequentially
on a longer time scale (Hu et al. 2021b). MCS A induced two passing over similar areas may contribute to/induce slow-rising
flash floods (episodes 94239 and 94240) and two hybrid floods floods by producing a higher fraction of saturation–excess
that lasted until 11 May (Table 1). It also contributed to the surface runoff that can continuously converge toward the
two slow-rising floods on the Red River in southeast Okla- basin outlet and extend the flood durations. Meanwhile, Fig. 2
homa and northeast Texas that were induced by the first MCS also indicates a wide span of flood locations that can possibly
cluster on 4–6 May (Figs. 1h, 2a and Table 1). The next MCS occur over each cluster’s PF coverage area, including flash
B (10 May) was weaker in terms of accumulated rainfall but a floods that were induced by MCSs in clustered form that pro-
slightly higher fraction (14.3%) of its rainfall became runoff duced rainfall intensity exceeding the infiltration rate of the
(Figs. 1c,f). It contributed to multiple preexisting slow-rising land surface.
and hybrid floods, but also induced a slow-rising flood in
Arkansas and a hybrid flood in South Dakota (Fig. 1i) due to 4. Establishing the links between floods and
the preexisting anomalously wet soil conditions at these loca- MCS clusters
tions (Fig. S3). On the next day, MCS C (11 May) struck the
same area by bringing substantial amounts of rainfall over
a. Seasonal mean of flood frequency associated with
MCS clusters
eastern Texas and Arkansas, with 16.5% of its rainfall turning
into runoff (Figs. 1d,g). Consequently, it enhanced the slow- Hu et al. (2021a) separately related flood frequency with
rising and hybrid floods in Oklahoma, Texas, and Arkansas, MCS and non-MCS storms and found that MCSs account for
and produced flash floods in these states and slow-rising most of the warm-season floods in the United States east of
floods in Missouri and Illinois (Fig. 1j and Table 1). Notice 1108W, especially for slow-rising and hybrid floods. But flash
1692 JOURNAL OF HYDROMETEOROLOGY VOLUME 23

FIG. 3. (a) Flood episode frequency in each month and flood attribution to MCSs and clustered MCSs aggregated
over the area to the east of 1108W. (b) Monthly frequency of MCSs and clustered MCSs. (c)–(h) Spatial distributions
of MCS-related flood frequency of each flood type and fraction of each flood type associated with clustered MCSs.
Panels (c), (e), and (g) share the same color bar at the bottom left, and (d), (f), and (h) share the same color bar at the
bottom right.
NOVEMBER 2022 HU ET AL. 1693

FIG. 4. (a)–(e) MCS frequency, (f)–(j) clustered MCS frequency, and (k)–(o) the fractions of clustered MCS to all MCS in each month.

floods in July and August are also frequently associated with The larger fractions of flood episodes associated with MCS
non-MCS convections (Fig. 3a) that occur often in the Rocky clusters in early months are consistent with the higher occur-
Mountains and Appalachian Mountains. As noted, we focus rence of MCS clusters in these months (Fig. 3b). Using the
on quantifying flood frequency associated with MCS clusters dx 5 0 and dt 5 4 h criteria, the fraction of clustered MCSs is
in the context of MCS-related floods rather than all floods in 57.2% in April, ∼40% in May and June, and 22.7% and
the following analysis. 30.7% in July and August. As a result, 23–29 MCSs in clus-
Using 11 years of data in 2007–17, we find that MCS-related tered form occur in each month from April–June, although
flooding is more likely to be associated with MCS clusters in fewer MCSs are observed in April and May. This can be par-
early warm season (April–June), regardless of flood types tially attributed to the fact that MCSs in the earlier warm sea-
(cf. black hatched and red stippled areas in Fig. 3a). For slow- son are subject to stronger synoptic forcing (Song et al. 2019),
rising floods that occur more frequently in April–June than which are more likely to support the development of multiple
later months over the northern Great Plains (north of 358N), spatially and/or temporally connected MCSs. MCS PF areas
more than 75% of MCS-related flooding is associated with are also larger in the early months, resulting in smaller PF dis-
MCS clusters in these months (83.3%, 85.0%, and 75.2% re- tances between MCSs. We also find that the frequency of clus-
spectively for April, May, and June; Figs. 3c, Figs. S4a–c, tered MCSs shows a northward migration from April to
S5a–c). In July and August, the fraction is reduced to 44.1% August (Fig. 6), similar to the northward migration of all
and 48.5%, resulting in less than 10 flood episodes per month. MCSs through the warm season. However, the fraction of
Similar fractions and decreasing trend with months during the MCSs associated with MCS clusters shows higher values south
warm season are found for hybrid floods observed in of ∼378N throughout the warm season (Figs. 4k–o). This ex-
378–438N (Fig. 3h and Figs. S5k–o). plains the greater fraction of MCS-related flash floods in the
For flash floods, most MCS-related floods occur between southern states (Figs. S5f–j), while no such preference is no-
1058 and 908W (Fig. 3e), while flash floods near the Rocky ticeable for slow-rising or hybrid floods that mostly occur in
Mountains and Appalachian Mountains attributable to non- the northern Great Plains (Figs. S5a–e,k–o). This southern
MCS thunderstorms are not included in Fig. 3e. In April and preference for clustered MCSs might be related to its proxim-
May when flash floods are often observed in the southern ity to the Gulf of Mexico that can provide abundant moisture
Great Plains, greater fractions of MCS-related floods are asso- supply for the redevelopment of MCSs in a short period of
ciated with MCS clusters: 79.3% of in April and 58.9% in time. More specific investigations of the synoptic environ-
May (Fig. 3a and Figs. S5f–j). But for June–August when flash ments in the future would be necessary in order to fully ex-
floods occur more frequently, the fraction is reduced to 56. plain the southern preference of MCS clusters. Notice that
5%, 31.0%, and 43.3%, respectively. While the flash floods the fractions of floods associated with clustered MCSs to total
shift northward in these months, we find that clustered MCSs MCS-related floods (Figs. 3d,f,h) always exceed the fractions
tend to favor flash floods in the southern states (Figs. S4f–j, of clustered MCSs to total MCSs (Figs. 4k–o), regardless of
S5f–j). flood types. This indicates that clustered MCSs are more
1694 JOURNAL OF HYDROMETEOROLOGY VOLUME 23

August with mostly flash floods because of the greater MCS


rainfall intensity in these months.

b. Differences between clustered and nonclustered MCSs


and relationship with flood frequency
To understand the factors contributing to the greater flood
occurrence associated with clustered MCSs relative to non-
clustered MCSs (Fig. 5c), we now examine their differences in
both MCS rainfall characteristics and land surface conditions.
For the MCS PF characteristics, MCSs that belong to clusters
have longer durations and larger instantaneous rainfall areas
compared to nonclustered MCSs, while propagation speeds
and instantaneous rain rates are not significantly different
(Figs. 6a–d; means and significance level indicated by Table S2).
As a result, each clustered MCS has significantly larger event-
total rainfall coverage area and greater total rainfall volume
than individual MCSs that are not clustered (Figs. 6e,f). The
greater rainfall areas and volumes per MCS in clustered MCSs
are likely contributed by favorable synoptic conditions that sup-
port sequential development of MCSs over consecutive days
[see case studies by Trier et al. (2006, 2014)]. In addition to the
fact that clustered MCSs are generally stronger in terms of PF
coverage area and rainfall volume, their preferential association
with flooding is also supported by the terrestrial conditions. We
find that clustered MCSs generally occur over wetter initial soil
moisture conditions (Figs. 6g,h) in the late summer months and
becomes especially important for generating floods as soils are
generally drier in those months. Such wetter initial soil moisture
FIG. 5. (a) MCS cluster frequency in each month color coded by
conditions can be a result of repeated wetting that we see in the
the dominant flood type associated with each MCS cluster. (b) Aver-
aged flood occurrence of each type associated with flood-producing May 2015 case (section 3), but it is also possible that MCS clus-
MCS clusters in each month. (c) Flood occurrence (left y axis) and du- ters are more likely to occur over a wetter surface as a result of
rations (right y axis) associated with nonclustered and clustered land–atmosphere interactions (Trier et al. 2008). However, the
MCSs, with boxplots showing durations shaded by orange. causes of these MCS clusters are beyond the scope of this study
but could be a very interesting topic for future research. Due to
effective in producing flooding. This is also supported by Fig. 5c the differences in rainfall characteristics and land surface condi-
showing significantly higher frequency of slow-rising and hybrid tions, each MCS in clustered form on average can produce sub-
floods induced by a clustered MCS compared with MCSs that stantially greater runoff than a nonclustered MCS (Fig. 6i).
are not clustered. Clustered MCSs also tend to produce floods Because not all MCSs are associated with flooding, it is also
with significantly longer durations. The factors that can contrib- important for us to understand the differences in flood-
ute to the more effective and long-lasting floods associated with producing MCSs and the rest that do not produce floods. Due
clustered MCSs is examined in section 4b. to the favorable conditions for flooding associated with clus-
On average, about 4–7 MCS clusters (each consisting of tered MCSs, we find that ∼64% clustered MCSs are associ-
4.3 MCSs on average) occur in each month during the warm ated with floods (807 out of 1262 clustered MCSs) while
season, with a peak frequency in June (Fig. 5a). Grouping the ∼50% of nonclustered MCSs are associated with floods (1102
MCS clusters by the dominant flood type associated with each out of 2185 nonclustered MCSs). Note that this fraction is dif-
MCS cluster, about 2.5 MCS clusters cause slow-rising flood ferent from the flood-producing MCS clusters (Fig. 5a) be-
in April–June and 2–4 MCS clusters cause flash floods in each cause each MCS belonging to an MCS cluster is treated
month, with a peak in June. Furthermore, in each month, less individually rather than treating each MCS cluster as an en-
than one MCS cluster is not associated with any type of flood- tity. Comparing MCSs that do not produce floods, perhaps
ing so there is a high likelihood of flooding when an MCS the most apparent difference is the PF areas in both clustered
cluster occurs. Notably, only 8 MCS clusters (out of 291 in and nonclustered MCSs (Figs. 7c,e). Significantly faster prop-
total) are not associated with flooding. For the MCS clusters agations are also observed in clustered MCSs for flood-
that lead to flooding, each MCS cluster in April can lead to producing ones but not seen in nonclustered MCSs (Fig. 7b).
24 flood events, with ∼12 slow-rising floods, 9 flash floods, Accordingly, flood-producing MCSs are characterized by signifi-
and 3 hybrid floods (Fig. 5b). The flood number decreases to cantly greater PF coverage areas and event-total rainfall volumes
∼18 in May and June with comparable number of slow-rising than nonflood producing ones (Figs. 7e,f). Also significantly dif-
and flash floods, and further reduces to ∼10 floods in July and ferent is the initial soil wetness that is supported by the wetter
NOVEMBER 2022 HU ET AL. 1695

FIG. 6. Boxplots of different characteristics for MCSs that are not clustered and clustered grouped by early warm-
season (April–June) and later warm season (July–August). Asterisks indicate where differences are significant (.99%
confidence level using the Kolmogorov–Smirnov test; also see Table S1).

initial soil moisture for flood-producing MCSs, which exists in MCS that is clustered. Similar increasing relationships are
both clustered and nonclustered MCSs (Figs. 7g,h). As a result, found for the total PF rain volume (Fig. 8b), which exceeds
flood-producing MCSs are characterized by substantially greater the rate of increase in the event-total PF coverage area thus
runoffs (Fig. 7i). resulting in a greater rainfall per area with increasing flood
The critical roles of MCS PF characteristics and wetter occurrence (Fig. 8c). For the land surface conditions, we
initial soils for floods are further supported by the relation- also find a clear increasing trend of flood occurrence for
ship between these factors and flood occurrence (Fig. 8). In MCSs occurring over wetter initial soils (Fig. 8d). As a con-
Fig. 8, we show the variations of MCSs’ rainfall and terres- sequence of the joint role of greater rainfall per area and
trial characteristics with slow-rising and flash flood occur- wetter soils, MCSs that produce greater runoff and also
rence separately in each panel, while the results of hybrid greater runoff generation ratios tend to result in more floods
floods are not included as hybrid floods have limited occur- (Figs. 8e,f). Such evident relationship between greater run-
rence and they display similar relationships associated with off and flooding likelihood is consistent with precipitation
slow-rising and flash floods. Flood occurrence is shown in excess as the dominant flood generating mechanism over
both percentile and actual number in the x axis of each the central United States (Berghuijs et al. 2016). The impor-
panel. This analysis further confirmed the predominant role tant roles of PF characteristics (dominated by PF coverage
of rainfall area on flood frequency identified by Hu et al. areas) and initial soil wetness are respectively tested by fur-
(2021a), with a larger MCS PF area favoring higher flood oc- ther subgrouping MCSs with similar initial soil conditions
currence. We see a similar increasing trend of flood fre- and those with similar PF coverage areas, and the relation-
quency with PF areas for each average MCS that is ships we see in Fig. 8 still hold (Figs. S6, S7), indicating that
nonclustered and clustered (Figs. 8a), while consistent with they can each play a distinctive role on flood frequency be-
Fig. 8e, the PF areas are generally larger for an average sides their joint effect.
1696 JOURNAL OF HYDROMETEOROLOGY VOLUME 23

FIG. 7. Boxplots of different characteristics for MCSs that are flood-producing (light blue and brown boxes) and
not flood-producing (gray boxes) grouped by nonclustered and clustered MCSs. Asterisks indicate where differences
are significant (.99% confidence level using the Kolmogorov–Smirnov test; also see Table S2).

c. Relationship between cluster-total characteristics and areas and thus increases rainfall per area over the whole clus-
floods and the role of overlapping PFs ter-total PF coverage areas (Fig. 9d).
Analysis of initial soil moisture conditions over the cluster-
With each MCS in clustered form characterized with stron-
total PF coverage area using percentiles also indicates an in-
ger PF features (PF area and rainfall volumes), integrating
creasing trend of floods with initial soil wetness (Fig. 10a) as we
over an entire MCS cluster that contains at least 3 MCSs can see in individual MCSs (Fig. 8d), reiterating the important role
be of more significant impact. Indeed, an MCS cluster is very of soil wetness to floods. Note that further testing by subgroup-
likely to produce severe floods as described earlier in Fig. 6a. ing MCS clusters with similar cluster-total PF coverage areas
Therefore, we integrate the individual MCS PF characteristics confirms such relationship (not shown). Subsequently, we see
within each MCS cluster treated as a unit (cluster-total PF an increasing trend of flood occurrence with MCS clusters oc-
characteristics) and investigate their relationship with floods. curring over wetter initial soils that generate greater amounts of
By integrating the PF features over the entire clusters, we find surface runoff per area and greater runoff ratios (Figs. 10b,c).
a similar and evident increase of flood occurrence with the We also find a more important role of terrestrial responses
cluster-total PF coverage areas and rain volumes (Figs. 9a,b). in flood durations in contrast to PF characteristics. In general,
Interestingly, MCS clusters associated with more floods are we find that floods associated with MCS clusters tend to last
also characterized with greater overlapping PF ratios (calcu- longer if more surface runoff generation is present (Figs. 10f,g),
lated as the ratio of PF areas with at least 2 MCS PF passages while such a relationship is not found for MCS PF areas or rain
to cluster-total PF coverage areas) among the MCSs in each volumes (not shown). Note that flood durations refer to the
cluster (Fig. 9c), with overlapping PF illustrated by the exam- mean durations of flood episodes associated with each MCS
ples we see in section 3. Such increase of overlapping areas cluster, and MCS clusters binned by flood durations in Figs. 9f,g
thus results in a greater rainfall per area over the overlapping only account for flood durations but do not jointly account for
NOVEMBER 2022 HU ET AL. 1697

FIG. 8. Boxplots of event-total characteristics for MCSs that are nonclustered and clustered, binned by the occur-
rence of each flood type they are associated with. The different percentiles of flooding occurrence are listed at the bot-
tom of each column, color coded by flood types. See median values and correlations in Table S3.

flood occurrence. MCS clusters causing one or two floods but areas, overlapping PF areas that receive rainfall from more
with long durations, presumably due to favorable local terres- than one MCS are characterized by a much greater rainfall
trial conditions, can fall into the 90%–100% duration bin, but per area (Fig. 11a). With a similar initial soil wetness, the
these clusters are not necessarily the strongest ones leading to overlapping areas end up producing greater runoff per area
more flood occurrence. This is the case for the relatively low with a much wetter soil condition after the passage of the entire
surface runoff ratios associated with the moderate MCS clusters, cluster than the cluster-total PF coverage areas (Figs. 11a,b).
leading to 1–2 slow-rising floods with .491.5 h of durations And such wetting occurs progressively throughout each cluster
(Figs. 9f,g). We also note that initial soil wetness can be impor- (Fig. 11c), providing a continuously favorable condition for
tant to other aspects of flood severity (e.g., inundation area) flooding for the subsequent MCSs in the same cluster, while a
that are not available from the Storm Events Database. similar wetting trend is not observed for the nonoverlapping
To better understand the effect of the overlapping PF areas areas (not shown).
on flooding indicated by Fig. 9c and the example in May 2015
(section 3) with successive MCS passages over the same area, 5. Sensitivity to MCS clusters definition
we quantify the flood frequency that occurs in the proximity
of the overlapping areas and analyze their rainfall and terres- As noted in section 2d, the detection of MCS clusters is
trial features over those areas. We find that 46.6% of floods subject to the threshold parameters (dx and dt) that we used
associated with MCS clusters occurs within or close to the to identify MCS clusters. In this section, we test the sensitivity
overlapping PF areas (43.4% for slow-rising, 45.8% for flash, of our results to the dx and dt parameters we use to identify
and 57.7% for hybrid floods). These fractions well exceed the MCS clusters. To achieve this, we perform two sets of analy-
fraction of overlapping areas to total PF coverage areas ses by varying the dx and dt values, respectively, and compar-
(20.9% on average; see variations in Fig. 9c), which indicates ing how the MCS characteristics and flood occurrences
a higher likelihood of floods occurring over overlapping areas. change with these parameters.
Such higher flood occurrence is also supported by Fig. 10d As expected, relaxing the dx or dt thresholds lead to more
and is closely related with the repeated wetting of clustered MCSs in clusters while restraining dx and dt to smaller values
MCSs. Compared with the entire cluster-total PF coverage would identify fewer MCSs in clusters (black lines in Figs. S8a,f).
1698 JOURNAL OF HYDROMETEOROLOGY VOLUME 23

FIG. 9. Boxplots of cluster-total precipitation feature characteristics for MCS clusters, binned
by the occurrences of each flood type they are associated with. The different percentiles of flood-
ing occurrence are listed at the bottom of each column, color coded by flood types. See median
values and correlations in Table S4.

The number of MCS clusters also increases by relaxing dt/dx, 6. Discussion and conclusions
but at a slower rate (red lines in Figs. S8a,f). This indicates
In this study, we extend the MCS–flood linkage established
more MCSs in each cluster with increasing dt/dx. However,
by Hu et al. (2021a) by focusing on the relationship between
we find that the monthly distribution of clustered MCSs is al-
flood frequency and MCS clusters, which can have a much
tered by the dt/dx changes (Figs. S8b,g). We see an evident
larger impact than each individual MCS and tend to produce
increase of clustered MCSs being identified in June with in-
more extreme floods. Importantly, we incorporate terrestrial
creasing dt/dx, which shifts the peak month of cluster MCSs
conditions in our analysis in addition to MCS PF characteris-
from April with the smallest dt/dx thresholds to June with the
tics, which are critical for flood generation but not discussed
largest dt/dx (Figs. S8b,g). This shift of peak month to June,
in Hu et al. (2021a). By linking floods reported by the NCEI
which is slightly more evident with the dx changes (Fig. S8g),
Storm Event Database with MCS clusters in the warm season
is related to the generally smaller MCS PF areas but higher
that occurred in the central-eastern United States, we quan-
MCS frequency in June (Figs. 3a, 8c). As a result, we see a
tify the flood occurrence associated with MCS clusters and ex-
more noticeable shift of the MCS characteristics of clustered
amine how the flood occurrence and duration changes with
MCSs toward the characteristics of June MCSs with increas-
the different MCS and terrestrial characteristics. In particular,
ing dx, which tend to have slower propagation and smaller
we examine the role of overlapping PF areas within each
PF rainfall volume (Figs. S8c,d,h,i). Consistent with the
MCS cluster that can experience sequential rainfall. We sum-
monthly changes of clustered MCS with dt/dx, more flood ep-
marize the key findings as below:
isodes in June are found to be associated with MCS clusters
with greater dt/dx values, while flood frequency associated with • MCS clusters occur more frequently in the earlier warm
MCS clusters peaks in April with smaller dt/dx (Figs. S8e,j). season (April–June) and over the southern states (Figs. 3b
However, the differences between nonclustered and clustered and 4). Accordingly, floods in early warm season are more
MCSs in Fig. 6 and the relationship between flood occurrence likely to be associated with MCS clusters, while the south-
and cluster-total characteristics shown in Fig. 9 still hold by ern preference for floods is less evident (Fig. 3).
changing the dt/dx thresholds (see examples in Figs. S9–14). • Each flood-producing MCS in clustered form lasts signifi-
Therefore, we conclude that our results showing the different cantly longer and can lead to more floods than MCSs that
features of clustered MCSs, particularly the greater PF rainfall are not clustered (Fig. 7). Clustered MCSs are more effective
per volume and greater runoff generation, and their relation- in producing floods due to their significantly greater rainfall
ship with flood occurrence are not sensitive to the dt and dx volumes as well as wetter initial soil conditions that cause
parameters we use to define MCS clusters. greater surface runoff than nonclustered MCSs (Fig. 8).
NOVEMBER 2022 HU ET AL. 1699

FIG. 10. Boxplots of cluster-total terrestrial characteristics for MCS clusters, binned by the
(a)–(d) occurrences and (e)–(h) durations (orange shaded) of each flood type they are associated
with. The different percentiles of flooding occurrence or duration (h) are listed at the bottom of
each column, color coded by flood types. See median values and correlations in Table S4.

• Integrating over the cluster, flood occurrence also increases greater runoffs can lead to floods with longer durations
with cluster-total PF coverage areas and initial soil wetness, (Fig. 10).
resulting in greater runoff ratios (Figs. 9 and 10).
In this study we have only examined the relationship between
• Floods are more likely to occur around the overlapping PF
MCS clusters and floods in terms of the flood occurrence and
areas in each MCS cluster, due to the successive rainfall
durations. This is because other metrics that may better quan-
from multiple MCSs over the overlapping areas and thus a tify flood severity (e.g., inundation area and depth) are not
continuous moistening of the soils, creating a more flood- available from the NCEI Storm Events Database. However, we
prone condition (Fig. 11). use a high-resolution land surface dataset to account for the
• Terrestrial runoff responses are not only important for land surface conditions and responses to the different MCS
flood occurrence associated with each MCS cluster, but events. We confirm an important role of initial soil moisture for
also to flood durations because MCS clusters producing flood occurrence related to individual MCSs as well as MCS
1700 JOURNAL OF HYDROMETEOROLOGY VOLUME 23

FIG. 11. Boxplots showing (a) the differences between the PF coverage area-averaged and overlapping area-averaged rainfall per area
and surface runoff per area and (b) soil moisture conditions before and after the entire MCS clusters. (c) The inner-cluster changes of soil
moisture states over the overlapping PF areas grouped by the order of occurrence (e.g., first 1/3 indicates MCSs that occur in the earliest
third of the cluster lifetime).

clusters. By incorporating the land surface conditions and exam- associated flooding risks can be important questions to ex-
ining the different aspects of PF characteristics of MCSs, we are plore. Second, MCS clusters producing significant amounts of
able to decompose the different factors contributing to flood oc- rainfall within consecutive days might have a distinctive effect
currence and durations. Building upon the close relationship be- on water storage in the terrestrial system compared to nonclus-
tween flooding and MCSs demonstrated by Hu et al. (2021a), tered MCSs and non-MCS storms. On one hand, a larger contri-
we further examined the links between floods and MCSs that
bution to surface runoff suggests a limited role of rainfall from
are clustered and not clustered. Because clustered MCSs tend
clustered MCSs in providing soil moisture and contributing to
to occur in more favorable meteorological conditions and can
ecosystem productivity, which would benefit from intermittent
produce substantially more precipitation than individual MCS
events, the link we establish between flooding and MCS clusters and moderate rainfall throughout the growing season. On the
can potentially help improve the prediction of flood occurrence other hand, a portion of the rainfall that saturates the soil might
by improving prediction of clustered MCSs and reduce uncer- infiltrate to the deeper soil layers due to the higher hydraulic con-
tainty in flooding occurrence in response to a stronger precipita- ductivity and stronger vertical water pressure gradient during
tion forcing. This linkage can also allow us to better evaluate flood events. Soil moisture at the deeper layers might contribute
variability of flood likelihood in response to changes in different to groundwater storage or moisten the upper layers through hy-
aspects (e.g., favorable for clustered/nonclustered MCSs, changes draulic redistribution. The long-term impacts of clustered MCS
in terrestrial conditions) with climate change. While recognizing events in the water cycle beyond the flooding impacts at short
the importance of initial soil conditions on floods, whether it can time scales can be important for seasonal predictions of hydro-
play a role on initiating or maintaining the MCS clusters is an meteorological events and agricultural outcomes.
open question as noted earlier.
While it is true that relaxing the spatial or temporal thresh-
Acknowledgments. This research is supported by the U.S.
olds to define clustered MCSs tends to shift the peak distribu-
Department of Energy Office of Science Biological and En-
tion of clustered MCSs from early warm-season (April–June)
vironmental Research as part of the Regional and Global
toward June, we confirm that the above conclusions are not
sensitive to the threshold parameters we use to define MCS Model Analysis and Multi-sector Dynamics program areas.
clusters. The effectiveness of clustered MCSs in producing PNNL is operated for the Department of Energy by Battelle
more floods with longer durations highlights the hydrological Memorial Institute under Contract DE-AC05-76RL01830.
and socioeconomic impacts of clustered MCS events. The im-
plication can be twofold. First, we need a better understand- Data availability statement. The 4-km hourly MCS data-
ing of the favorable synoptic conditions of MCS clusters (e.g., base covering 2004–17 is obtained from the U.S. Department
the May 2015 case), their relationships with different climate of Energy Atmospheric Radiation Measurement program
modes of variability (e.g., ENSO), and their potential changes (https://doi.org/10.5439/1571643). The NCEI Storm Events
in the future to understand the predictability of clustered Database is obtained from the NOAA’s National Centers for
MCSs and to better anticipate variability and changes in flood Environmental information (ftp://ftp.ncdc.noaa.gov/pub/data/
risks associated with them. Case studies indicate an important swdi/stormevents/csvfiles/).
role of large-scale external factors (e.g., tropospheric anticy-
clone position, Great Plains low-level jet) and internal feed-
REFERENCES
backs (Trier et al. 2014). With a consistent strengthening of
the Great Plains low-level jet projected by climate models in AP, 2015: Texas floods: Damage tops $45m in Houston as state
the future (Tang et al. 2017; Torres-Alavez et al. 2021), its im- starts to rebuild. Guardian, 31 May, https://www.theguardian.
plication for possible changes of MCS clusters and their com/us-news/2015/may/31/texas-floods-damage-missing-deaths.
NOVEMBER 2022 HU ET AL. 1701

Ashley, S. T., and W. S. Ashley, 2008: Flood fatalities in the }}, R. A. Houze Jr, L. R. Leung, F. Song, J. C. Hardin,
United States. J. Appl. Meteor. Climatol., 47, 805–818, https:// J. Wang, W. I. Gustafson Jr, and C. R. Homeyer, 2019: Spa-
doi.org/10.1175/2007JAMC1611.1. tiotemporal characteristics and large-scale environments of
Ashley, W. S., T. L. Mote, P. G. Dixon, S. L. Trotter, E. J. Powell, mesoscale convective systems east of the Rocky Mountains.
J. D. Durkee, and A. J. Grundstein, 2003: Distribution of me- J. Climate, 32, 7303–7328, https://doi.org/10.1175/JCLI-D-19-
soscale convective complex rainfall in the United States. 0137.1.
Mon. Wea. Rev., 131, 3003–3017, https://doi.org/10.1175/1520- Fritsch, J. M., R. J. Kane, and C. R. Chelius, 1986: The contribu-
0493(2003)131,3003:DOMCCR.2.0.CO;2. tion of mesoscale convective weather systems to the warm-
Bartels, D. L., and R. A. Maddox, 1991: Midlevel cyclonic vortices season precipitation in the United States. J. Appl. Meteor.
generated by mesoscale convective systems. Mon. Wea. Rev., Climatol., 25, 1333–1345, https://doi.org/10.1175/1520-0450
119, 104–118, https://doi.org/10.1175/1520-0493(1991)119,0104: (1986)025,1333:TCOMCW.2.0.CO;2.
MCVGBM.2.0.CO;2. }}, J. D. Murphy, and J. S. Kain, 1994: Warm core vortex am-
Barthold, F. E., T. E. Workoff, B. A. Cosgrove, J. J. Gourley, plification over land. J. Atmos. Sci., 51, 1780–1807, https://doi.
D. R. Novak, and K. M. Mahoney, 2015: Improving flash org/10.1175/1520-0469(1994)051,1780:WCVAOL.2.0.CO;2.
flood forecasts: The HMT-WPC flash flood and intense rain- Geerts, B., 1998: Mesoscale convective systems in the southeast
fall experiment. Bull. Amer. Meteor. Soc., 96, 1859–1866, United States during 1994–95: A survey. Wea. Forecasting,
https://doi.org/10.1175/BAMS-D-14-00201.1. 13, 860–869, https://doi.org/10.1175/1520-0434(1998)013,0860:
Berghuijs, W. R., R. A. Woods, C. J. Hutton, and M. Sivapalan, MCSITS.2.0.CO;2.
2016: Dominant flood generating mechanisms across the Getirana, A. C. V., A. Boone, and C. Peugeot, 2014: Evaluating
United States. Geophys. Res. Lett., 43, 4382–4390, https://doi. LSM-based water budgets over a West African basin assisted
org/10.1002/2016GL068070. with a river routing scheme. J. Hydrometeor., 15, 2331–2346,
Best, M. J., and Coauthors, 2015: The plumbing of land surface https://doi.org/10.1175/JHM-D-14-0012.1.
models: Benchmarking model performance. J. Hydrometeor., Gochis, D. J., and Coauthors, 2018: The WRF-Hydro modeling
16, 1425–1442, https://doi.org/10.1175/JHM-D-14-0158.1. system technical description (version 5.0). NCAR Tech. Note,
Boone, A., and Coauthors, 2009: The AMMA Land Surface Model 107 pp., https://ral.ucar.edu/sites/default/files/public/WRF-
Intercomparison Project (ALMIP). Bull. Amer. Meteor. Soc., HydroV5TechnicalDescription_update512019_0.pdf.
Gourley, J. J., and Coauthors, 2013: A unified flash flood database
90, 1865–1880, https://doi.org/10.1175/2009BAMS2786.1.
across the United States. Bull. Amer. Meteor. Soc., 94, 799–
Bosart, L. F., and F. Sanders, 1981: The Johnstown flood of July
805, https://doi.org/10.1175/BAMS-D-12-00198.1.
1977: A long-lived convective system. J. Atmos. Sci., 38,
Haberlie, A. M., and W. S. Ashley, 2019: A radar-based climatology
1616–1642, https://doi.org/10.1175/1520-0469(1981)038,1616:
of mesoscale convective systems in the United States. J. Cli-
TJFOJA.2.0.CO;2.
mate, 32, 1591–1606, https://doi.org/10.1175/JCLI-D-18-0559.1.
Bowman, K. P., and C. R. Homeyer, 2017: GridRad}Three-
Houze, R. A., Jr, 2004: Mesoscale convective systems. Rev. Geo-
dimensional gridded NEXRAD WSR-88D radar data. Re-
phys., 42, RG4003, https://doi.org/10.1029/2004RG000150.
search Data Archive at the National Center for Atmospheric
}}, B. F. Smull, and P. Dodge, 1990: Mesoscale organization
Research, Computational and Information Systems Labora-
of springtime rainstorms in Oklahoma. Mon. Wea. Rev.,
tory, accessed 13 October 2022, https://doi.org/10.5065/
118 613–654, https://doi.org/10.1175/1520-0493(1990)118,0613:
D6NK3CR7.
MOOSRI.2.0.CO;2.
Cui, W., X. Dong, B. Xi, and Z. Feng, 2021: Climatology of linear
Hu, H., L. R. Leung, and Z. Feng, 2020a: Observed warm-season
mesoscale convective system morphology in the United characteristics of MCS and non-MCS rainfall and their recent
States based on random forests method. J. Climate, 34, 7257– changes in the central United States. Geophys. Res. Lett., 47,
7276, https://doi.org/10.1175/JCLI-D-20-0862.1. e2019GL086783, https://doi.org/10.1029/2019GL086783.
Dougherty, E., and K. L. Rasmussen, 2019: Climatology of flood- }}, }}, and }}, 2020b: Understanding the distinct impacts
producing storms and their associated rainfall characteristics of MCS and non-MCS rainfall on the surface water balance
in the United States. Mon. Wea. Rev., 147, 3861–3877, https:// in the central US using a numerical water-tagging technique.
doi.org/10.1175/MWR-D-19-0020.1. J. Hydrometeor., 21, 2343–2357, https://doi.org/10.1175/JHM-
Fan, Y., and Coauthors, 2019: Hillslope hydrology in global D-20-0081.1.
change research and Earth system modeling. Water Resour. }}, Z. Feng, and L.-Y. R. Leung, 2021a: Linking flood fre-
Res., 55, 1737–1772, https://doi.org/10.1029/2018WR023903. quency with mesoscale convective systems in the US. Geo-
Feng, Z., 2019: Mesoscale convective system (MCS) database over phys. Res. Lett., 48, e2021GL092546, https://doi.org/10.1029/
United States. USDOE Office of Science, accessed 13 October 2021GL092546.
2022, https://doi.org/10.5439/1571643. }}, L. R. Leung, and Z. Feng, 2021b: Early warm-season meso-
}}, S. Hagos, A. K. Rowe, C. D. Burleyson, M. N. Martini, and scale convective systems dominate soil moisture–precipitation
S. P. de Szoeke, 2015: Mechanisms of convective cloud orga- feedback for summer rainfall in central United States. Proc.
nization by cold pools over tropical warm ocean during the Natl. Acad. Sci. USA, 118, e2105260118, https://doi.org/10.
AMIE/DYNAMO field campaign. J. Adv. Model. Earth 1073/pnas.2105260118.
Syst., 7, 357–381, https://doi.org/10.1002/2014MS000384. Janowiak, J. E., R. J. Joyce, and Y. Yarosh, 2001: A real-time
}}, L. R. Leung, R. A. Houze Jr, S. Hagos, J. Hardin, Q. Yang, global half-hourly pixel-resolution infrared dataset and its ap-
B. Han, and J. Fan, 2018: Structure and evolution of meso- plications. Bull. Amer. Meteor. Soc., 82, 205–218, https://doi.
scale convective systems: Sensitivity to cloud microphysics in org/10.1175/1520-0477(2001)082,0205:ARTGHH.2.3.CO;2.
convection-permitting simulations over the United States. J. Kundzewicz, Z. W., and Coauthors, 2014: Flood risk and climate
Adv. Model. Earth Syst., 10, 1470–1494, https://doi.org/10. change: Global and regional perspectives. Hydrol. Sci. J., 59,
1029/2018MS001305. 1–28, https://doi.org/10.1080/02626667.2013.857411.
1702 JOURNAL OF HYDROMETEOROLOGY VOLUME 23

Kunkel, K. E., D. R. Easterling, D. A. R. Kristovich, B. Gleason, Mon. Wea. Rev., 136, 3964–3986, https://doi.org/10.1175/
L. Stoecker, and R. Smith, 2012: Meteorological causes of the 2008MWR2471.1.
secular variations in observed extreme precipitation events Sharma, A., C. Wasko, and D. P. Lettenmaier, 2018: If precipita-
for the conterminous United States. J. Hydrometeor., 13, tion extremes are increasing, why aren’t floods? Water Resour.
1131–1141, https://doi.org/10.1175/JHM-D-11-0108.1. Res., 54, 8545–8551, https://doi.org/10.1029/2018WR023749.
Laing, A. G., and J. M. Fritsch, 1997: The global population of Song, F., Z. Feng, L. R. Leung, R. A. Houze Jr., J. Wang, J. Hardin,
mesoscale convective complexes. Quart. J. Roy. Meteor. Soc., and C. R. Homeyer, 2019: Contrasting spring and summer
123, 389–405, https://doi.org/10.1002/qj.49712353807. large-scale environments associated with mesoscale convective
Lin, Y., 2011: GCIP/EOP surface: Precipitation NCEP/EMC 4KM systems over the U.S. Great Plains. J. Climate, 32, 6749–6767,
Gridded Data (GRIB) Stage IV Data, version 1.0. UCAR/ https://doi.org/10.1175/JCLI-D-18-0839.1.
NCAR-Earth Observing Laboratory, accessed 13 October Stevenson, S. N., and R. S. Schumacher, 2014: A 10-year survey
2022, https://data.eol.ucar.edu/dataset/21.006. of extreme rainfall events in the central and eastern United
Maddox, R. A., L. R. Hoxit, C. F. Chappell, and F. Caracena, States using gridded multisensor precipitation analyses. Mon.
1978: Comparison of meteorological aspects of the Big Wea. Rev., 142, 3147–3162, https://doi.org/10.1175/MWR-D-
Thompson and Rapid City flash floods. Mon. Wea. Rev., 13-00345.1.
106, 375–389, https://doi.org/10.1175/1520-0493(1978)106,0375: Tang, Y., J. Winkler, S. Zhong, X. Bian, D. Doubler, L. Yu, and
COMAOT.2.0.CO;2. C. Walters, 2017: Future changes in the climatology of the
Martinez, M., and B. Brunfeld, 2015: Texas floods: Enough rain fell Great Plains low-level jet derived from fine resolution multi-
in May to cover the entire state 8 inches deep. CNN, 31 May model simulations. Sci. Rep., 7, 5029, https://doi.org/10.1038/
2015, https://www.cnn.com/2015/05/31/us/severe-weather/index. s41598-017-05135-0.
html. Torres-Alavez, J. A., and Coauthors, 2021: Future projections in
Merz, B., and Coauthors, 2014: Floods and climate: Emerging per- the climatology of global low-level jets from CORDEX-
spectives for flood risk assessment and management. Nat. CORE simulations. Climate Dyn., 57, 1551–1569, https://doi.
Hazards Earth Syst. Sci., 14, 1921–1942, https://doi.org/10. org/10.1007/s00382-021-05671-6.
Trier, S. B., and C. A. Davis, 2002: Influence of balanced motions
5194/nhess-14-1921-2014.
on heavy precipitation within a long-lived convectively gener-
Niu, G.-Y., and Coauthors, 2011: The community Noah land sur-
ated vortex. Mon. Wea. Rev., 130, 877–899, https://doi.org/10.
face model with multiparameterization options (Noah-MP):
1175/1520-0493(2002)130,0877:IOBMOH.2.0.CO;2.
1. Model description and evaluation with local-scale measure-
}}, }}, and W. C. Skamarock, 2000: Long-lived mesoconvec-
ments. J. Geophys. Res., 116, D12109, https://doi.org/10.1029/
tive vortices and their environment. Part II: Induced thermo-
2010JD015139.
dynamic destabilization in idealized simulations. Mon. Wea.
NWS, 2018: National Weather Service Instruction 10-1065. Na-
Rev., 128, 3396–3412, https://doi.org/10.1175/1520-0493(2000)
tional Oceanic and Atmospheric Administration, http://www.
128,3396:LLMVAT.2.0.CO;2.
nws.noaa.gov/directives/.
}}, }}, D. A. Ahijevych, M. L. Weisman, and G. H. Bryan,
Parker, M. D., and R. H. Johnson, 2000: Organizational modes of
2006: Mechanisms supporting long-lived episodes of propa-
midlatitude mesoscale convective systems. Mon. Wea. Rev.,
gating nocturnal convection within a 7-day WRF Model sim-
128, 3413–3436, https://doi.org/10.1175/1520-0493(2001)129
ulation. J. Atmos. Sci., 63, 2437–2461, https://doi.org/10.1175/
,3413:OMOMMC.2.0.CO;2.
JAS3768.1.
Pinto, J. O., J. A. Grim, and M. Steiner, 2015: Assessment of the
}}, F. Chen, K. W. Manning, M. A. LeMone, and C. A. Davis,
High-Resolution Rapid Refresh Model’s Ability to predict 2008: Sensitivity of the PBL and precipitation in 12-day simu-
mesoscale convective systems using object-based evaluation. lations of warm-season convection using different land sur-
Wea. Forecasting, 30, 892–913, https://doi.org/10.1175/WAF- face models and soil wetness conditions. Mon. Wea. Rev.,
D-14-00118.1. 136, 2321–2343, https://doi.org/10.1175/2007mwr2289.1.
Prein, A. F., C. Liu, K. Ikeda, R. Bullock, R. M. Rasmussen, G. J. }}, }}, C. A. Davis, and R. E. Carbone, 2014: Mechanisms
Holland, and M. Clark, 2017: Simulating North American governing the persistence and diurnal cycle of a heavy rainfall
mesoscale convective systems with a convection-permitting corridor. J. Atmos. Sci., 71, 4102–4126, https://doi.org/10.1175/
climate model. Climate Dyn., 55, 95–110, https://doi.org/10. JAS-D-14-0134.1.
1007/s00382-017-3993-2. Tuttle, J. D., and C. A. Davis, 2006: Corridors of warm season
Ralph, F. M., P. J. Neiman, G. N. Kiladis, K. Weickmann, and precipitation in the central United States. Mon. Wea. Rev.,
D. W. Reynolds, 2011: A multiscale observational case study 134, 2297–2317, https://doi.org/10.1175/MWR3188.1.
of a Pacific atmospheric river exhibiting tropical–extratropical Villarini, G., 2016: On the seasonality of flooding across the conti-
connections and a mesoscale frontal wave. Mon. Wea. Rev., nental United States. Adv. Water Resour., 87, 80–91, https://
139, 1169–1189, https://doi.org/10.1175/2010MWR3596.1. doi.org/10.1016/j.advwatres.2015.11.009.
Saharia, M., P.-E. Kirstetter, H. Vergara, J. J. Gourley, and Vischel, T., T. Lebel, S. Massuel, and B. Cappelaere, 2009: Condi-
Y. Hong, 2017: Characterization of floods in the United tional simulation schemes of rain fields and their application
States. J. Hydrol., 548, 524–535, https://doi.org/10.1016/j. to rainfall–runoff modeling studies in the Sahel. J. Hydrol.,
jhydrol.2017.03.010. 375, 273–286, https://doi.org/10.1016/j.jhydrol.2009.02.028.
Schumacher, R. S., and R. H. Johnson, 2005: Organization and Wang, S.-Y. S., W.-R. Huang, H.-H. Hsu, and R. R. Gillies, 2015:
environmental properties of extreme-rain-producing meso- Role of the strengthened El Niño teleconnection in the May
scale convective systems. Mon. Wea. Rev., 133, 961–976, 2015 floods over the southern Great Plains. Geophys. Res.
https://doi.org/10.1175/MWR2899.1. Lett., 42, 8140–8146, https://doi.org/10.1002/2015GL065211.
}}, and }}, 2008: Mesoscale processes contributing to ex- Wright, D. B., G. Yu, and J. F. England, 2020: Six decades of rain-
treme rainfall in a midlatitude warm-season flash flood. fall and flood frequency analysis using stochastic storm
NOVEMBER 2022 HU ET AL. 1703

transposition: Review, progress, and prospects. J. Hydrol., Hydrometeor., 18, 1997–2009, https://doi.org/10.1175/JHM-D-
585, 124816, https://doi.org/10.1016/j.jhydrol.2020.124816. 16-0207.1.
Ye, S., H.-Y. Li, L. R. Leung, J. Guo, Q. Ran, Y. Demissie, and Zhu, Z., D. B. Wright, and G. Yu, 2018: The impact of rainfall
M. Sivapalan, 2017: Understanding flood seasonality and its space-time structure in flood frequency analysis. Water Resour.
temporal shifts within the contiguous United States. J. Res., 54, 8983–8998, https://doi.org/10.1029/2018WR023550.
Copyright of Journal of Hydrometeorology is the property of American Meteorological
Society and its content may not be copied or emailed to multiple sites or posted to a listserv
without the copyright holder's express written permission. However, users may print,
download, or email articles for individual use.

You might also like