Agricultural and Forest Meteorology 247 (2017) 414–423

Contents lists available at ScienceDirect

Agricultural and Forest Meteorology

journal homepage: www.elsevier.com/locate/agrformet

Meteorological conditions associated with the onset of flash drought in the MARK
Eastern United States
⁎
Trent W. Forda, , Christopher F. Labosierb
a
Department of Geography and Environmental Resources Southern Illinois University, Carbondale, IL 62901, United States
b
Department of Biological and Environmental Sciences Longwood University, Farmville, VA 23909, United States

A R T I C L E I N F O A B S T R A C T

Keywords: Rapid onset droughts, termed “flash droughts”, present a series of unique challenges for drought monitoring,
Flash drought forecasting, and mitigation. Due to the rapid onset and lack of early warning systems, stakeholders can be caught
Soil moisture deficit off-guard by flash droughts and suffer disproportionate impacts. Despite these impacts, little is known about the
Mid-level ridging physical drivers of flash droughts. The purpose of this study is to determine antecedent meteorological condi-
tions prior to the onset of flash drought in the Eastern United States. Emphasizing the agricultural impacts, flash
droughts were defined as periods when the pentad-average 0–40 cm volumetric water content declines from at
least the 40th percentile to below the 20th percentile in 4 pentads or less. Meteorological variables from 125
stations in the Eastern U.S. from March − October 1979 − 2010 were analyzed for their relationships with flash
drought onset. Consistent with previous findings, flash drought was associated with decreased precipitation and
humidity, increased solar radiation, and elevated temperatures. However logistic regression results suggest
variables that accounted for surface moisture balance and/or atmospheric evaporative demand were more
closely linked with the likelihood of flash drought than temperature and/or precipitation. Associated surface
conditions are likely driven by ridging in the mid to upper level troposphere, which is shown to be more per-
sistent leading up the flash droughts in the northern half of the study region. Our results elucidate the me-
teorological conditions immediately prior to the onset of one type or “flavor” of flash drought, defined by
characteristic rapid intensification. Arguably, one could define flash drought with soil moisture thresholds
varying from those used in this study and/or different time scales of soil moisture depletion. Therefore, we
additionally argue that absences of both a standard flash drought definition and consistent precedent for
identifying flash drought complicates monitoring and predicting these events.

1. Introduction from observations, satellite remote sensing, and models to generate a

Drought is one of the most damaging and costly natural hazards in
the United States because of the dependence on agriculture and the
impact that drought has on water resources and ecosystems. For ex-
ample, the 2011 drought caused an estimated $7.62 billion of agri-
cultural losses and nearly $800 million in timber resources in Texas
alone (Hoerling et al., 2013). Because of the risk drought poses, stra-
tegies have been developed worldwide, with particular foci on effective
drought monitoring and communication (Hayes et al., 2011). One of
the most comprehensive drought monitoring systems is the United
States Drought Monitor (USDM), which was developed to track and
communicate the severity and extent of drought across the United
States (Svoboda et al., 2002). The USDM and other state-of-the-art
drought monitoring systems assimilate information related to water
resource availability, ecosystem health, and meteorological conditions
product that can be effective for a wide variety of end-users. However,
despite the utility of comprehensive drought monitoring tools such as
the USDM, they often have difficulty capturing rapidly evolving
drought events commonly referred to as “flash droughts” (Svoboda
et al., 2002; Senay et al., 2008; Otkin et al., 2013), first coined by
Svoboda et al. (2002) .The rapid onset of flash droughts significantly
reduces time available for impact mitigation, potentially resulting in
greater adverse agricultural and societal effects than a slowly evolving
drought event (Otkin et al., 2015). Concurrently, the physical drivers of
flash droughts, common to regions of the United States east of the
Rocky Mountains (hereafter regarded as the Eastern United States), are
not well understood (Mo and Lettenmaier 2016). This is in contrast to
conventional droughts such as the 2011 Texas event and drought events
of the 1930s and 1950s, as the causes of these droughts are known to be
remote factors such as sea-surface temperature (SST) anomalies

⁎
Corresponding author.
E-mail address: twford@siu.edu (T.W. Ford).

http://dx.doi.org/10.1016/j.agrformet.2017.08.031
Received 12 April 2017; Received in revised form 16 August 2017; Accepted 25 August 2017
Available online 06 October 2017
0168-1923/ © 2017 Elsevier B.V. All rights reserved.
T.W. Ford, C.F. Labosier
augmented by land-atmosphere interactions and soil moisture de-
siccation (Schubert et al., 2004; Fernando et al., 2016). The primary
consequence of the knowledge gap of the drivers of and meteorological
conditions that lead to flash drought is a lack of effective monitoring
and forecasting infrastructure. Indeed, the 2012 Great Plains flash
drought arrived without early warning from seasonal climate forecast
models and operational drought monitoring systems (Hoerling et al.,
2014; Kam et al., 2014). The economic losses directly attributed to the
2012 drought in the United States exceeded $12 billion (Hoerling et al.,
2014).
The frequency with which flash droughts occur in the Eastern
United States (Mo and Lettenmaier 2015) combined with the significant
lack of knowledge and capability to predict and monitor flash drought
underscore the need for investigation of the meteorological conditions
that lead to flash drought onset. One consistent theme amid the studies
examining flash droughts in the United States is the importance of soil
moisture observations and evapotranspiration for flash drought mon-
itoring and forecasting (Otkin et al., 2014; Ford et al., 2015). Ad-
ditionally it is known that inception of flash drought in the Eastern
United States occurs with rapid declines in soil moisture availability,
attributed to decreased precipitation and amplified by increased air
temperature and elevated evaporative demand (Otkin et al., 2013; Mo
and Lettenmaier 2015). With this in mind the purpose of this project is
to determine both the antecedent meteorological conditions most
strongly associated with flash drought occurrence and the time-scale at
which these connections occur. The scope of this analysis is climato-
logical, such that we do not focus on one or two particular flash drought
events, but identify flash droughts over a 32-year time period and de-
scribe the atmospheric conditions associated with a station’s flash
drought climatology.
2. Material and methods
2.1. Meteorological observations
Hourly observations of temperature, dew point temperature, station
pressure, relative humidity, precipitation, and wind speed between
415
Agricultural and Forest Meteorology 247 (2017) 414–423
1979 and 2010 were obtained from 125 “first order” weather stations
across the Eastern United States (Fig. 1), defined as all states east of the
Rocky Mountains or approximately 105°W longitude. The stations were
selected as a compromise between 1) spanning the geographic entirety
of the Eastern United States, 2) working with nearly serially-complete,
high quality observations, and 3) inclusion in the National Solar Ra-
diation Database program (see below). Hourly observations were as-
signed to the nearest rounded hour and partitioned into the traditional
eight 3-h time blocks (0000–0300 LST, 0300–0600 LST, etc.). Daily
maximum temperature (TMAX, °C), mean 2-m wind speed (Wspd, ms−1),
mean relative humidity (RH, %), total solar radiation (Srad, Wm−2),
and total precipitation (Precip, mm) were computed from the 3-h
average or total values; however, they were only considered if at least
one valid hourly observation was included in each 3-h block. If one of
the 3-h blocks was missing for a variable, the variable was not com-
puted for that day. In this situation, all other variables were computed
for that particular day, given all 8, 3-h blocks were valid. TMAX was
included for each 3-h block as the maximum hourly temperature re-
corded in that block. Therefore, daily TMAX represents the actual max-
imum hourly temperature value for that day. A station was considered
for the analysis if less than 10% of daily observations were missing.
Additionally, we computed daily average vapor pressure deficit (VPD,
mb) as the difference between the saturation vapor pressure at the daily
maximum and minimum air temperatures (es) and the actual vapor
pressure (ea). The former is calculated such that:
eTmax + eTmin
es =
2
where
eT = 0.6108e ( 17.67T T+237.3 ),
and the latter is computed using the empirical relationship of Bolton
(1980) from daily average dew point temperature (Td, °C):
17.67Td ⎞
ea = 6.112exp ⎛ ⎜ . ⎟
⎝ Td + 243.5 ⎠
Fig. 1. Map of first-order weather stations,
from which meteorological conditions are
composited. The stations are colored based
on their respective cluster (clusters based on
National Centers for Environmental
Information (NCEI) climatic regions, as in
Karl and Koss (1984)).
T.W. Ford, C.F. Labosier
2.2. Evaporative demand
Recent studies have elaborated on the importance of evapo-
transpiration (ET) and atmospheric evaporative demand, otherwise
referred to as potential evapotranspiration (PET), for the rapid deple-
tion of soil moisture in flash drought conditions (Otkin et al., 2016;
McEvoy et al., 2016). Indeed physically-based methods of estimating
PET, such as the Penman-Monteith method (Monteith 1965) which
incorporates radiation and aerodynamic components, are preferred
over those methods only incorporating temperature. The primary lim-
itation of the Penman-Monteith method is its need for data resources,
including components that are often not available or of poor quality,
such as net radiation (Rn). We use daily Rn observations from the same
125 first-order weather stations to fulfill the Penman-Monteith ne-
cessity of good quality net radiation data. These observations are part of
the National Solar Radiation Database (NSRDB, Maxwell et al., 1995;
Wilcox 2007), which comprises hourly Rn measurements from hundreds
of National Weather Service stations. Most stations archive measured
Rn from 1961 until 2010, although because of the lack of robust soil
moisture datasets prior to 1979, we only obtain Rn from the period
1979–2010. Unfortunately, NSRDB data after 2010 is derived using the
Physical Solar Model and is therefore less compatible with direct ob-
servations in the early part of the record. Daily Rn is obtained from
totaling hourly observations of horizontal irradiance (direct + diffuse,
Wm−2). Days in which any hourly observation was missing or erro-
neous were considered missing. Daily PET was computed using ob-
servations from the 125 weather stations, based on the Food and
Agricultural Organization (FAO) Penman-Monteith method (Allen
et al., 1998):
900
0.408Δ(Rn − G ) + γ T + 273 U2 (es − ea)
PET = ,
Δ + γ (1 + 0.34U2)
where G is soil heat flux, U2 is wind speed, Δ is the saturated vapor
pressure slope, and γ is the psychrometric constant. Daily observations
from the weather stations were used for U2, Δ, Rn, and T, and G was
considered null for daily estimates (e.g., Allen et al., 1998; Cai et al.,
2007; Yuan and Quiring, 2014). The FAO Penman-Monteith method
widely used for estimating PET because it is physically based, in-
corporates equations for estimating bulk surface resistance and aero-
dynamic resistance (Allen et al., 1998), and can be calculated from
readily available meteorological observations (Sentelhas et al., 2010).
The coefficients used in the FAO-PM method can introduce uncertainty
in PET estimates as they are not applicable to all landscapes. However,
the FAO-PM method has been widely validated in diverse surface
conditions and climates, demonstrating its robust applicability for es-
timating PET from meteorological applications (McMahon et al., 2013;
Peel and McMahon, 2014). A simple surface moisture balance was
subsequently calculated as the difference of daily precipitation and PET
(P − PET, mm).
Daily anomalies of Tmax, Wspd, RH, Srad, PET, Precip, P − PET, and
VPD were derived using the climatological (1979–2010) mean and
standard deviation of that calendar day. The daily anomalies were then
averaged to pentad-scale; we use pentads in order to both 1) minimize
daily variability in noisy variables such as Precip and Wspd and 2)
capture the rapid onset of flash drought events. Pentad and weekly time
scales have been employed previously in flash drought investigations
(Ford et al., 2015; Mo and Lettenmaier 2016).
2.3. Synoptic-Scale conditions
Along with surface meteorological conditions, we composite sy-
noptic-scale mid-level and upper-level atmospheric circulation features
associated with flash drought onset. The North American Regional
Reanalysis (NARR, Mesinger et al., 2006) dataset is used to characterize
atmospheric conditions immediately prior to flash drought events.
416
Agricultural and Forest Meteorology 247 (2017) 414–423
NARR is a moderate resolution (32 km) atmospheric and land surface
reanalysis product that covers all of North America and spans the time
period 1979–2016. We composite 500- and 300-hPa geopotential
heights to describe the synoptic- and larger-scale circulation features
associated with the surface weather and soil moisture conditions that
characterize flash drought.
2.4. Soil moisture
Despite its relatively short evolution, flash drought frequently re-
sults in significant damage to natural and agricultural systems (Svoboda
et al., 2002; Hunt et al., 2014; Anderson et al., 2016). Therefore soil
moisture, in particular the rapid depletion of soil moisture, is an im-
portant indicator of flash drought occurrence. Several considerations
must be made when selecting sources of soil moisture information for
this type of analysis, including 1) the historical record length and
temporal resolution, 2) data quality, and 3) representativeness of re-
levant soil moisture conditions. The first consideration precludes ana-
lysis of in situ soil moisture observations, as very few stations have been
monitoring soil moisture for 20+ years, and the majority of these
longer-term monitoring stations are concentrated in one or two regions.
Satellite remote sensing soil moisture datasets exist over a longer time
period and are advantageous, relative to in situ observations, in their
spatial continuity. However, most microwave, visible, and thermal-
band soil moisture products only provide information on the most
surficial soil layer (i.e., < 10 cm). Although observations of surficial
soil moisture are important for drought monitoring, particularly in
sparsely-vegetated regions, ideally analysis of flash drought in the
Eastern United States should be grounded in root zone soil moisture
(Hunt et al., 2009). Accounting for these considerations, we used soil
moisture from the Noah land surface model, as part of the North
American Land Data Assimilation System phase 2 (NLDAS-2, Xia et al.,
2012a,b). The NLDAS-2 soil moisture products have been evaluated and
validated with a wide-variety of in situ observations and are able to
capture the broader features of soil moisture variability in the Eastern
United States (Xia et al., 2014,2015). With that being said, the soil
moisture dataset used in this study is still a modeled dataset and
therefore is less accurate than actual in situ observed soil moisture. The
results, with respect to soil moisture, should be viewed with this caveat
in mind.
As part of NLDAS-2, the Noah model supplies daily soil moisture
fields in three layers; 0–10 cm, 10–40 cm, and 40–100 cm, at a 1/8°
resolution. Daily NLDAS-2 0–10 cm and 10–40 soil moisture fields (in
units of mm) were combined into one 0–40 cm soil moisture dataset
and subsequently converted to volumetric water content (cm−3 cm−3).
The 0–40 cm depth interval was chosen to generally represent the “root
zone” over the seasonal time period assessed. It should be noted, though
that depending on the geographic location this interval may extend too
deep in the spring and potentially too shallow in the mid-to-late
summer. The soil moisture was then standardized using percentiles of
volumetric water content, and the percentiles were then averaged to the
pentad-scale, to match the resolution of the meteorological data. Each
of the 125 weather stations was matched to the soil moisture time series
of the NLDAS-2 grid cell in which the station resided. All analyses of
meteorological conditions and soil moisture was done using the pentad-
scale data.
2.5. Flash drought
The surge in flash drought research over the last 10–15 years has,
curiously, not been accompanied by a widely-accepted or standard
definition of flash drought. The lack of precedent and resulting ambi-
guity has forced researchers to either specify a flash drought definition
for their own purpose (e.g., Mo and Lettenmaier 2015; Ford et al.,
2015) or identify flash drought case studies using a posteriori knowledge
(e.g., Otkin et al., 2013; Anderson et al., 2013; Hunt et al., 2014). We
T.W. Ford, C.F. Labosier
argue that flash drought determination should first and foremost be
based on soil moisture. With that being said, other environmental in-
formation such as remote sensing of vegetation health and ET are
better-suited for near-real time flash drought monitoring and prediction
than is soil moisture. An additional consideration is the short time scale
over which flash drought can establish and intensify. For this reason,
flash drought identification should include a built-in intensification-
time component that is appropriate for effective flash drought detec-
tion. For this project we identify a flash drought if the pentad-average
0–40 cm soil moisture percentile at a station declines from at or above
the 40th percentile to at or below the 20th percentile in 4 pentads or
less. These thresholds were subjectively determined, based both on
previous work with soil moisture-identified flash droughts (e.g., Ford
et al., 2015) and significant experience working with in situ soil
moisture data (e.g., Ford and Quiring, 2014; Quiring et al., 2016). The
20th percentile of soil moisture was selected as it represents the onset of
moderate drought, according to the USDM. The 40th percentile of soil
moisture was selected as the upper threshold because it represents drier
than normal soils, but not drought. This allowed us to evaluate me-
teorological conditions leading to the onset of flash drought separately
from situations where pre-existing drought rapidly intensifies into a
larger magnitude event. We contend that both of these conditions (i.e.,
rapid onset and rapid intensification) should be considered flash
drought, but the former is the phenomenon of interest here. The four
pentad intensification period sits squarely in the subseasonal-to-sea-
sonal timescale on which flash drought operates. Additionally, we
tested a range of soil moisture percentile thresholds (50th to 30th
percentile for top threshold, 30th to 10th percentile for bottom
threshold) and drought-intensification thresholds (3 pentads to 8 pen-
tads). The meteorological composites were mostly independent of the
thresholds selected. Therefore we determined these thresholds to be
appropriate to accomplish the objectives of this particular investigation.
Flash droughts between March and October are identified at each of the
125 stations over the period 1979–2010.
2.6. Logistic regression
We use composite analysis to determine the meteorological condi-
tions associated with the onset of flash drought in the Eastern United
States. Additionally, we employ logistic regression, as described by
Peng et al. (2002) to examine the statistical, predictive relationship
between meteorological conditions and the probability of antecedent
flash drought. Logistic regression is advantageous when exploring the
relationship between ratio, ordinal, and binary independent and de-
pendent variables. In this case, logistic regression measures the logit
transformation of the odds ratio of flash drought onset given a change
in various meteorological variables. The logistic regression model takes
the form of:
ρ or lead at which these anomalies occur prior to flash drought imply the
ln = α + βX ,
1−ρ
ρ
where is the odds ratio of flash drought occurrence. The logistic
1−ρ
regression results provide evidence of what meteorological variables
are most strongly associated with flash drought onset, and at what time
lags the association peaks.
3. Results
3.1. Flash drought climatology
Flash drought at stations across the Eastern United States, as iden-
tified from pentad-average soil moisture, occurred in between 0.1% and
1.5% of all March − October pentads (Fig. 2a), corresponding with
flash drought return intervals between 1.5 and 10 years. Stations in the
Southeast region tended to have a higher overall flash drought occur-
rence, particularly compared to stations in the Great Plains region. This
417
Agricultural and Forest Meteorology 247 (2017) 414–423
pattern is somewhat counterintuitive given the prevalence of notable
Great Plains flash droughts in the recent past (Senay et al., 2008; Otkin
et al., 2015, 2016). However, more frequent flash droughts in the
Southeast is strongly (R2 = 0.53) related to shorter soil moisture
“memory” (Supplementary Fig. 1), defined here as the number of
pentads it takes the lagged 0–40 cm soil moisture autocorrelation to
drop to 1 e (e.g., Dirmeyer et al., 2016). Fewer flash droughts identified
in the Great Plains region, therefore is a direct result of longer soil
moisture memory, making it less likely that volumetric water content in
the Great Plains will degrade from the 40th to the 20th within the
imposed 4-pentad maximum time period. However, longer soil
moisture memory also dictates that dry soil anomalies will persist
longer, perhaps at least partly explaining the distinction of Great Plains
flash droughts (Svoboda et al., 2002; Hunt et al., 2008; Ford et al.,
2015; Otkin et al., 2016). Differences in soil moisture memory between
stations seem to be somewhat driven by differences in soil texture
(Supplementary Fig. 2), such that stations in the Southeast U.S. exhibit
lower soil porosity than those farther west in the Great Plains. Physi-
cally, this makes sense as the sandier soils of the Southeast U.S. will
permit faster infiltration and, more relevant here, desiccation
throughout the 0–40 cm soil column than the higher clay-content soils
of the Great Plains and the loamy soils of the Midwest.
Although there is a notable geographic variation in the climatolo-
gical flash drought occurrence between 1979 and 2010, the absolute
frequencies of occurrence are less than that reported in Mo and
Lettenmaier (2015, 2016). This could partially be attributed to the
difference in study period length, with the latter study identifying flash
droughts over a nearly–100 year timeframe. However, we argue the
primary difference is the method of identifying flash droughts, as Mo
and Lettenmaier (2015, 2016) require soils drier than the 40th per-
centile (20th percentile in this study) and mandate thresholds of tem-
perature and ET anomalies. We also find a considerable geographic
distribution in the time of the (March − October) season in which flash
drought occurs, with stations in the Great Plains and Upper Midwest
occurring later than those stations farther east (Fig. 2b). Flash drought
seasonality patterns in Fig. 2b are somewhat reversed from frequency
patterns in Fig. 2a, suggesting soil moisture memory could be partly
attributable for differences in the seasonal timing of flash drought onset
across the study region. Stations with sandier, lower porosity soils are
able to dry faster in the presence of the right meteorological forcings
(discussed below), and therefore flash droughts are able to occur earlier
in the March − October season (Fig. 2b).
3.2. Meteorological conditions
Pentad-scale surface meteorological conditions between 1 and 4
pentads prior to flash drought onset were composited for each of the
125 stations. The magnitude of the anomalies, as well as the time scale
strength of the statistical association between that particular variable
and the subsequent onset of flash drought. The relative lead times of
significant anomalies are particularly important, as the meteorological
variables composited here often co-vary. For example, prolonged pre-
cipitation deficits for several pentads could not only lead to the onset of
flash drought but also contribute to significantly elevated Tmax
anomalies in the pentads leading up to the drought onset.
Composites of Tmax, PET, Srad, and Wspd anomalies are shown at 4,
3, 2, and 1 pentad(s) prior to flash drought occurrence at each station
(Fig. 3). Unsurprisingly, the anomalies of the former 3 variables are
dominantly positive (i.e., increased temperatures, evaporative demand,
solar radiation) leading up to flash drought onset; however noticeable
geographic disparities exist in the lead time at which this pattern first
appears. Positive, albeit weaker anomalies of PET are found throughout
the Southern Great Plains out to a 4-pentad lead, and across the
southern half of the region at a 3-pentad lead. Composites of RH, Precip,
P − PET, and VPD show similar patterns, with positive anomalies of
T.W. Ford, C.F. Labosier
Fig. 2. Left panel: frequency of flash drought occurrence (% of March-October pentads) from 1979 to 2010. Right panel: median flash drought onset day of the year.
VPD and negative anomalies of RH, Precip, and P − PET out to a 2-
pentad lead (Fig. 4). Similar persistence of positive evaporative demand
anomalies in the Southern Great Plains have been previously connected
with flash drought onset (Otkin et al., 2013). This suggests that flash
droughts in the Southern Great Plains region are associated with rela-
tively short (< two weeks) periods of very dry conditions (i.e., negative
RH and Precip anomalies), but with longer precedent time periods of
elevated evaporative demand. Interestingly, P − PET exhibits a rapid
transition from positive anomalies (soil moisture surplus) across the
Eastern half of the study region to negative anomalies (soil moisture
deficit) in a period of 2–3 pentads. In the same part of the study region
RH, Srad, and PET show similar patterns at a 4 pentad lead, demon-
strating the quick evolution of flash drought conditions in this area.
These results suggest a difference in the timescales of flash drought
development between the eastern and western halves of the study re-
gion. Much of the area east of the Mississippi River exhibits a rapid
418
Agricultural and Forest Meteorology 247 (2017) 414–423
transition from near-to-above normal soil moisture conditions to flash
drought in a matter of 1–2 pentads. This is contrasted by the western
extent of the region, particularly the Southern Great Plains, where soil
moisture and atmospheric moisture deficit conditions appear in the
form of negative (positive) P − PET (VPD) anomalies 3–4 pentads prior
to the official flash drought onset. These timescale differences could
potentially be related to the sensitivity of soil water storage to relatively
quick changes in atmospheric water supply and demand, such that only
5- to 10-day periods of negative P − PET anomalies are necessary to
instill drought conditions in the Midwest, Mid-Atlantic, and Northeast
regions of the U.S. Such information is useful to determine the time-
scales necessary to monitor and forecast potential flash drought con-
ditions.
Fig. 3. Pentad anomalies of Tmax, PET, Srad,
and Wspd prior to flash drought onset. From
left to right the panels show anomaly com-
posites 1, 2, 3, and 4 pentads prior to flash
drought onset.
T.W. Ford, C.F. Labosier
3.3. Synoptic conditions
Surface meteorological conditions immediately prior to flash
droughts in the Eastern United States are consistent with increased
evaporative demand, decreased precipitation and humidity, and de-
velopment of soil moisture deficits. These conditions are, unsurpris-
ingly, related to the establishment of intense ridging in the mid-to-
upper-level troposphere (Fig. 5). The 125 weather stations were clus-
tered into 5 regions (Fig. 1), based on the National Centers for En-
vironmental Information (NCEI) climate regions (Karl and Koss, 1984).
For our purposes, we include all stations in Nebraska, South Dakota,
and North Dakota in the Upper Midwest cluster. Additionally, the only
station in Mississippi, Meridian, was included in the Southeast cluster
because it was geographically and climatologically closer to other sta-
tions in the Southeast region. Anomalies of pentad-averaged 500 hPa
geopotential heights are composited for all stations within each of the
five regions, demonstrating the consistency with which flash droughts
can be attributed to strong mid-level ridging to the west-to-northwest of
the study area. In each sub-region, the ridge axis is located immediately
to the west/northwest of the region’s geographic center, corresponding
with weak northerly-to-northeasterly wind patterns (not shown). The
synoptic-scale atmospheric patterns leading up to flash drought in the
South region differ somewhat from those in other regions, such that the
ridge is weaker even for the 5-day composite, and the axis of the ridge
shifts to the east as the flash drought onset approaches (Fig. 5). The
South region’s 500 hPa height anomaly pattern’s most noticeable de-
viation from the other regions is the lack of ridging at 30- and 20-day
periods ahead of the flash drought. The consistent appearance of rid-
ging to the north-northwest of each region (the South region omitted) at
multiple-pentad lead times suggests the persistence, and not necessarily
just the establishment, of a mid-level ridge is necessary for flash
drought onset.
We examine the persistence of ridging associated with flash
droughts for each region, in the context of the normal or climatological
persistence of these features. To accomplish this, we subset a 10° x 10°
(lat x lon) area to the immediate northwest of each of the 5 regions. We
419
Agricultural and Forest Meteorology 247 (2017) 414–423
Fig. 4. Pentad anomalies of RH, Precip,
P − PET, and VPD prior to flash drought
onset. From left to right the panels show
anomaly composites 1, 2, 3, and 4 pentads
prior to flash drought onset.
then average daily 500 hPa heights within these areas over randomly-
selected 30, 20, 10, and 5 day periods. Within each time period, we
denote the number of days in which the area-average 500 hPa height
anomaly exceeds its climatological 75th percentile. We then bootstrap
this procedure over 10,000 iterations to obtain a distribution of “rid-
ging days” within randomly-selected 30, 20, 10, and 5 day periods.
These distributions are then compared to the average number of “rid-
ging days” in 30, 20, and 10 day periods immediately prior to flash
droughts in each region (Fig. 6). Statistical significance of ridging day
occurrence preceding flash drought is noted if the actual frequency of
ridging days is above the 95th percentile of the distribution of randomly
selected sets. Up to 30 days prior to flash drought onset, the Upper
Midwest, Ohio Valley, and Northeast regions experience ridging days at
significantly elevated rates. The frequency of ridging days both up to
20 days and 10 days prior to flash drought onset is also significantly
higher than expected. The Southeast region exhibits an above-average,
although not significant, frequency of ridging days preceding flash
drought at 30, 20, and 10 days prior to the events. The South region
exhibits a well-below normal frequency of ridging days out to 30-days
prior to flash drought; however out to 10-days prior to the events, the
frequency of ridging days becomes slightly above normal. This is con-
sistent with the lack of persistent ridging in the South region beyond a
10-day lead to flash drought onset (Fig. 5), and could potentially be
attributed to differences in warm season moisture availability in these
regions. For example, the P − PET moisture balance in the South region
is frequently below 0 between March and October, signifying soil
moisture deficit. Additionally, the lag in the establishment of a mid-
level ridge in the South region could indicate a feedback from the soil
moisture desiccation, itself. Large negative (i.e., dry) root zone soil
moisture anomalies have been shown to intensify and prolong mid-to-
upper level ridging and anti-cyclonic conditions in semi-arid regions
(Ferranti and Viterbo, 2006; Zampieri et al., 2009), including the
Southern Great Plains (Oglesby and Erickson, 1989). A potential land
surface-atmosphere feedback combined with precedent soil moisture
deficit in the South region could magnify the impact of ridging circu-
lation features and corresponding surface conditions, desiccating soils
T.W. Ford, C.F. Labosier
to the point of drought faster than in more humid regions. On the other
hand, it is obvious that ridging persistence or the frequency with which
mid-level ridging occurs is strongly related to the onset of flash drought
in the Midwest and Northeast United States.
3.4. Logistic regression
Our results demonstrate that flash drought is forced by mid-level
ridging and corresponding 1) decreased precipitation and humidity and
2) increased solar radiation, maximum temperatures, and overall eva-
porative demand. We use logistic regression to identify the surface
weather variables with which flash drought onset is most strongly re-
lated. Each meteorological variable is separately regressed and the
output is the change in flash drought probability corresponding with a
change in that variable. The resulting flash drought probability changes
are grouped by variable, along with the fraction of stations that exhibit
a statistically significant fit, as determined using Ward’s Method
(Fig. 7). Very few statistically significant relationships appear at 4-
pentad and 3-pentad lead times; P − PET has the most significant with
17% of stations. The results at 2-pentad and 1-pentad lead times,
though, are dramatically different. At the 2-pentad lead time, P − PET,
PET, and VPD have the strongest relationship with the probability of
antecedent flash drought, as a 1-unit reduction in P − PET at a 2-
pentad lead increases this probability by 35%. This relationship be-
tween P − PET and flash drought is significant at over 90% of stations
in the study region. At a 1-pentad lead, logistic regression models for all
variables except TMAX, Wspd, and Precip are significant at over 80% of
stations. These results suggest that variables representing surface
moisture balance (P − PET) and atmospheric demand (PET and VPD)
are better predictors of flash drought than are temperature or pre-
cipitation alone. Additionally, the logistic regression results suggest the
predictive power of these three variables could extend to a 2-pentad
lead time.
420
Agricultural and Forest Meteorology 247 (2017) 414–423
Fig. 5. Anomalies of 500 hPa geopotential
heights (m) prior to flash drought onset.
Composites are grouped by region: (top to
bottom) South, Upper Midwest, Ohio Valley,
Southeast, and Northeast. Height anomalies
are composited over (left to right) 30-, 20-,
10-, and 5-day periods immediately prior to
flash drought onset.
4. Discussion and conclusions
While many conventional droughts evolve over a period of weeks or
months, recent studies suggest that drought can evolve and intensify
quickly (Otkin et al., 2015; Ford et al., 2015; Mo and Lettenmaier 2015,
2016). Such events have been termed “flash droughts” and leave little
time for stakeholders to make adequate preparations, potentially re-
sulting in more detrimental impacts (Otkin et al., 2015) than conven-
tional droughts driven by remote factors, like SST anomalies. Despite
the potential for disproportionate impacts from flash droughts, little
attention has been paid in understanding the nature and characteristics
of flash droughts and their meteorological drivers.
Using meteorological variables from 125 first-order weather sta-
tions across the Eastern United States, a climatology of flash droughts is
developed for the period of March-October 1979–2010. While flash
drought occurrence is overall low, there is spatial and temporal varia-
bility with locations in the Southeast region having an overall higher
frequency than the rest of the Eastern United States, particularly in the
Upper Midwest and South regions. Additionally, locations in the
Northeast and Southeast regions tend to see flash droughts earlier in the
March-October period whereas flash droughts occur later in the Great
Plains and Upper Midwest.
Meteorological conditions associated with flash droughts include
decreased precipitation and relative humidity, along with increases in
solar radiation, maximum temperatures, and atmospheric demand.
Below normal precipitation accumulation over a relatively-long time
period is requisite for any drought; however the combination of pre-
cipitation deficits with large, positive evapotranspiration anomalies are
associated with rapid soil moisture desiccation and the rapid in-
tensification that characterizes flash droughts in the United States
(Otkin et al., 2013). Our results are consistent with those from Otkin
et al. (2013), suggesting that both moisture supply and demand must be
accounted for in order to properly monitor and forecast flash drought.
T.W. Ford, C.F. Labosier
However, we did find that temperature and precipitation were, by
themselves, not strongly associated with conditions prior to flash
drought occurrence. This weak association implies that the two vari-
ables have limited predictive capacity. Instead, variables that accounted
for surface moisture balance and/or atmospheric demand were more
closely linked with conditions leading up to flash droughts. These sur-
face conditions are likely driven by ridging in the mid to upper level
troposphere. Composites of 500 hPa geopotential heights suggest that
ridging to the west and northwest of the regions contributes to the onset
of flash droughts. Furthermore, this ridging pattern may be more per-
sistent in the northern clusters of the study region.
It is important to note that we define only one type or “flavor” of
flash drought. Indeed the soil moisture thresholds of 40th and 20th
percentiles, as well as the necessary 4-pentad soil moisture depletion
period were subjectively determined, based both on previous work with
soil moisture-identified flash droughts (e.g., Ford et al., 2015) and
significant experience working with in situ soil moisture data (e.g., Ford
and Quiring, 2014; Quiring et al., 2016). With that being said, we
421
Agricultural and Forest Meteorology 247 (2017) 414–423
Fig. 6. Boxplots show distribution of the
frequency of “ridging days” within ran-
domly-selected (panel a) 30, (panel b) 20,
and (panel c) 10 day groups. The dashed,
black lines denote the frequency of “ridging
days” in (panel a) 30, (panel b) 20, and
(panel c) 10 day periods leading up to flash
drought occurrence. Results are shown se-
parately for each region.
contend that although these thresholds will naturally vary between
studies, flash drought identification or classification must be driven by
the intensification time and not overall length of prevailing dry con-
ditions. Additionally, the lack of a consistent flash drought definition
complicates monitoring and forecasting of flash drought events, and
represents a major limitation in drought science.
Acknowledgement
We would like to acknowledge the reviewers and associated editor
for their assistance in evaluating this paper.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in the
online version, at http://dx.doi.org/10.1016/j.agrformet.2017.08.031.
T.W. Ford, C.F. Labosier
References
Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop Evapotranspiration-Guidelines
for Computing Crop Water Requirements-FAO Irrigation and Drainage Paper 56 300.
FAO, Rome, pp. D05109.
Anderson, M.C., Hain, C., Otkin, J., Zhan, X., Mo, K., Svoboda, M., Wardlow, B., Pimstein,
A., 2013. An intercomparison of drought indicators based on thermal remote sensing
and NLDAS-2 simulations with U.S. Drought Monitor Classifications. J.
Hydrometeorol. 14, 1035–1056.
Anderson, M.C., Zolin, C.A., Sentelhas, P.C., Hain, C.R., Semmens, K., Tugrul Yilmaz, M.,
Gao, F., Otkin, J.A., Tetrault, R., 2016. The evaporative stress index as an indicator of
agricultural drought in Brazil: an assessment based on crop yield impacts. Rem. Sens.
Env. 174, 82–99.
Cai, J., Liu, Y., Lei, T., Pereira, L.S., 2007. Estimating reference evapotranspiration with
the FAO Penman-Monteith equation using daily weather forecast messages. Ag. For.
Meteorol. 145, 22–35.
Dirmeyer, P.A., Wu, J., Norton, H.E., Dorigo, W.A., Quiring, S.M., Ford, T.W., Santanello
Jr., J.A., Bosilovich, M.G., Ek, M.B., Koster, R.D., Balsamo, G., 2016. Confronting
weather and climate models with observational data from soil moisture networks
over the United States. J. Hydrometeorol. 17, 1049–1067.
Fernando, D.N., Mo, K.C., Fu, R., Pu, B., Bowerman, A., Scanlon, B.R., Solis, R.S., Yin, L.,
422
Agricultural and Forest Meteorology 247 (2017) 414–423
Fig. 7. Boxplots (panel a) show the change
in probability of flash drought onset given a
one-unit increase in each meteorological
variable. Bar plots (panel b) show the per-
cent of all stations in the study region ex-
hibiting a statistically significant logistic
regression fit. Panels show relationships at
1-, 2-, 3-, and 4-pentad lead times.
Mace, R.E., Mioduszewski, J.R., Ren, T., Zhang, K., 2016. What caused the spring
intensification and winter demise of the 2011 drought over Texas? Clim. Dynam. 47,
3077–3090.
Ferranti, L., Viterbo, P., 2006. The European summer of 2003: Sensitivity to soil water
initial conditions. J. Clim. 19, 3659–3680.
Ford, T.W., Quiring, S.M., 2014. Comparison and application of multiple methods of
temporal interpolation of daily soil moisture. Int. J. Climatol. 34, 2604–2621.
Ford, T.W., McRoberts, D.B., Quiring, S.M., Hall, R.E., 2015. On the utility of in situ soil
moisture observations for flash drought early warning in Oklahoma, USA. Geophys.
Res. Lett. 42, 9790–9798.
Hayes, M., Svoboda, M., Wall, N., Widhalm, M., 2011. The Lincoln declaration on drought
indices. Bull. Amer. Met. Soc. 92, 485–488.
Hoerling, M., Kumar, A., Dole, R., Nielsen-Gammon, J.W., Eischeid, J., Perlwitz, J., Quan,
X., Zhang, T., Pegion, P., Chen, M., 2013. Anatomy of an extreme event. J. Clim. 26,
2811–2832.
Hoerling, M., Eischeid, J., Kumar, A., Leung, R., Mariotti, A., Mo, K., Schubert, S., Seager,
R., 2014. Causes and predictability of the 2012 Great Plains drought. Bull. Amer. Met.
Soc. 95, 269–282.
Hunt, E.D., Hubbard, K.G., Wilhite, D.A., Arkebauer, T.J., Dutcher, A.L., 2009. The de-
velopment and evaluation of a soil moisture index. Int. J. Climatol. 29, 747–759.
Hunt, E.D., Svoboda, M., Wardlow, B., Hubbard, K., Hayes, M., Arkebauer, T., 2014.
Monitoring the effects of rapid onset of drought on non-irrigated maize with
T.W. Ford, C.F. Labosier
agronomic data and climate-based drought indices. Ag. For. Meteorol. 191, 1–11.
Kam, J., Sheffield, J., Yuan, X., Wood, E.F., 2014. Did a skillful prediction of sea surface
temperatures help or hinder forecasting of the 2012 Midwestern US drought? Env.
Res. Lett. 9, 034005.
Karl, T.R., Koss, W.J., 1984. Regional and national monthly, seasonal, and annual tem-
perature weighted by area, 1895–2983. Historical Climatology Series 4-3. National
Climatic Data Center Asheville, NC, pp. 38.
Maxwell, E.L., Marion, W., Myers, D., Rymes, M., Wilcox, S., 1995. In: Final Technical
Report, National Solar Radiation Data Base (1961–1990). National Renewable Energy
Laboratory, Golden, Colorado,NREL/TP-463-5784..
McEvoy, D.J., Huntington, J.L., Hobbins, M.T., Wood, A., Morton, C., Anderson, M., Hain,
C., 2016. The evaporative demand drought index. Part II: CONUS-wide assessment
against common drought indicators. J. Hydrometeorol. 17, 1763–1779.
McMahon, T.A., Peel, M.C., Lowe, L., Srikanthan, R., McVicar, T.R., 2013. Estimating
actual, potential, reference crop and pan evaporation using standard meteorological
data: a pragmatic synthesis. Hydrol. Earth Syst. Sci. 17, 1331.
Mesinger, F., DiMego, G., Kalnay, E., Mitchell, K., Shafran, P.C., Ebisuzaki, W., Jovic, D.,
Woollen, J., Rogers, E., Berbery, E.H., Ek, M.B., Fan, Y., Grumbine, R., Higgins, W.,
Li, H., Lin, Y., Manikin, G., Parrish, D., Shi, W., 2006. North american regional re-
analysis. Bull. Amer. Met. Soc. 87, 343–360.
Mo, K.C., Lettenmaier, D.P., 2015. Heat wave flash droughts in decline. Geophys. Res.
Lett. 42, 2823–2829.
Mo, K.C., Lettenmaier, D.P., 2016. Precipitation deficit flash droughts over the United
States. J. Hydrometeorol. 17, 1169–1184.
Monteith, J.L., 1965. Evaporation and environment. Symp. Soc. Exp. Biol. 19, 4.
Oglesby, R.J., Erickson III, D.J., 1989. Soil moisture and the persistence of North
American drought. J. Clim. 2, 1362–1380.
Otkin, J.A., Anderson, M.C., Hain, C., Mladenova, I.E., Basara, J.B., Svoboda, M., 2013.
Examining rapid onset drought development using the thermal infrared-based eva-
porative stress index. J. Hydrometeorol. 14, 1057–1074.
Otkin, J.A., Anderson, M.C., Hain, C., Svoboda, M., 2014. Examining the relationship
between drought development and rapid changes in the evaporative stress index. J.
Hydrometeorol. 15, 938–956.
Otkin, J.A., Shafer, M., Svoboda, M., Wardlow, B., Anderson, M.C., Hain, C., Basara, J.,
2015. Facilitating the use of drought early warning information through interactions
with agricultural stakeholders. Bull. Amer. Met. Soc. 96, 1073–1078.
Otkin, J.A., Anderson, M.C., Hain, C., Svoboda, M., Johnson, D., Mueller, R., Tadesse, T.,
Wardlow, B., Brown, J., 2016. Assessing the evolution of soil moisture and vegetation
conditions during the 2012 United States flash drought. Ag. For. Meteorol. 218–219,
230–242.
Peel, M.C., McMahon, T.A., 2014. Estimating evaporation based on standard meteor-
ological data −progress since 2007. Prog. Phys. Geog. 38, 241–250.
423
Agricultural and Forest Meteorology 247 (2017) 414–423
Peng, C., Lee, K.L., Ingersoll, G.M., 2002. An introduction to logistic regression analysis
and reporting. J. Educ. Res. 96, 3–14.
Quiring, S.M., Ford, T.W., Wang, J.K., Khong, A., Harris, E., Lindgren, T., Goldberg, D.W.,
Li, Z., 2016. The North American soil moisture database: development and applica-
tions. Bull. Amer. Met. Soc. 97, 1441–1459.
Schubert, S.D., Suarez, M.J., Pegion, P.J., Koster, R.D., Bacmeister, J.T., 2004. Causes of
long-term drought in the US Great Plains. J. Clim. 17, 485–503.
Senay, G.B., Buddle, M.B., Brown, J.F., Verdin, J.P., 2008. Mapping flash drought in the
US: southern great plains. In: 22nd Conference on Hydrology. AMS, New Orleans,
Louisiana.
Sentelhas, P.C., Gillespie, T.J., Santos, E.A., 2010. Evaluation of FAO Penman-Monteith
and alternative methods for estimating reference evapotranspiration with missing
data in Southern Ontario, Canada. Ag. Wat. Man. 97, 635–644.
Svoboda, M., LeComte, D., Hayes, M., Heim, R., Gleason, K., Angel, J., Rippey, B., Tinker,
R., Palecki, M., Stooksbury, D., Miskus, D., Stephens, S., 2002. The drought monitor.
Bull. Am. Met. Soc. 83, 1181–1190.
Wilcox, S., 2007. In: National Solar Radiation Database 1991–2005 Update: User’s
Manual National Renewable Energy Laboratory (NREL). Golden, Colorado. (NREL/
TP-581-41364).
Xia, Y., Mitchell, K., Ek, M., Sheffield, J., Cosgrove, B., Wood, E., Luo, L., Alonge, C., Wei,
H., Meng, J., Livneh, B., Lettenmaier, D., Koren, V., Duan, Q., Mo, K., Fan, Y., Mocko,
D., 2012a. Continental-scale water and energy flux analysis and validation for the
North American Land Data Assimilation System project phase 2 (NLDAS-2): 1.
Intercomparison and application of model products. J. Geophys. Res. 117, D3.
Xia, Y., Mitchell, K., Ek, M., Cosgrove, B., Sheffield, J., Luo, L., Alonger, C., Wei, H., Meng,
J., Livneh, B., Duan, Q., Lohmann, D., 2012b. Continental-scale water and energy flux
analysis and validation for the North American Land Data Assimilation System pro-
ject phase 2 (NLDAS-2): 2. Validation of model-simulated streamflow. J. Geophys.
Res. 117, D3.
Xia, Y., Sheffield, J., Ek, M.B., Dong, J., Chaney, N., Wei, H., Meng, J., Wood, E.F., 2014.
Evaluation of multi-model simulated soil moisture in NLDAS-2. J. Hydrol. 512,
107–125.
Xia, Y., Ek, M.B., Wu, Y., Ford, T.W., Quiring, S.M., 2015. Comparison of NLDAS-2 si-
mulated and NASMD observed daily soil moisture. Part I: comparison and analysis. J.
Hydrometeorol. 16, 1962–1980.
Yuan, S., Quiring, S.M., 2014. Drought in the US Great Plains 1980–2012: A sensitivity
study using different methods for estimating potential evapotranspiration in the
Palmer Drought Severity Index. J. Geophys. Res 119. http://dx.doi.org/10.1002/
2014JD021970.
Zampieri, M., D’Andrea, F., Vautard, R., Ciais, P., de Noblet-Ducoudré, N., Yiou, P., 2009.
Hot European summers and the role of soil moisture in the propagation of
Mediterranean drought. J. Clim. 22, 4747–4758.