You are on page 1of 14

HYDROLOGICAL PROCESSES

Hydrol. Process. 25, 27712784 (2011)


Published online 15 March 2011 in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/hyp.8040
A review of advances in ash ood forecasting
H. A. P. Hapuarachchi,* Q. J. Wang and T. C. Pagano
CSIRO Land & Water, Graham Road, Highett, Victoria 3190, Australia
Abstract:
Flash ooding is one of the most hazardous natural events, and it is frequently responsible for loss of life and severe damage
to infrastructure and the environment. Research into the use of new modelling techniques and data types in ash ood
forecasting has increased over the past decade, and this paper presents a review of recent advances that have emerged from
this research. In particular, we focus on the use of quantitative precipitation estimates and forecasts, the use of remotely
sensed data in hydrological modelling, developments in forecasting models and techniques, and uncertainty estimates. Over
the past decade ash ood forecast lead-time has expanded up to six hours due to improved rainfall forecasts. However the
largest source of uncertainty of ash ood forecasts remains unknown future precipitation. An increased number of physically
based hydrological models have been developed and used for ash ood forecasting and they have been found to give more
plausible results when compared with the results of conceptual, statistical, and neural network models. Among the three
methods for deciding ash ood occurrence discussed in this review, the rainfall comparison method (ash ood guidance)
is most commonly used for ash ood forecasting as it is easily understood by the general public. Unfortunately, no existing
model is capable of making reliable ash ood forecasts in urban watersheds even though the incidence of urban ash ooding
is increasing due to increasing urbanisation. Copyright 2011 John Wiley & Sons, Ltd.
KEY WORDS ash oods; advances; forecasts; remote sensing; hydrological models; quantitative precipitation forecasts
Received 10 February 2010; Accepted 28 January 2011
INTRODUCTION
Flash oods can be caused by excessive rainfall or by the
sudden release of water due to a dam breach or glacier
lake outburst. However, as sudden-release events are
uncommon and occur under extraordinary circumstances
that require considerations outside the scope of this
review, only rainfall-induced ash oods are considered
here.
In general, ash oods are characterized by their rapid
onset (within six hours of rainfall), which leaves very
limited opportunity for effective response. They are often
accompanied by other phenomena such as landslides and
mudows, and can cause bridge collapses, damage to
buildings and businesses, psychological harm to people
and, in exceptional circumstances, fatalities. Projections
of climate change indicate an increase in the intensity
of rainfall in some parts of the world, which may lead
to more severe ash ooding. In addition, changing
demography (i.e. increased urbanisation) will result in
larger segments of the population being prone to ash
ooding.
In order to identify the occurrence of ash ooding,
estimate the risk and implement effective mitigation mea-
sures, high spatial resolution ash ood forecasting with
useful lead time is necessary. The starting point in effec-
tive ash ood forecasting is accurate rainfall forecasts
with useful lead times. The next step is the representation
* Correspondence to: H. A. P. Hapuarachchi, CSIRO Land & Water,
Graham Road, Highett, Victoria 3190, Australia.
E-mail: Prasantha.Hapuarachchi@csiro.au
of the hydrologic and hydraulic processes within a catch-
ment that determine how rainfall-runoff accumulates. The
static physical properties of a catchment (e.g. permeabil-
ity, relief, fraction of impervious areas, land use, and
soil types) and its time-varying states (e.g. soil moisture,
ground water decit) will modulate the ash ood poten-
tial of heavy rainfall (Davis, 2001). Usually, ash oods
occur in streams and small catchments with a drainage
area of a few hundred square kilometres (Kelsch, 2001),
with Davis (1998) suggesting that for the United States
(US), this cut-off threshold is about 260 km
2
(100 mi
2
).
Such catchments often respond rapidly to intense rain-
fall rates because of steep slopes, saturated soils, re-
induced alterations to natural drainage, and impermeable
surfaces due to soil crusting or man-made structures (i.e.
pavement). It becomes necessary, therefore, to exam-
ine the rainfall-to-runoff conversion process in headwater
catchments. However, these small catchments are often
poorly gauged or ungauged, which presents additional
challenges in hydrological modelling.
Over the past decade, there have been increasing
calls to improve ash ood forecasts in many parts
of the world, including the US, the European Union,
and Australia (Penning-Rowsell et al., 2000; Handmer,
2001). In response, advances in ash ood forecast-
ing have been achieved through a range of develop-
ments in observing capabilities and modelling techniques.
This review rst discusses these developments, empha-
sising recent improvements in quantitative precipitation
estimates (QPEs) and quantitative precipitation forecasts
(QPFs), remotely sensed data products, and ow forecast
Copyright 2011 John Wiley & Sons, Ltd.
2772 H. A. P. HAPUARACHCHI, Q. J. WANG AND T. C. PAGANO
models, and their use for ash ood forecasting. This is
followed by a focus on uncertainty estimates and urban
ash oods and the article is nished with conclusions.
INPUT DATA
Quantitative precipitation estimates (QPEs)
The quality of any ood forecast depends to a high
degree on the quality of the rainfall input. During the
past decade, observation networks (e.g. radar and satel-
lite) have been expanded, and new techniques have been
developed for deriving rainfall from multi-sensor obser-
vations. Extensive research has gone into assimilating
radar data for producing QPEs. In general, radar pro-
vides useful information on the spatial distribution of
the precipitation eld, but QPEs directly derived from
radar reectivity measurements are subject to errors and
uncertainties (e.g. Collier, 1996; German and Joss, 2003).
New techniques have been developed for real-time cor-
rection of systematic biases in radar precipitation elds,
by adjusting radar precipitation to rain gauge measure-
ments (Ahnert et al., 1986; Smith and Krajewski, 1991;
Seo, 1998; Anagnostou and Krajewski, 1999; Sinclair and
Pegram, 2005; Mazzetti and Todini, 2009).
With the availability of high-resolution remotely
sensed data, advanced algorithms for retrieving rain-
fall from satellite-based microwave and infrared obser-
vations have been developed (Sorooshian et al., 2000;
Huffman et al., 2002; Kidd et al., 2003; Joyce et al.,
2004; Kubota et al., 2007). Consequently, a number of
general-use satellite-based precipitation products have
emerged (e.g. CMORPH, 3B42RT, PERSIANN, GSMaP)
that allow much improved temporal and spatial resolu-
tions and reduced latency. Typically, satellite-based pre-
cipitation products provide QPEs of the global area of
60

N60

S at 01

025

spatial resolution at 13 h
intervals. Although the accuracy of QPEs varies, they
are generally useful for hydrometeorological modelling,
particularly in poorly gauged catchments. Moreover,
advanced techniques have been developed to improve
the accuracy of QPEs by blending multiple sources of
information (radar, satellite and gauged data) (Seo and
Breidenbach, 2002; Gjertsen et al., 2004), thus expanding
the limits of hydrological modelling.
Burton and OConnell (2002) used radar observation
elds to continuously update a function to convert from
infrared (IR) satellite observations of cloud-top temper-
atures to precipitation rates. The technique was adapted
from the raw histogram-matching technique developed by
Turk et al. (2000) and described by Grose et al. (2002).
Instead of using microwave observations of precipitation
from satellite to determine xed relationships between
IR satellite observations and precipitation rates, Burton
and OConnell (2002) used ground-based precipitation
radar datasets to estimate an instantaneous relationship.
Overall, QPEs produced by blending multiple sources
of information have high temporal and spatial accuracy,
thus, this approach deserves future research.
Quantitative precipitation forecasts (QPFs)
Accuracy and lead time in rainfall prediction are the
most important components of ash ood forecasting,
and recent developments enable the production of high-
resolution QPFs with 16 h lead times. These techniques
include linear regression (Antolik, 2000), quantile regres-
sion (Bremnes, 2004; Friederichs and Hense, 2007),
logistic regression (Applequist et al., 2002; Hamill et al.,
2004), hierarchical models based on prior climatic distri-
butions (Krzysztofowicz and Maranzano, 2006), hybrid
approaches that statistically combine radar and numerical
weather prediction (NWP) model outputs (Golding, 2000;
Ganguly and Bras, 2003; Sokol, 2006), articial neu-
ral network applications (Hsu et al., 1997; Kuligowski
and Barros, 2001; Ramirez et al., 2005), and statistical
methods based on Bayesian techniques (Sloughter et al.,
2007).
The spatial (<10 km) and temporal (<1 h) resolu-
tions of NWP model rainfall forecasts have signicantly
improved in the past few years (e.g. ALADIN, SKIRON).
With these developments, the most promising approach
for developing QPFs with useful lead times appears to
be combining NWP model forecasts with blends of the
advected patterns of recent radar, satellite and gauged
rainfall data. In the merged product, the contribution of
the NWP model forecasts usually increases with increas-
ing lead time. The operational Nimrod system (Golding,
1998) in the UK Met Ofce, designed for 16 h QPF,
combines radar advection with NWP-based QPF using
relative weights (Smith and Austin, 2000).
Recently developed ensemble rainfall forecasting tech-
niques have demonstrated encouraging results that sug-
gest their accuracy for lead times of 16 h is sufcient
to improve forecasts of ash oods (Kondragunta and
Seo, 2004; Yuan et al., 2007; Rezacova et al., 2007).
Pierce et al. (2004) presented a rst-step approach to
probabilistic forecasting by generating an ensemble of
radar rainfall forecasts from a stochastic advection-based
scheme. This approach allows the probability of precipi-
tation to be derived from an ensemble of forecasts for
several hours ahead. It has been further improved by
Bowler et al. (2006) as the Short-Term Ensemble Pre-
diction System (STEPS), which merges an extrapolated
nowcast with downscaled NWP model forecasts.
The ensembles can be used to generate forecasts of
the probability density function of areal and temporal
averages of precipitation. However, accurate model fore-
casts of higher rain rates are limited to a 6-h lead time.
Ebert (2001) discussed a different technique to produce
ensemble rainfall forecasts, called Poor Mans Ensem-
ble (PME) in which the components are obtained from
independent NWP model forecasts from several different
operational forecasting centres. The PME then forecasts
the probability and distribution of rainfall in the short
term (12 days), although the accuracy of the forecasts
rapidly declines with increasing lead time. To date, PME
has been applied on coarse (i.e. 100 km) resolution mod-
els. Nonetheless, these improved QPF techniques have
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
A REVIEW OF ADVANCES IN FLASH FLOOD FORECASTING 2773
increased the potential for more accuracy and lead time
in ash ood forecasting.
Remotely sensed catchment data
Over the years, developments in observational tech-
niques have led to improvements in high-quality remotely
sensed data for better dening catchment characteristics.
Different sensors have been used to gather information
about properties at the surface and in shallow layers of
the Earth. New techniques have been developed to utilize
such remotely sensed information for deriving physi-
cal characteristics of a catchment, such as near-surface
soil moisture, land surface temperature, snow cover and
melt, topographical characteristics, land use, and vegeta-
tion cover. This data is useful for hydrometeorological
modelling, particularly for poorly gauged and ungauged
catchments.
Soil moisture content is an important variable in
hydrological modelling, but it is hard to measure on a
large scale via in situ measurements. However, advances
in remote sensing techniques using microwave satel-
lite observations have enabled the estimation of surface
(10 cm depth) soil moisture easily (Owe et al., 2001;
Meesters et al., 2005). A number of studies have demon-
strated fairly good agreement between soil moisture
derived from microwave-based models and eld observa-
tions (Owe et al., 1992; Drusch et al., 2001; Jackson and
Hsu, 2001) as well between microwave-based models and
distributed hydrological modelling approaches (Vischel
et al., 2008). More recently, a historical climatology of
continuous satellite-derived global land surface soil mois-
ture data has been developed (Owe et al., 2008), which
consists of surface soil moisture retrievals derived from
all available historical and active satellite microwave sen-
sors from 1978 to the present. The accuracy of the derived
soil moisture data was tested using the observed data
from the Global Soil Moisture Data Bank (Robock et al.,
2000). The comparisons between satellite and ground
data appear quite good, considering the considerable dif-
ferences in spatial coverage and the vertical sampling
characteristics between the two datasets.
A variety of models and methods have recently
been developed that are capable of processing remotely
sensed data for estimating evapotranspiration (Basti-
aanssen et al., 1998; Richard et al., 2005; Chavez et al.,
2009). The surface temperature is derived from thermal
infrared and passive microwave data, and utilized to esti-
mate actual evapotranspiration (Coll and Caselles, 1997).
However, there is no clear consensus at present as to the
accuracy of evapotranspiration estimates derived using
remotely sensed data, and a proper validation of the
methods is required before the data could be accepted.
Satellite observations have also been used to monitor the
areal extent of snow cover and glacier inventories. Steven
(2006) examined the feasibility of the Ensemble Kalman
Filtering (EnKF) data assimilation approach for snow-
pack characterisation using multi-spectral remote sensing
observations. He developed a framework for estimating
the snow water equivalent using remotely sensed data.
The primary ndings in this study are that the assimila-
tion approach is capable of providing estimates of SWE
in cases with deep snow, wet snow, and light to moderate
vegetation cover. These are three of the primary con-
founding problems with traditional snow retrieval algo-
rithms. Moreover, satellite images are being increasingly
used for deriving ood inundation extent, and for esti-
mating lake and reservoir volume.
All these data are useful for hydrological modelling
and consequently for ash ood forecasting, particu-
larly in data-poor regions. However, it should be noted
that current remote sensing capabilities cannot replace
ground-based methods for providing high-quality prole
data at a given point. Also the operational delay due
to processing time of satellite-based observations (>2 h)
is not favourable for real-time forecasting. The present
advantage of remote sensing is in mapping conditions at
the large scale (e.g. regional, continental and even global)
and on a repetitive basis to detect possible changes.
FLOW FORECASTING MODELS
Data-driven models
Data-driven (e.g. neural network, statistical) models
use statistical relationships derived from rainfall and
river ow data to generate ow forecasts. Generally,
these models perform better than others in situations
where the underlying interactions and dependencies of
physical processes are only partially understood or are
unknown. These models are easy to set up and are able
to produce acceptable results with minimum input data
(rainfall and discharge). Owing to this simplicity, data-
driven modelsparticularly neural networks (NN)have
been widely used for ow forecasting (e.g. Thirumalaiah
and Deo, 1998; Jain and Srinivasulu, 2004; Piotrowski
et al., 2006; Sahoo et al., 2006).
Using four types of NN models (one-hidden-layer
back-propagation NN, two-hidden-layer back-pro-
pagation NN, radial basis function network, and Fuzzy
Inference System Network), Sahoo and Ray (2006)
demonstrated that an optimized two-hidden-layer back-
propagation NN model outperforms the other three in
terms of prediction performance efciency (coefcient
of correlation, root mean square error and absolute mean
error). Recently developed NN models incorporate mul-
tiple sources of information to produce accurate ow
forecasts with useful lead times. Chiang et al. (2007)
introduced a recurrent NN model that merges multi-
ple rainfall sources (gauged and satellite-based data) for
ow forecasting. They found that satellite-derived pre-
cipitation (PERSIANN) (Hong et al., 2004) had limited
contribution (5%) to the merging procedure in their par-
ticular study. This implies the importance of gauged data,
although the contribution from gauged precipitation to
the merging procedure depends on the number of gauges
and the quality of data. Nevertheless, the study exam-
ined a potential method for extending satellite-derived
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
2774 H. A. P. HAPUARACHCHI, Q. J. WANG AND T. C. PAGANO
precipitation to those catchments where gauge observa-
tions are limited. A more complicated forecast method
that uses NWP model forecasts, radiosonde data (direc-
tional wind and pressure levels) and rain gauge data was
proposed by Kim and Barros (2001). The special fea-
ture of this method is that inputs include satellite-derived
characteristics of storm systems such as tropical cyclones,
meso-scale convective complex systems and convective
cloud clusters. Forecasts are primarily derived using a
particular NN model, which is selected based on the type
of convective weather system. Although it is claimed that
this method could produce accurate ow forecasts with
lead times up to 24 h, it requires experienced manual
intervention in operational use. Categorising convective
weather systems is labour intensive and can be subjective.
In general, the main disadvantage of using data-driven
models for ash ood forecasting is that they require
long-term data records to train or calibrate. Furthermore,
the derived relationships are site specic. Therefore, data-
driven models are hard to apply to general ash ood
forecasting, since ash oods usually occur in small
catchments (<300 km
2
) where gauged data is often poor
or unavailable.
Lumped and distributed hydrological models
Many of the hydrological models developed before the
1990s have been adequately discussed (e.g. Singh, 1995),
and therefore, only a few recently developed models
and their applicability to ash ood forecasting are dis-
cussed. Until recently, lumped hydrological models were
primarily used for ow forecasting; however, spatially
distributed hydrological models are gaining in popular-
ity, in part, due to the increased availability of new data.
Nevertheless, lumped models remain a reliable opera-
tional tool. Some operational ood forecasting systems
use lumped models due to their simplicity, computational
efciency and lower data requirements. For example, the
Sacramento soil moisture accounting model (SAC-SMA)
(Burnash et al., 1973) is used by the US National Weather
Services (NWS) for producing ash ood guidance, and
Kobold and Brilly (2006) recently demonstrated the sat-
isfactory use of the HBV model (Bergstrom, 1976) in
ash ood forecasting for a mountainous catchment on
an hourly timestep.
Sirdas and Sen (2007) presented a new lumped
approach that combines isohyetal rainfall maps, kine-
matic wave and rational methods to produce synthetic
ash ood hydrographs for arid regions. This empirical
approach relates ood peaks to the average temporal and
areal intensities of storm rainfalls. Based on the applica-
tions results, the authors concluded that the reciprocal
of rainfall intensity is proportional to its duration, and
that the proportionality is constant; the time of concen-
tration of oods is also a function of catchment size,
slope, geology, soil types, and topography. Foody et al.
(2004) introduced a model incorporating the Soil Con-
servation Services empirical concept of hill slopes, and
a Muskingum scheme for river routing. Its key parame-
ters are derived from conventional topographic data, eld
surveys and land-cover maps. Acceptable results were
obtained when the model was used to predict sites at risk
from large peak ows associated with ash ooding in
the Wadi catchment in the eastern desert of Egypt.
In general, the usefulness of lumped hydrological
models for ash ood forecasting is limited by their
coarse resolution, the need for long-term historical data
for calibration, and inapplicability to poorly gauged
catchments. To overcome these limitations, some lumped
models have been improved upon and are now considered
as semi-distributed or distributed hydrological models.
For example, the SAC-SMA model was modied to
become the distributed HL-RDHM model (Koren et al.,
2004); the TOPKAPI model (Todini, 1995) was also
improved by Ciarapica and Todini (2002) to become a
distributed model. Simple event-based models have also
been applied in a semi-distributed manner, such as the
URBS model (Carroll, 1992) in Australia. These modied
models provide similar or slightly better performance
compared to lumped versions, and are capable of better
utilising distributed rainfall input and representing a
catchments physical characteristics.
A number of physically based distributed hydrologi-
cal models of high temporal and spatial resolution have
been developed in the recent past. El-Hames and Richards
(1998) introduced a physically based model for predict-
ing ash foods and associated transmission losses in arid
regions. The model is based on coupling numerical solu-
tions of the Saint Venant equations for channel ow rout-
ing and the kinematic wave equation for overland ow
routing, with the Richards equation for the determination
of inltration. To assess its robustness, the model was
applied to a medium-size catchment in an arid region,
and the results indicated that good simulations can be
achieved with minimum calibration.
Borrell et al. (2006) introduced a distributed hydro-
logical model called MARINE (Model of Anticipation
of Runoff and INundations for Extreme events) and,
after testing for ash ood modelling of the Orbieu
River catchment in France, the authors concluded that
the model produces realistic simulations. It is particu-
larly well suited for real-time forecasting as it requires
relatively short computation time, and model outputs
can be formulated in accordance with the forecasters
needs. More recently, England et al. (2007) investigated
the applicability of the TREX (Two-dimensional, Runoff,
Erosion and Export) model to simulate extreme oods
in semi-arid regions in the western US, and concluded
that the model could generate good results with lim-
ited data. Other recently developed physically based dis-
tributed hydrological models with potential in ash ood
modelling include LISFLOOD (Roo et al., 2000), tRIBS
(Ivanov et al., 2004) and BTOP (Takeuchi et al., 2008).
In general, the spatial scale used in physically based
distributed hydrological modelling is well suited for ash
ood forecasting. As most of the model parameters are
physically based, long historical records are not required
for parameter calibration, and there is more potential
for physically based models to be successfully applied
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
A REVIEW OF ADVANCES IN FLASH FLOOD FORECASTING 2775
to poorly gauged catchments. Recent work by Moore
et al. (2006) suggests that for extreme events, the dis-
tributed models may give more plausible results than the
lumped models. However, physically based distributed
hydrological models are computationally inefcient and
need high quality catchment data (e.g. DEM, land-use
and soil maps, and soil characteristics). Thus, it remains
unclear whether the use of more complex models for ash
ood forecasting has much advantage over the use of
simple models.
The suitability of a model to a particular applica-
tion depends on factors such as data availability, pro-
cess complexity, temporal and spatial scales and the
required outputs of the application (e.g. forecast vari-
ables such as river ow, soil moisture, and inundation).
Most hydrological models perform well in humid zones,
but not in arid and semi-arid zones due to highly com-
plex hydro-meteorological processes. Generally, greater
spatial and temporal variations of rainfall occur in the
arid zones than occur in more humid zones. Pilgrim
et al. (1988) describes distinctive features of arid zone
hydrology which may be different from humid zones. He
states that in arid zones, there is a mix of hydrological
processes; some humid zone processes (e.g. baseow)
are essentially absent while channel transmission losses
are critically important. Usually arid and semi-arid zones
are highly dynamic, and a prolonged wet or dry sequence
may change the character of the hydrology (and hence
calibrated model parameters). The vegetation cover is
sparse and the spatial extent changes largely with the sea-
son in arid and semi-arid zones. Lack of organic matter in
arid and semi-arid zones probably has a signicant effect
on many processes including interception, inltration,
evapotranspiration, and runoff. In these zones, surface
soil largely is the rst point of contact by rainfall. Thus,
surface soil properties likely play a major role in runoff
production, especially as soil saturation occurs relatively
rarely compared to humid zones. Instead, inltration-
excess runoff is a more dominant process during heavy
rainfall events. Hydrophobic soils, armouring, dispersive
soils, cracks, scald, or claypan areas, sand dunes and bare
surface rock are some of the features which are inuen-
tial in arid zone runoff production (Pilgrim et al., 1988).
Most of the above mentioned physical catchment charac-
teristics can be considered more realistically using physi-
cally based distributed hydrological models than lumped
or statistical models; thus, physically based hydrologi-
cal models are more suitable for arid and semi-arid zone
hydrological modelling (El-Hames and Richards, 1998).
Furthermore, a comprehensive method for ow routing
is needed for arid and semi-arid zones, such as those that
employ a detailed solution of the St Venant equations,
coupled with an inltration model to deal with transmis-
sion losses.
CRITERIA FOR DECIDING FLASH FLOOD
OCCURANCE
Estimating ooding ow
Flooding is generally associated with damaging con-
ditions due to the rising and overowing of a body of
water onto normally dry land, and it is often site specic
and difcult to quantify. Floods can also be associated
with exceedance of stormwater capacity in the urban
environment. However, urban ooding is extremely com-
plex to model due to interactions with various man-made
infrastructures such as buildings, roads, culverts, chan-
nels, tunnels, and underground structures. Therefore, a
simple approach often used to dene thresholds for ood-
ing based on overbank-ow is described herein.
The bank-full ow is assumed as the starting point
of ooding ow (Q
p
). However, more than bank-full
ow is usually needed to cause ood damage. Assuming
a wide rectangular channel cross-section, the bank-full
ow Q
bf
is usually computed from channel geometry
and roughness characteristics using Mannings steady,
uniform ow resistance formula (Chow et al., 1988):
Q
p
D Q
bf
D bh
5/3
S
05
c
/n 1
where b is the channel width (m) at bank-full, h is the
hydraulic depth at bank-full (m), S
c
is the local channel
slope (dimensionless), and n is the Mannings rough-
ness coefcient. The computation of bank-full discharge
requires information on channel cross-sectional parame-
ters, but measured data for these parameters is often not
available for small streams. However, some of the param-
eters can be derived using other available data such as
digital elevation maps (DEMs), land-use maps and soil
characteristics. For example, S
c
can be derived from a
DEM, and n can be assumed based on river bank (land
use) and river bed (soil) characteristics. Regional relation-
ships between the stream cross-sectional parameters and
other catchment and stream characteristics (e.g. catch-
ment area, stream length) can be used for identifying b
and h.
Carpenter et al. (1999) derived parameters b and h
through regional regression relationships using a power-
law relationship with the predictor variables of A (km
2
),
S
c
, stream length L (m), and average annual rainfall P
(mm). They found that both b and h are related to A in
the form:
b D
1
A

1
2
h D
2
A

2
3
where
k
and
k
(k D 1, 2) are coefcients. In two
regions of the US (Iowa and Oklahoma) the relationship
between b and A showed a higher correlation (8291%)
than the relationship between h and A (4050%).
An alternative statistically based denition of ooding
ow is the ow of a certain return period, which considers
the risk and uncertainty associated with ooding. Hender-
son (1966) showed that there is a good statistical relation-
ship between the bank-full ow and a ow with a return
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
2776 H. A. P. HAPUARACHCHI, Q. J. WANG AND T. C. PAGANO
b = a
1
A
b
1
h = a
2
A
b
2
Derive regional regression
equations for b and h
Estimate bank-full flow at
a particular location
Derive regional regression
equations for Q
2
, Q
3
,
Q
4
,Q
n
Compute Q
2
, Q
3
,Q
4
,Q
n
at a particular location
Long-term rainfall data and
adequate discharge data
available to calibrate a
hydrological model
Generate long-term river flow
data using a hydrological
model
Compare with available flood
flow data and local information
Q
bf
= bh
5/3
S
0.5
/ n
Q
2
= a
3
A
b
3
S
c
l
c
River cross-sectional
information available at
bank-full condition at
some locations
Adequate observed
discharge data available
at some locations to
deriveQ
2
, Q
3
, Q
4
,Q
n
Figure 1. Different methods to estimate bank-full ow at a data-poor location, depending on available data
period of between one and two years. Many researchers
have since used this range to assume bank-full ow (e.g.
see discussion in Riggs, 1990). For instance, two-year
return period ow (Q
2
) can be calculated using histor-
ical observed ow records. In the absence of observed
ow data, a hydrological model-simulated ow series
can be used (Reed et al., 2007). Carpenter et al. (1999)
conducted a regression analysis (similar to the above) to
derive Q
2
based on regional data (A, S
c
and L). They
found that Q
2
is related to A and S
c
with 85% correlation
coefcient in the form:
Q
2
D
3
A

3
S

c
4
where
3
,
3
and are coefcients. However, deriv-
ing bank-full ow at each grid-cell of a catchment is
challenging, since a river cross-section may vary greatly
over short distances and may also change in time with
the occurrence of oods. Therefore, assuming Q
2
as the
bank-full ow at every grid-cell is questionable. Figure 1
presents a ow chart for estimating bank-full ow at a
data-poor location of a catchment, depending on available
data.
Flow comparison method
Given a forecast ow at a particular point in a catch-
ment (such as the catchment outlet), there needs to
be a criterion for deciding whether ooding should be
expected. The simplest approach is to compare the mod-
elled ow value with the observed ooding threshold.
An alternative statistical-distributed modelling approach
is to compare the modelled ow value with the long-term
simulated record (i.e. climatology) of the model (Reed
et al., 2007).
Figure 2 shows the ow chart of the statistical-
distributed modelling approach for ash ood
forecasting. In the historical mode (Figure 2), a grid-
based distributed hydrological model (calibrated) is run
using archived multisensory QPE data to generate a long
term ow series at each grid cell in a particular catch-
ment. Basing on the statistical characteristics derived
from the long-term simulated ow data, a unique ow
frequency curve is developed for each cell using a ow
frequency analysis method.
In the real-time mode, for any forecast period (e.g. 12,
24 h), the model produces many grids of ow forecasts
(e.g. hourly) of the catchment. Using these ow grids,
a grid containing peak ow forecast at each cell is
prepared for a given forecast time. The grid of peak
ow forecasts is then converted to a ow frequency grid
based on the statistical characteristics of the frequency
curves derived in the historical mode. Finally, the ow
frequency forecasts at each cell is compared with a
particular threshold ow frequency value, determined
from local characteristics, to assess the level of severity
of the event and then to identify ash ood risk areas
where warnings need be issued. The rationale for this is
that model simulations can be biased and a comparison
Flow
Frequency
curves
Frequency
thresholds
Local/regional
knowledge
Simulated
historical
peaks
Grid-based
hydrological
model
Archived
QPE
Historical
Real-time
QPE/QPF
Max
forecast
peaks
Statistical
Post-processor
Forecast
frequencies
Initial hydro
model states
Grid-based
hydrological
model
Real-time
Compare
Figure 2. Flow chart of the statistical-distributed modelling approach for
ash ood forecasting (from Reed et al., 2007)
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
A REVIEW OF ADVANCES IN FLASH FLOOD FORECASTING 2777
Figure 3. Simplied representation of the ash ood guidance system and the main sources of uncertainty considered (Ntelekos et al., 2006). Gray
fonts mark the variables for which uncertainty was quantied and thick dark gray arrows mark the propagation of uncertainty throughout the system.
WSR-88D is radar data of the Next-Generation Weather Radar (NEXRAD) network. SAC-SMA is the Sacramento soil moisture accounting model.
Thresh-R is threshold runoff
with bank-full ow in exceedance frequency diminishes
the effect of the bias. However, this approach requires
long historical data to run simulations to establish both
the simulated and observed ow frequency distributions.
Rainfall comparison method
Instead of comparing the ow forecast with ooding
ow, one can compare the rainfall required over a specic
area to produce ooding ow at its outlet with the rainfall
forecast. This method is commonly known as ash ood
guidance (FFG), and is dened as the volume of rainfall
of a given duration and distributed uniformly over a small
catchment (<300 km
2
) that is just enough to cause bank-
full ow at the catchment outlet. The estimation of FFG
requires the determination of the amount of effective
rainfall, or threshold runoff (R
T
), which is dened as
R
T
D
Q
p
q
pR
A
5
where q
pR
is the unit hydrograph peak for a specic
duration t
R
normalized by catchment area (m
2
s
1
/ km
2
).
The value of q
pR
can be determined using a unit
hydrograph method (Rodiguez-Iturbe et al., 1982). To
determine the FFG for a particular duration (e.g. 3, 6 h),
a hydrological model is run with increasing amounts of
accumulated rainfall using the same initial catchment
conditions (e.g. soil moisture, groundwater store). The
runoff produced is plotted against the required rainfall,
and FFG can be determined for a known threshold
runoff value using this plot Georgakakos (2006). As the
guidance is in terms of rainfall, it is easily understood
by the general public, and it allows decision-makers to
consider risks in different time-frames (e.g. 3, 6, 12 h).
However the following limitations are recognized:
FFG is provided as a lumped value across a given
catchment and as uniform rainfall over the period.
This is a signicant disadvantage, as the effects of
spatial rainfall distribution over a catchment cannot be
properly accounted for, and consequently the ash ood
risk at critical areas inside a catchment could be missed;
There is uncertainty in R
T
values. R
T
values can have
signicant local variations due to differences in the
catchment and stream morphology (Carpenter et al.,
1999). Determining a R
T
value using 12-year ood
frequency values is unrealistic;
The FFG method does not address areas where land
characteristics and rainfall intensity inuence the occur-
rence of ash ooding more than soil moisture. In such
areas, ash ooding is intensity driven and is inu-
enced more by terrain, land cover, soil type, geology,
and land-use characteristics than soil moisture.
FFG was developed in the US during the 1970s, 1980s,
and 1990s (e.g. Mogil et al., 1978; Georgakakos, 1987;
Sweeney, 1992), and the more recent development of
FFG in conjunction with a distributed hydrological model
has led to gridded ash ood guidance (GFFG) (Smith
et al., 2000). GFFG better represents the distribution of
the degree of soil saturation inside a catchment, however,
the nal FFG associated with a particular ow point is
still a lumped valuethe required rainfall is assumed
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
2778 H. A. P. HAPUARACHCHI, Q. J. WANG AND T. C. PAGANO
to be uniformly distributed over an area above the ow
point to produce bank-full ow.
Martina et al. (2006) introduced a ood forecasting
method for given river sections, based on the direct
comparison of QPF with critical rainfall threshold values,
without the need for an online real-time forecasting
system. The estimation of critical rainfall thresholds
requires derivation of a joint probability function of
rainfall totals over the contributing area and water stages
(or discharges) at a relevant river section. To derive
the probability function, a long series (10 000 years)
of hourly average rainfall data is generated using a
stochastic model, and used to force a hydrological model
to generate soil moisture and water stage data. Soil
is categorized into three antecedent moisture classes
(AMCs) as dry, moderately saturated, and wet (based
on 033 and 066 percentiles), and the corresponding
rainfall and water stage is sorted into the three AMCs
for determining joint probability density functions. The
most appropriate rainfall threshold is determined based
on the minimum expected cost under uncertainty using a
Bayesian utility function. This approach is simple and can
be used by non-technical operators. Its main limitation is
that rainfall forecast uncertainty is not considered.
Flash ood susceptibility assessment procedure
While more complex hydrometeorological models are
used for quantifying ash oods, a simple ash ood
susceptibility assessment can be considered as a useful
rst step in determining the contributing factors to the
ash ood vulnerability of a catchment using limited
data. It is particularly useful in an urban context. Collier
and Fox (2003) proposed a simple decision-support
procedure for assessing the vulnerability of catchments to
ooding due to extreme rainfall. The procedure considers
the following:
1. The likelihood that a heavy rain event will become
stationary and protracted over the area
2. The availability of signicant precipitable water in the
lower atmosphere
3. The likelihood that heavy cells embedded in the main
area will move in parallel to the main watercourse, in
which case, the ood peak is likely to be enhanced
4. The steepness of the catchment leading to a short time
to peak
5. The soil moisture condition of the catchment
6. The likelihood of unimpeded ow to the main water-
course: whether or not signicant vegetation and chan-
nel debris are likely to be problematic, and whether or
not there are constrictions in the channel that will facil-
itate a build-up of water for later release as a wave
(surge)
7. Snowmelt
The variables listed above include both hydrometeoro-
logical and catchment morphological characteristics. In
an assessment, these variables are identied and each
variable is assessed on a scale of 14, where a score
of 4 strongly favours ood development. The individual
scores of all variables are summed, and the ood category
is determined based on the overall score. These variables
need to be continuously updated in real time. Collier and
Fox (2003) found the procedure to be a useful indicator
of ash oods in tests on a number of major ood events.
One of the challenges with this procedure is assessing
whether a precipitation system will become stationary.
In practice, this can be accomplished by integrating real-
time radar reectivity data, which indicates the general
direction of cell tracks and cell splitting throughout an
event.
Dale et al. (2004) found that the main weakness of the
procedure is that all variables have equal weights in the
score, yet naturally some variables are more inuential
than others in ash ood formation. For example, the
rainfall depth over a relatively short period on a steep
catchment may result in an extreme event. However,
low scores for factors such as urban extent, snow depth
and direction of motion result in a lowering of the
nal score to a value considered as major rather than
extreme. Therefore, appropriate weights to different
variables need to be assigned carefully through the
experience gained by many trials. The procedure could be
improved signicantly by splitting the xed, antecedent
and real-time variables, and overlaying the risk levels
(Dale et al., 2004).
UNCERTAINTY ESTIMATES
Flash ood forecasts are necessarily uncertain due to
input data errors, and modelling errors. Understanding
of the uncertainty from all these sources is particularly
important for decision making in ood warning.
Historically, probability theory has been the primary
tool for representing uncertainty in mathematical models
(Ross, 1995). With the introduction of fuzzy set the-
ory (Zadeh, 1965) uncertainty could be represented using
non-probabilistic approaches. Zadeh (1978) developed a
broader framework for uncertainty representation called
possibility theory, which is also known as a fuzzy mea-
sure. Basing on these theories, various methodologies
have been developed for the treatment of uncertainty
in ood forecasting, such as the Generalised Likelihood
Uncertainty Estimation (GLUE) by Beven and Binley
(1992), the Bayesian Forecasting System by Krzyszto-
fowicz (1999) and a methodology based on fuzzy exten-
sion principle by Maskey et al. (2004).
Krzysztofowicz (2002) developed a Bayesian system
for probabilistic river stage forecasting. In this system,
short-term probabilistic river stage forecasts were pro-
duced using probabilistic QPFs as input to a determin-
istic hydrological model. The overall system includes a
precipitation uncertainty processor, a hydrologic model
uncertainty processor, and an integrator. The system is
computationally efcient and capable of quantifying the
total uncertainty that exists at the forecast time. Notably,
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
A REVIEW OF ADVANCES IN FLASH FLOOD FORECASTING 2779
Figure 4. Probabilistic stage forecasting with 95% condence interval (from Chen and Yu, 2007)
the general methodology developed in this study can be
applied to any hydrological model. One of the weak-
nesses of this method is that it considers only the total
precipitation amount as the dominant source of uncer-
tainty. The assumption of a single dominant source of
uncertainty is not universally valid.
A rst step towards the development of a proba-
bilistic FFG forecast method (Figure 3) was presented
by Ntelekos et al. (2006). They considered uncertainty
in the threshold-runoff calculation methodology, in the
hydrologic models parameters, and initial states. Dif-
ferent sources of uncertainty were propagated to study
their joint effect on the FFG System (Figure 3). Their
results conrmed that ash ood guidance estimates are
most uncertain under dry initial soil moisture conditions.
The uncertainty of threshold-runoff dominated the out-
come of the numerical experiments, while the hydrologic
model uncertainty had limited inuence on FFG results.
The threshold-runoff is particularly sensitive to stream-
top width, hydraulic depth, the drainage area, and the
length of the main stream; a decrease in their uncertainty
levels would automatically translate to a decrease to the
overall threshold-runoff uncertainty.
Chen and Yu (2007) proposed a different type of proba-
bilistic river stage forecast method that consists of a deter-
ministic forecast derived from support vector regression
and a probability distribution of forecast error based on
fuzzy inference. This method can account for total uncer-
tainty involved in the forecasting process. The results
indicated that for short lead time (13 h) forecasts, the
predicted condence intervals were smooth and their
amplitudes were small, thus useful for practical appli-
cations. However, for forecasts with lead-times greater
than 3 h, the condence intervals exhibited undesirable
uctuations (Figure 4).
Yatheendradas et al. (2008) investigated the predictive
uncertainty of a physically based distributed hydrological
model KINEROS2 (Smith et al., 1995) driven by high-
resolution radar rainfall input for ash ood forecasting
in a semi-arid catchment. They considered rainfall esti-
mates, model parameters, and initial moisture conditions
as the sources of uncertainty. The GLUE approach and
variance-based Sobol global sensitivity analysis method
(Sobol, 1993), which accounts for factor interactions at
all orders, was used for uncertainty analysis. Follow-
ing are the main conclusions drawn from the results
(Yatheendradas et al., 2008).
1. The uncertainty in the radar rainfall estimates almost
completely dominated the uncertainty in the modelled
response
2. Initial hillslope soil moisture signicantly inuenced
the modelled response
3. The predictive uncertainty in the modelled ash ood
response is often likely to be much higher than what
would be considered acceptable for accurate ash ood
forecasting
In general, good uncertainty estimates of ash ood
forecasts can add credibility to the forecast system. How-
ever the development of methodologies capable of incor-
porating forecast uncertainty into the decision-making
process (whether to issue ood warning or not) remains a
major challenge. Even the simple assignment of appropri-
ate exceedance thresholds presents problems: a too low
threshold would lead to a high number of false alarms;
conversely, a too high threshold would lead to a low
probability of detection and a greater number of surprises.
In either circumstance, miscalculation of risk could have
considerable social and economic consequences.
URBAN FLASH FLOOD FORECASTING
Urban areas typically have high risks of ash ooding
due to the presence of large impervious areas, block-
ing of stormwater ow or insufcient drainage capacity,
and sometimes location (vicinity of a river). Blockage of
canals and stormwater drainage systems by debris and
sediment may signicantly increase the risk of ooding.
During an urban ash ood event, streets can become
fast-moving rivers that can cause damage to infrastruc-
ture, while water-lled basements and viaducts can cause
fatalities. Some structures such as highways, railway
lines, large buildings, concrete walls, and pavements can
act as temporary embankments that may trap water for
several days.
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
2780 H. A. P. HAPUARACHCHI, Q. J. WANG AND T. C. PAGANO
Flash flood
(FF) forecast
Rainfall
Comparison
method
Flow comparison
method
FF Guidance
Direct flow
comparison
FF susceptibility
assessment
procedure
Flow frequency
comparison
Flow forecast
model
Forcing Data
Rainfall forecasts
Remotely
sensed data
Historical
climatology
Geomorphologic
data
Neural Networks
Statistical
models
Conceptual
models
Distributed
physically-based
models
Figure 5. Different approaches available for making a ash ood forecast
Increasing urbanisation due to population growth,
combined with potential climate change impacts, means
that there is an increased risk of more frequent and
severe urban ash oods. As land is converted from
elds or woodlands to residential areas, roads, and
parking lots, its capacity to absorb rainfall is diminished.
On average, urbanisation results in a two- to six-fold
increase in runoff compared to what would occur on
natural terrain (Ramachandra and Kumar, 2008). In
addition, the high population densities and related socio-
economic infrastructure in urban areas usually leads to
more signicant damage from ash ood events than that
associated with ooding in non-urban areas.
Modelling urban ash oods is particularly difcult
due to the absence of natural ow paths and the pres-
ence of man-made structures. Snell and Gregory (2002)
applied a ash ood forecast model for rural and urban
catchments in New Mexico, using information on the
road networks to alter the ow direction map derived
from a DEM alone, since streets serve as major drainage
conduits in urban catchments. Flow direction cells that
did not correspond to the street drainage network were
edited to correspond to the drainage network. However,
the results indicated that additional work is necessary to
improve the estimates of runoff and discharge. A recent
study of ash ood forecasting in the Dead Run water-
shed of Baltimore County in Maryland, USA (Javier
et al., 2007), conrmed that a major limitation on the
accuracy of ash ood forecasting in urban areas is
imposed by stormwater management infrastructure. The
model analyses also suggest that there is potential for
improving model forecasts through the use of information
on initial soil moisture storage. The authors further com-
mented that distributed hydrological models and high-
resolution radar rainfall can provide important elements
of site-specic ash ood forecasting systems in small
urban watersheds.
Very high-resolution DEMs such as LiDAR (Light
Detection And Ranging) (Lin, 1997) are useful for
urban ash ood modelling. Haile and Rientjes (2005)
examined the effects of LiDAR DEM resolution in
ood modelling, and concluded that it has a signicant
effect on inundation extent, ow velocity, ow depth,
and ow patterns across the model domain. Therefore,
the spatial resolution of ash ood forecasts in urban
areas requires special attention. Though urban ash ood
modelling is very important, no existing model is capable
of addressing all these issues to reliably forecast ash
oods in urban catchments.
CONCLUSIONS
This review discusses recent advances in ash ood
forecasting, including new developments in QPEs, QPFs,
remotely sensed data, ow forecast models and methods,
forecast uncertainty estimates, and ash ood forecasts
in the urban context. Summary of different approaches
discussed in this review for ash ood forecasting is
shown in Figure 5. The main conclusions drawn from
this review are listed below.
1. Effective ash ood forecasting with useful lead times
is one of the most challenging areas in hydrology,
particularly due to the uncertainties associated with
rainfall forecasts. Figure 6(a) shows the current qual-
ity versus the importance of QPF, QPE, and remotely
sensed data. Quality means the ability of the informa-
tion source to portray the actual scenario. Importance
reects the contribution of this product towards over-
all ash ood forecast accuracy, if the product reached
its full potential. Improvements are being made in
remotely sensed data retrieval techniques (radar- and
satellite-based) and in merging rainfall data derived
from different sources (gauged, radar, satellite, NWP
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
A REVIEW OF ADVANCES IN FLASH FLOOD FORECASTING 2781
I
m
p
o
r
t
a
n
c
e
Current quality
(a) Data
QPF
RSD
QPE
T
u
n
i
n
g

n
e
e
d
e
d
Ease of use
(b) Flow forecasting models
PDM
LHM
DDM
RCM
FCM
FSA
R
e
l
i
a
b
i
l
i
t
y
Operational simplicity
(c) Criteria for deciding
flash flood occurrence
LHM = Lumped hydrological models
FSA = Flash flood susceptibility assessment procedure
QPF = Quantitative precipitation forecasts
RSD = Remotely sensed data and deriving techniques
QPE = Quantitative precipitation forecasts
PDM = Physically-based distributed hydrological models
RCM = Rainfall comparison methods
FCM = Flow comparison methods
DDM = Data driven models
Figure 6. Current states of data, ow forecasting models, and criteria for deciding ash ood occurrence used for ash ood forecasting
model forecasts). These improvements have facili-
tated the production of sufciently accurate QPFs with
16 h lead-times and high quality QPEs. However fur-
ther improvements in lead-time (1224 h) and accu-
racy of QPFs is essential (Figure 6(a)) for providing
meaningful ash ood forecasts. Observed informa-
tion such as rain gauge, radar, and satellite-based data
are only partially useful for producing QPFs, whereas,
high-resolution-quality NWP products have the poten-
tial to improve the quality of QPFs signicantly
2. Flash oods usually occur in small catchments
(<300 km
2
) that are often poorly gauged or ungauged,
thus, remotely sensed data is particularly useful for
hydrological modelling in such catchments. Remote
sensing techniques have signicantly improved and
have been used for estimating precipitation, evapo-
transpiration, snow cover and melt, soil moisture, land
use and its change, ood inundation, etc. Some stud-
ies have demonstrated fairly good agreement between
remotely sensed data and ground observations. How-
ever, remotely sensed data such as soil moisture, snow
cover and melt, and evapotranspiration are rarely used
for operational ash ood forecasting. This is partially
because of their poor quality and inadequate resolu-
tion for hydrological applications. Therefore, improv-
ing the quality of remotely sensed data is essential
(Figure 6(a)) for general ash ood forecasting
3. Advances have been made in the development of ow
forecast models for ash ood modelling. In the recent
past, more physically based distributed hydrological
models (PDMs) that can be used for ash ood fore-
casting have been developed. The spatial scale used
in PDMs is well suited for ash ood forecasting and
for extreme events; the distributed models seem to give
plausible results (Moore et al., 2006). As most of PDM
parameters are physically based, long historical records
are not required for parameter calibration (as shown
in Figure 6(b)), and there is more potential for PDMs
to be successfully applied to poorly gauged catch-
ments. However PDMs need high-resolution sophis-
ticated input data, hence, applicability is limited. In
general, applicability of LHMs and DDMs
(Figure 6(b)) for ash ood forecasting is limited as
most of ash oods occur in small catchments that are
usually poorly gauged, hindering model tuning efforts
4. Three methods available for deciding ash ood occur-
rence discussed in this review are the ow comparison
method, rainfall comparison method, and ash ood
susceptibility assessment method (Figure 5). The rain-
fall comparison (RCM) method (e.g. ash ood guid-
ance) is a good tool to warn of an imminent ash
ood. RCM is commonly used for ash ood fore-
casting as it is easily understood by the general public
(Figure 6(c)). It allows decision makers to consider
risks in different time frames (e.g. 3, 6, 12 h). How-
ever, RCM poorly conveys the magnitude of the ash
ooding. On the other hand, ow comparison methods
(FCM, Figure 6(c)) combined with a physically based
distributed hydrological model, if properly calibrated
and with high-quality QPF, can potentially successfully
predict specic crest stage and ow for small catch-
ments. In particular, runoff can be modelled on the
same scale as convective storms, an important driver
of ash oods. However, such a system is hard to
operationalize. In comparison, the Flash Flood Sus-
ceptibility Assessment (FSA) procedure (Figure 6(c))
is operationally simple although forecasts might be less
reliable because the method is essentially a heuristic
5. Uncertainty estimates are important for decision mak-
ing in ash ood warning. During the past two decades,
new methods have been introduced to estimate the
forecast uncertainty (e.g. Beven and Binley, 1992;
Krzysztofowicz, 1999; Maskey et al., 2004) and for
making probabilistic forecasts (e.g. Krzysztofowicz,
2002; Ntelekos et al., 2006; Chen and Yu, 2007). The
largest source of uncertainty in quantitative ash ood
forecasting is errors in the rainfall eld. One of the
major challenge is to develop methodologies capable
of incorporating forecast uncertainty into the decision-
making process (whether to issue ood warning or not)
6. Urban areas are highly vulnerable to ash oods,
mainly due to the existence of large impervious areas.
Increasing urbanisation due to population growth, com-
bined with potential climate change impacts, implies
an increased risk of more frequent and severe urban
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
2782 H. A. P. HAPUARACHCHI, Q. J. WANG AND T. C. PAGANO
ash oods in the future. Thus, ash ood forecast-
ing in urban areas needs special attention in order to
reduce severe losses. Unfortunately, no existing model
is capable of making reliable ash ood forecasts in
urban watersheds.
ACKNOWLEDGEMENTS
The authors thank Jim Elliott, Soori Sooriyakumaran, and
their colleagues at the Bureau of Meteorology, Australia,
for their support and guidance on this review. Cathy
Bowditch of CSIRO is gratefully acknowledged for edit-
ing this paper. We thank the two anonymous reviewers
for their valuable comments on this manuscript. Permis-
sion for gure use (Figure 2) came from Seann Reed,
Alexandros Ntelekos (Figure 3), and Shien-Tsung Chen
(Figure 4). This review was undertaken as part of the
Water Information Research and Development Alliance, a
joint initiative of the CSIRO Water for a Healthy Country
Flagship and the Bureau of Meteorologys Water Divi-
sion. The funding support from the CSIRO OCE Science
Leader Scheme is also gratefully acknowledged.
REFERENCES
Ahnert P, Krajewski W, Johnson E. 1986. Kalman lter estimation
of radar-rainfall eld bias. In Preprints of 23d Conf. on Radar
Meteorology, Snowmass, CO, American Meteorological Society,
pp. 3337.
Anagnostou EN, Krajewski WF. 1999. Real-time radar rainfall estima-
tion: Part IICase study. Journal of Atmospheric and Oceanic Tech-
nology 16: 198205.
Antolik MS. 2000. An overview of the National Weather Services
centralized statistical quantitative precipitation forecasts. Journal of
Hydrology 239: 306337.
Applequist S, Gahrs GE, Pfeffer RL, Niu X. 2002. Comparison of
methodologies for probabilistic quantitative precipitation forecasting.
Weather and Forecasting 17(4): 783799.
Bastiaanssen WGM, Menenti M, Feddes RA, Holtslag AAM. 1998. A
remote sensing surface energy balance algorithm for land (SEBAL):
1Formulation. Journal of Hydrology 212: 198212.
Bergstrom S. 1976. Development and application of a conceptual runoff
model for Scandinavian catchments. Bulletin Series A 52: University
of Lund.
Beven KJ, Binley A. 1992. The future of distributed models: model
calibration and uncertainty prediction. Hydrological Processes 6:
279298.
Borrell VE, Dartus D, Ababou R. 2006. Flash ood modelling with the
MARINE hydrological distributed model. Hydrology and Earth System
Sciences 3: 33973438.
Bowler NE, Pierce CE, Seed AW. 2006. STEPS: A probabilistic
precipitation forecasting scheme which merges an extrapolation
nowcast with downscaled NWP. Quarterly Journal of the Royal
Meteorological Society 132: 21272155.
Bremnes JB. 2004. Probabilistic forecasts of precipitation in terms of
quantiles using NWP model output. Monthly Weather Review 132:
338347.
Burnash RJC, Ferral RL, McGuire RA. 1973. A generalized stream ow
simulation system. In Concerned Modelling for Digital Computers,
Joint Federal State River Forecast Centre: Sacramento, California,
USA.
Burton A, OConnell PE. 2002. Report on the performance of the
new nowcasting methodology. Deliverable 54. MUSIC Project,
http://www.geomin.unibo.it/hydro/music/index2.htm.
Carpenter TM, Sperfslage JA, Georgakakos KP, Sweeney T, Fread DL.
1999. National threshold runoff estimation utilizing GIS in support of
operational ash ood warning systems. Journal of Hydrology 224:
2144.
Carroll DG. 1992. URBS: The Urbanized Catchment Runoff Routing
Model . Report prepared for Brisbane City Council, Queensland,
Australia.
Chavez JL, Gowda PH, Howell TA, Copeland KS. 2009. Radiometric
surface temperature calibration effects on satellite based evapotran-
spiration estimation. International Journal of Remote Sensing 30(9):
23372354.
Chen ST, Yu PS. 2007. Real-time probabilistic forecasting of ood
stages. Journal of Hydrology 340: 6377.
Chiang Y, Hsu K, Chang F, Hong Y, Sorooshian S. 2007. Merging
multiple precipitation sources for ash ood forecasting. Journal of
Hydrology 340: 183196.
Chow VT, Maidment DR, Mays LW. 1988. Applied Hydrology.
McGraw-Hill: New York, 572.
Ciarapica L, Todini E. 2002. TOPKAPI: a model for the representation of
the rainfall-runoff process at different scales. Hydrological Processes
16: 207229.
Coll C, Caselles V. 1997. A split-window algorithm for land surface
temperature from AVHRR data: validation and algorithm comparison.
Journal of Geophysical Resources 102: 1669716712.
Collier CG. 1996. Applications of Weather Radar Systems: A Guide to
Uses of Radar Data in mMeteorology and Hydrology. 2nd edn. John
Wiley & Sons: New Jersey; 390.
Collier CG, Fox NI. 2003. Assessing the ooding susceptibility of river
catchments to extreme rainfall in the United Kingdom. International
Journal of River Basin Management 1: 111.
Dale M, Dempsey P, Dent J. 2004. Extreme Rainfall Event Recognition:
Phase 2 Work Package 5Establishing a User Requirement for a
Decision-Support Tool. Research and Development Technical Report
FD2208 of Defra/Environment Agency Flood and Coastal Defence
R&D Programme. Department of the Environment, Food and Rural
Affairs: London, UK.
Davis RS. 1998. Detecting time duration of rainfall: a controlling factor
of ash ood intensity. In Proceedings of Special Symposium on
Hydrology, Phoenix, AZ, pp. 258263.
Davis RS. 2001. Flash ood forecast and detection methods: severe
convective storms. AMS Meteorological Monograph Series 28(50):
481525.
Drusch, M, Wood EF, Jackson TJ. 2001. Vegetative and atmospheric
corrections for the soil moisture retrieval from passive microwave
remote sensing data: results from the Southern Great Plains Hydrology
Experiment 1997. Journal of Hydrometeorology 2: 181192.
Ebert EE. 2001. Ability of a Poor Mans Ensemble to predict the
probability and distribution of precipitation. Monthly Weather Review
129: 24612480.
El-Hames AS, Richards KS. 1998. An integrated, physically-based model
for arid region ash ood prediction capable of simulating dynamic
transmission loss. Hydrological Processes 12: 12191232.
England JF, Velleux ML, Julien PY. 2007. Two-dimensional simulations
of extreme oods on a large watershed. Journal of Hydrology 347:
229241.
Foody GM, Ghoneim EM, Arnell NW. 2004. Predicting locations
sensitive to ash ooding in an arid environment. Journal of Hydrology
292: 4858.
Friederichs P, Hense A. 2007. Statistical downscaling of extreme
precipitation events using censored quantile regression. Monthly
Weather Review 135: 23652378.
Ganguly AR, Bras RL. 2003. Distributed quantitative precipitation
forecasting using information from radar and numerical weather
prediction models. Journal of Hydrometeorology 4: 11681180.
Georgakakos KP. 1987. Real-time ash ood prediction. Journal of
Geophysical Research 92: 96159629.
Georgakakos KP. 2006. Analytical results for operational ash ood
guidance. Journal of Hydrology 317: 81103.
German U, Joss J. 2003. Operational measurement of precipitation in
mountainous terrain. In Weather Radar: Principles and Advanced
Applications, Meischner P (ed). Springer Verlag: 5276.
Gjertsen U, Salek M, Michelson D. 2004. Gauge adjustment of radar-
based precipitation estimates in Europe. In Proceedings of European
conference on radar in meteorology and hydrology (ERAD), Visby,
Sweden, pp. 711, http://publications.copernicus.org/other publica
tions/proceedings.html.
Golding BW. 1998. Nimrod: a system generating automatic very short-
range forecasts. Meteorological Applications 5: 116.
Golding BW. 2000. Quantitative precipitation forecasting in the UK.
Journal of Hydrology 239: 286305.
Grose AME, Smith EA, Chung H, Ou M, Sohn B, Turk FJ. 2002.
Possibilities and limitations for quantitative precipitation forecasts
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
A REVIEW OF ADVANCES IN FLASH FLOOD FORECASTING 2783
using nowcasting methods with infrared geosynchronous satellite
imagery. Journal of Applied Meteorology 41(7): 763785.
Haile AT, Rientjes THM. 2005. Effects of LiDAR DEM resolution in
ood modelling: a model sensitivity study for the city of Tegucigalpa,
Honduras. In Proceedings of ISPRS WG III/3, III/4, V/3 Workshop on
Laser Scanning 2005, Enschede: The Netherlands; 1214.
Hamill TM, Whitaker JS, Wei X. 2004. Ensemble reforecasting:
improving medium-range forecast skill using retrospective forecasts.
Monthly Weather Review 132: 14341447.
Handmer J. 2001. Improving ood warnings in Europe: a research and
policy agenda. Global Environmental Change Part B: Environmental
Hazards 3(1): 1928.
Henderson FM. 1966. Open Channel Flow. MacMillan: New York; 522.
Hong Y, Hsu KL, Sorooshian S, Gao XG. 2004. Precipitation estimation
from remotely sensed imagery using an articial neural network
cloud classication system. Journal of Applied Meteorology 43(12):
18341852.
Hsu KL, Gao X, Sorooshian S, Gupta HV. 1997. Precipitation estimation
from remotely sensed information using articial neural networks.
Journal of Applied Meteorology 36: 11761190.
Huffman GJ, Adler RF, Stocker EF, Bolvin DT, Nelkin EJ. 2002. A
TRMM-based system for real-time quasi-global merged precipitation
estimates. In Proceedings of TRMM International Science Conference,
Honolulu, 2226 July, NASA/TM-2002- 211605, USA, pp. 7.
Ivanov VY, Vivoni ER, Bras RL, Entekhabi D. 2004. Catchment
hydrologic response with a fully distributed triangulated irreg-
ular network model. Water Resources Research 40: W11102.
DOI:10.1029/2004WR003218.
Jackson TJ, Hsu AY. 2001. Soil moisture and TRMM microwave
imager relationships in the Southern Great Plains 1999 (SGP99)
Experiment. IEEE Transactions on Geoscience and Remote Sensing
39(8): 16321642.
Jain A, Srinivasulu S. 2004. Development of effective and efcient
rainfall-runoff models using integration of deterministic, real-coded
genetic algorithms and articial neural network techniques. Water
Resources Research 40: W04302. DOI:10.1029/2003WR002355.
Javier JRN, Smith JA, Meierdiercks KL, Baeck ML, Miller AJ. 2007.
Flash ood forecasting for small urban watersheds in the Baltimore
metropolitan region. Weather and Forecasting 22: 13311344.
Joyce RJ, Janowiak JE, Arkin PA, Xie P. 2004. CMORPH: a method
that produces global precipitation estimates from passive microwave
and infrared data at high spatial and temporal resolution. Journal of
Hydrometeorology 5: 487503.
Kelsch M. 2001. Hydro-meteorological characteristics of ash oods.
In Coping with Flash Floods, NATO Science Series, Environmental
Security, Gruntfest E, Handmer J (eds). Kluwer Academic Publishers:
Dordrecht, Boston; 77: 181193.
Kidd C, Kniveton DR, Todd MC, Bellerby TJ. 2003. Satellite rainfall
estimation using combined passive microwave and infrared algorithms.
Journal of Hydrometeorology 4: 10881104.
Kim G, Barros AP. 2001. Quantitative ood forecasting using
multisensor data and neural networks. Journal of Hydrology 246:
4562.
Kobold M, Brilly M. 2006. The use of HBV model for ash ood
forecasting. Natural Hazards and Earth System Sciences 6: 407417.
Kondragunta CR, Seo DJ. 2004. Toward integration of satellite
precipitation estimates into the multisensor precipitation estimator
algorithm. In Preprints 19th Conference on Hydrology. American
Meteorological Society: Seattle.
Koren VI, Reed S, Smith M, Zhang Z, Seo DJ. 2004. Hydrology
laboratory research modelling system (HL-RMS) of the US National
Weather Service. Journal of Hydrology 291: 297318.
Krzysztofowicz R. 1999. Bayesian theory of probabilistic forecasting
via deterministic hydrologic model. Water Resources Research 35(9):
27392750.
Krzysztofowicz R. 2002. Bayesian system for probabilistic river stage
forecasting. Journal of Hydrology 268: 1640.
Krzysztofowicz R, Maranzano CJ. 2006. Bayesian Processor of Output
for Probabilistic Quantitative Precipitation Forecasts. Working paper,
Department of Systems Engineering and Department of Statistics,
University of Virginia.
Kubota T, Shige S, Hashizume H, Aonashi K, Takahashi N, Seto
S, Hirose M, Takayabu YN, Nakagawa K, Iwanami K, Ushio T,
Kachi M, Okamoto K. 2007. Global precipitation map using satellite-
borne microwave radiometers by the GSMaP Project: production and
validation. IEEE Transactions on Geoscience and Remote Sensing
45(7): 22592275.
Kuligowski RJ, Barros AP. 2001. Blending multi-resolution satellite data
with application to the initialization of an orographic precipitation
model. Journal of Applied Meteorology 40: 15921606.
Lin CS. 1997. Waveform sampling LiDAR application in complex
terrain. International Journal of Remote Sensing 18(10): 20872104.
Martina MLV, Todini E, Libralon A. 2006. A Bayesian decision
approach to rainfall thresholds based ood warning. Hydrology and
Earth System Sciences 10: 413426.
Maskey S. 2004. Modelling Uncertainty in Flood Forecasting Systems.
PhD Thesis. A.A. Balkema Publisher: The Netherlands.
Mazzetti C, Todini E. 2009. Combining weather radar and raingauge data
for hydrologic applications. In Flood Risk Management: Research and
Practice, Samuels P, Huntington S, Allsop W, Harrop J (eds) Taylor
& Francis Group: London; ISBN 978-0-415-48507-4.
Meesters AGCA, Dejeu RAM, Owe M. 2005. Analytical derivation of
the vegetation optical depth from the microwave polarization difference
index. IEEE Transactions on Geoscience and Remote Sensing 2(2):
121123.
Mogil HM, Monro JC, Groper HS. 1978. NWSs ash ood warning and
disaster preparedness programs. Bulletin of American Meteorological
Society 59: 690699.
Moore RJ, Cole SJ, Bell VA, Jones DA. 2006. Issues in ood forecasting:
ungauged basins, extreme oods and uncertainty. In Frontiers in
Flood Research, IAHS Publication No. 305. International Association
of Hydrological Sciences Press, Centre for Ecology and Hydrology:
Wallingford, UK; pp. 103122.
Ntelekos AA, Georgakakos KP, Krajewski WF. 2006. On the Uncertain-
ties of Flash Flood Guidance: Toward Probabilistic Forecasting of Flash
Floods. Journal of Hydrometeorology 7: 896915.
Owe M, Dejeu R, Holmes T. 2008. Multisensor historical climatology of
satellite-derived global land surface moisture. Journal of Geophysical
Research 113: F01002. DOI:10.1029/2007JF000769.
Owe M, Dejeu R, Walker J. 2001. A methodology for surface soil
moisture and vegetation optical depth retrieval using the microwave
polarization difference index. IEEE Transactions on Geoscience and
Remote Sensing 39(8): 16431654.
Owe M, Griend Van De AA, Chang ATC. 1992. Surface moisture and
satellite microwave observations in semiarid southern Africa. Water
Resources Research 28(3): 829839.
Penning-Rowsell EC, Tunstall SM, Tapsell SM, Parker DJ. 2000. The
benets of ood warnings: real but elusive, and politically signicant.
Journal of Chartered Institution of Water and Environmental Engineers
14: 714.
Pierce C, Bowler N, Seed A, Jones A, Jones D, Moore R. 2004. Use of
a stochastic precipitation nowcast scheme for uvial ood forecasting
and warning. In Proceedings of 6th International Symposium on
Hydrological Applications of Weather Radar, Melbourne, Australia.
Pilgrim DH, Chapman TG, Doran DG. 1988. Problems of rainfall-runoff
modelling in arid and semiarid regions. Hydrological Sciences Journal
33(4): 379400.
Piotrowski A, Napiorkowski JJ, Rowinski PM. 2006. Flash-ood fore-
casting by means of neural networks and nearest neighbour
approacha comparative study. Nonlinear Processes in Geophysics
13: 443448.
Ramachandra TV, Kumar U. 2008. Wetlands of Greater Bangalore, India:
automatic delineation through pattern classiers. Electronic Green
Journal 26: Spring http://1441693203/energy/water/paper/P25 1/
index.htm, ISSN 10767975.
Ramirez MCV, Velho HFDC, Ferreira NJ. 2005. Articial neural
network technique for rainfall forecasting applied to the Sao Paulo
region. Journal of Hydrology 301: 146162.
Reed S, Schaake J, Zhang Z. 2007. A distributed hydrologic model
and threshold frequency-based method for ash ood forecasting at
ungauged locations. Journal of Hydrology 337: 402420.
Rezacova D, Sokol Z, Pesice P. 2007. A radar-based verication of
precipitation forecast for local convective storms. Atmospheric
Research 83: 211224.
Richard GA, Masahiro T, Anthony M, Ricardo T. 2005. Satellite-based
evapotranspiration by energy balance for Western States water
management. In Proceedings of 2005 World Water and Environmental
Resources Congress, Anchorage, Alaska, 1519 May, Raymond W.
(ed). ASCE Publications. DOI:10.1061/40792(173)556.
Riggs HC. 1990. Estimating ow characteristics at ungauged sites.
In Regionalization in Hydrology, Beran MA, Brilly M, Becker A,
Bonacci O (eds). IAHS Publication 191: Wallingford, UK; pp.
150161.
Robock A, Konstantin YV, Srinivasan G, Entin JK, Hollinger SE,
Speranskaya NA, Liu S, Namkhai A. 2000. The global soil moisture
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)
2784 H. A. P. HAPUARACHCHI, Q. J. WANG AND T. C. PAGANO
data bank. Bulletin of American Meteorological Society 81:
12811299.
Rodiguez-Iturbe I, Gonzalez-Sanabria M, Bras RL. 1982. A geomorpho-
climatic theory of the instantaneous unit hydrograph. Water Resources
Research 18(4): 886887.
Roo APJD, Wesseling CG, Deursen WPAV. 2000. Physically based river
basin modelling within a GIS: the LISFLOOD model. Hydrological
Processes 14: 19811992.
Ross TJ. 1995. Fuzzy Logic with Engineering Applications. McGraw-Hill,
Inc.: USA.
Sahoo GB, Ray C. 2006. Flow forecasting for a Hawaii stream using
rating curves and neural networks. Journal of Hydrology 317: 6380.
Sahoo GB, Ray C, De Carlo EH. 2006. Use of neural network to predict
ash ood and attendant water qualities of a mountainous stream on
Oahu, Hawaii. Journal of Hydrology 327: 525538.
Seo DJ. 1998. Real-time estimation of rainfall elds using radar rainfall
and rain gage data. Journal of Hydrology 208: 3752.
Seo DJ, Breidenbach JP. 2002. Real-time correction of spatially non-
uniform bias in radar rainfall data using rain gauge measurements.
Journal of Hydrometeorology 3: 93111.
Sinclair S, Pegram G. 2005. Combining radar and rain gauge rainfall
estimates using conditional merging. Atmospheric Science Letters 6:
1922.
Singh VP. 1995. Computer Models of Watershed Hydrology. Water
Resources Publications: Highlands Ranch, Colorado.
Sirdas S, Sen Z. 2007. Determination of ash oods in Western Arabian
Peninsula. Journal of Hydrologic Engineering 12(6): 676681.
Sloughter JM, Raftery AE, Gneiting T, Fraley C. 2007. Probabilistic
quantitative precipitation forecasting using Bayesian model averaging.
Monthly Weather Review 135: 32093220.
Smith JA, Krajewski WF. 1991. Estimation of the mean eld bias of
radar rainfall estimates. Journal of Applied Meteorology 30: 397412.
Smith KT, Austin GL. 2000. Nowcasting precipitation: a proposal for a
way forward. Journal of Hydrology 239: 3445.
Smith RE, Goodrich DC, Woolhiser DA, Unkrich CL. 1995.
KINEROS2A KINematic Runoff and EROSion Model. Computer
Models of Watershed Hydrology. Singh VP (ed). Water Resources
Publication: Highlands Ranch, Colorado.
Snell S, Gregory K. 2002. A Flash Flood Prediction Model for Rural and
Urban Basins in New Mexico. Technical report, New Mexico Water
Resources Research Institute. http://wrri.nmsu.edu/publish/techrpt/
tr321/tr321.pdf.
Smith SB, Filiaggi MT, Churma M, Roee J, Glaudemans M, Erb R,
Xin L. 2000. Flash ood monitoring and prediction in AWIPS Build 5
and beyond. In Preprints of 15th Conference on Hydrology, American
Meteorological Society: Long Beach, CA.
Sobol IM. 1993. Sensitivity estimates for nonlinear mathematical
models. Mathematical Modeling & Computational Experiment (Engl.
Transl.) 1(4): 407414.
Sokol Z. 2006. Nowcasting of 1-h precipitation using radar and NWP
data. Journal of Hydrology 328: 200211.
Sorooshian S, Hsu KL, Gao X, Gupta HV, Imam B, Braithwaite D.
2000. Evaluation of PERSIANN system satellite-based estimates of
tropical rainfall. Bulletin of American Meteorological Society 81:
20352046.
Steven AM. 2006. Feasibility of Snowpack Characterization Using
Remote Sensing and Advanced Data Assimilation Techniques. Technical
Completion Report 980, University of California Water Resources
Centre. http://repositories.cdlib.org/wrc/tcr/980.
Sweeney TL. 1992. Modernized Areal Flash Flood Fuidance. NOAA
Technical Report NWS HYDRO 44. Hydrologic Research Laboratory,
National Weather Service, NOAA: Silver Spring, MD; p. 21 and
appendix.
Takeuchi K, Hapuarachchi P, Zhou M, Ishidaira H, Magome J. 2008.
A BTOP model to extend TOPMODEL for distributed hydrological
simulation of large basins. Hydrological Processes 22: 32363251.
Thirumalaiah K, Deo MC. 1998. River stage forecasting using articial
neural networks. Journal of Hydrologic Engineering 3(1): 2632.
Todini E. 1995. New trends in modelling soil processes from hillslope to
GCM scales. In The Role of Water and the Hydrological Cycle in Global
Change, Oliver HR, Oliver SA (eds). Global Environmental Change,
NATO ASI Series, Series I, vol. 31. Springer-Verlag; 317347.
Turk J, Rohaly G, Hawkins J, Smith EA, Grose A, Marzano FS,
Mugnai A, Levizzani V. 2000. Analysis and assimilation of rainfall
from blended SSM/I, TRMM and geostationary satellite data.
10
th
Conf. on Satellite Meteorol. and Oceanography. American
Meteorological Society: Long Beach, California; 6669.
Vischel T, Pegram GGS, Sinclair S, Wagner W, Bartsch A. 2008.
Comparison of soil moisture elds estimated by catchment modelling
and remote sensing: a case study in South Africa. Hydrology
and Earth System Sciences 12: 751767, www.hydrol-earth-syst-
sci.net/12/751/2008/.
Weltz MA, Ritchie JC, Fox HD. 1994. Comparison of lidar and eld
measurements of vegetation heights and canopy cover. Water Resources
Research 30: 13111319.
Yatheendradas S, Wagener T, Gupta H, Unkrich C, Goodrich D, Schaffner
M, Stewart A. 2008. Understanding uncertainty in distributed ash
ood forecasting for semiarid regions. Water Resources Research 44:
W05S19. DOI:10.1029/2007WR005940.
Yuan H, Mullen SL, Gao X, Sorooshian S, Du J, Juang HMH. 2007.
Short-range probabilistic quantitative precipitation forecasts over the
southwest United States by the RSM ensemble system. Monthly
Weather Review 135: 16851698.
Zadeh LA. 1965. Fuzzy sets. Information and Control 8: 338353.
Zadeh LA. 1978. Fuzzy Sets as a Basis for Theory of Possibility. Fuzzy
Sets and Systems 1: 328.
Copyright 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 27712784 (2011)

You might also like