You are on page 1of 10

Advances in Water Resources 27 (2004) 889–898

www.elsevier.com/locate/advwatres

Treatment of precipitation uncertainty in rainfall-runoff


modelling: a fuzzy set approach
a,*
Shreedhar Maskey , Vincent Guinot b, Roland K. Price c

a
UNESCO-IHE Institute for Water Education, P.O. Box 3015, 2601 DA Delft, The Netherlands
b
Université Montpellier 2, Maison des Sciences de l’Eau, 34095 Montpellier Cedex 5, France
c
UNESCO-IHE Institute for Water Education, P.O. Box 3015, 2601 DA Delft, The Netherlands

Received 22 December 2003; received in revised form 9 June 2004; accepted 6 July 2004
Available online 12 September 2004

Abstract

The uncertainty in forecasted precipitation remains a major source of uncertainty in real time flood forecasting. Precipitation
uncertainty consists of uncertainty in (i) the magnitude, (ii) temporal distribution, and (iii) spatial distribution of the precipitation.
This paper presents a methodology for propagating the precipitation uncertainty through a deterministic rainfall-runoff-routing
model for flood forecasting. It uses fuzzy set theory combined with genetic algorithms. The uncertainty due to the unknown tem-
poral distribution of the precipitation is achieved by disaggregation of the precipitation into subperiods. The methodology based on
fuzzy set theory is particularly useful where a probabilistic forecast of precipitation is not available. A catchment model of the
Klodzko valley (Poland) built with HEC-1 and HEC-HMS was used for the application. The results showed that the output uncer-
tainty due to the uncertain temporal distribution of precipitation can be significantly dominant over the uncertainty due to the
uncertain quantity of precipitation.
 2004 Elsevier Ltd. All rights reserved.

Keywords: Disaggregation; Flood forecasting; Fuzzy sets; Genetic algorithm; Precipitation; Uncertainty

1. Introduction uncertainty in a deterministic rainfall-runoff-routing


model. The various sources of uncertainty associated
Since flooding is an inherently uncertain natural with the model outputs can be classified as [18]:
process, the reliability and credibility of flood forecast-
ing and warning systems cannot be guaranteed without (1) Model uncertainty (due to assumptions in the model
properly incorporating the sources of uncertainty into equations, model building, and other forms of
the forecasting and warning systems. Estimating uncer- imprecision in conceptualising the real system)
tainty in forecasts enables an authority to set risk-based (2) Input uncertainty (due to imperfect forecasts of
criteria for flood warning and offers potential for addi- future precipitation, evaporation, etc.)
tional economic benefits of forecasts to every rational (3) Parameter uncertainty (due to imperfect assess-
decision maker [15]. The present discussion deals with ments of model parameters)
(4) Natural and operational uncertainty (due to unfore-
seen causes, e.g. glacier lake overflow, landslides
*
and debris flows, etc., malfunctioning of system
Corresponding author. Tel.: +31 15 2151755; fax: +31 15 2122921.
E-mail addresses: s.maskey@unesco-ihe.org (S. Maskey), guinot@
components (hardware and software), erroneous
msem.univ-montp2.fr (V. Guinot), r.price@unesco-ihe.org (R.K. and missing data, human errors and mistakes,
Price). etc.).

0309-1708/$ - see front matter  2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.advwatres.2004.07.001
890 S. Maskey et al. / Advances in Water Resources 27 (2004) 889–898

The latter category is broad and is usually not ad- 2.1. Temporal disaggregation of precipitation for uncer-
dressed by uncertainty assessment in operational flood tainty assessment
forecasting systems. Model uncertainty may be assessed
by comparing results from different models [22]. The In catchment modelling, some of the forcings may be
parameter uncertainty can reflect part of the uncertainty observed with a period larger than the typical reaction
associated with the model structure [7]. Such assess- time of the catchment. A typical example is a rainfall
ments of model uncertainty however, remain largely measured or forecasted on an hourly basis when the
subjective. One of the important developments for the catchment response time is half an hour. In this case,
treatment of uncertainty in water systems modelling is using the (average) measured forcing directly as an input
the generalised likelihood uncertainty estimation to the model may lead to underestimation of the ampli-
(GLUE) by Beven and Binley [2]. Some examples of tude of the model response (because the variations in the
the application of the GLUE methodology in rainfall- signal are smoothed out into the averaged measure-
runoff modelling are [3] and [5]. ment). This is particularly true when peak values in
The present discussion is limited to the treatment of the output are related to peak values in the inputs (e.g.
uncertainty in precipitation as an input uncertainty. peak rainfall resulting in peak runoff). In order to esti-
Real time flood forecasting using a rainfall-runoff-rout- mate better the peak values of the model output, the in-
ing model requires the forecasted precipitation for the puts must be reconstructed at a time scale smaller than
forecast period [4]. The precipitation uncertainty repre- the typical reaction time of the hydrological system
sents the major part of the input uncertainty [14]. The under study. Failing to generate an input at smaller time
uncertainty in the forecasted precipitation results from scale than the catchment response time may introduce
the uncertainty in (i) the magnitude, (ii) the temporal error/uncertainty in the model outputs. If a model with
distribution over the forecast period, and (iii) the spatial multiple inputs is to be used (e.g. a catchment model
distribution over the catchment. using records from several rainfall gauges) the spatial
The present paper proposes a methodology for the distribution of the inputs is also a source of uncertainty.
propagation of the precipitation uncertainty through a In addition, the average rainfall rate over a given sub-
deterministic rainfall-runoff-routing model. The meth- basin is not necessarily equal to the point value meas-
odology is independent of the structure of the model ured at the corresponding rainfall gauge. Besides the
and can therefore be used with any deterministic, rain- imprecision attached to the measuring device, the point
fall-runoff-routing model. It uses the temporal disaggre- measurement introduces uncertainty in the average in-
gation of the precipitation (over a period) into put over the subbasin to which the measuring device is
subperiods and allows the uncertainty in both the tem- attached.
poral and spatial distributions of the rainfall to be ac- The idea presented here consists in dividing the tem-
counted for. The methodology is detailed in the poral period over which the accumulated rainfall is
framework of fuzzy set theory for uncertainty propaga- known into a fixed number of subperiods and to ran-
tion using the extension principle (see Appendix A). The domly disaggregate the accumulated sum over subperi-
methodology however can be easily adapted for imple- ods. The disaggregated rainfall should aggregate up to
mentation in the framework of the Monte Carlo tech- the (measured) accumulated sum. The disaggregated pre-
nique for a probabilistic approach. The method is cipitation signal is then used as an input to the rainfall-
applied to a model of the Klodzko (Poland) watershed, runoff-routing model. Let the accumulated sum of the
based on the HEC-1 [28] and HEC-HMS [29,30] model- precipitation for the subbasin i (i = 1, . . ., m) during a
ling software. given time period of T be Pi. Dividing the period T into
n subperiods, Pi is disaggregated over the subperiods as
Xn
2. Methodology Pi ¼ pi;j i ¼ 1; . . . ; m ð1Þ
j¼1
The methodology of uncertainty assessment in flood
where pi,j is the disaggregated quantity for the jth subpe-
forecasting due to precipitation is presented in this sec-
riod (Fig. 1).
tion. The principle of disaggregation is presented in Sec-
For the sake of simplicity coefficients bi,j are intro-
tion 2.1 and the representation of the uncertainty in the
duced, such that
magnitude of the precipitation by a fuzzy membership
function is presented in Section 2.2. The algorithm for pi;j ¼ P i bi;j ð2Þ
the method is presented in Section 2.3, whereas the algo-
where
rithm used to determine the minimum and the maximum
)
values of the model outputs and a simplification of the 0 6 bi;j 6 1 8ði; jÞ
methodology are described in Sections 2.4 and 2.5 Pn ð3Þ
respectively. j¼1 bi;j ¼ 1 8i
S. Maskey et al. / Advances in Water Resources 27 (2004) 889–898 891

Pi µ (Pi)

Pi /n p i,1 p i,j p i,n Pi


t Pi,min Pi,mc Pi,max
j=1 j j=n
T Fig. 2. A triangular membership function with three values: Pi,min,
Pi,mc and Pi,max.
Fig. 1. Average and disaggregated precipitations.
the 3 values Pmc, Pmin and Pmax a triangular MF (Fig. 2)
is assumed, given as
In the present approach the coefficients bi,j are generated 8
>
> 0 if P i < P i;min
randomly and independently from each other, subject to >
> P i P i;min
the constraints (3); the method is described in Section >
< P i;mcP if P i;min 6 P i 6 P i;mc
i;min
2.3. Eqs. (2) and (3) allow the quantity pi,j to take any lðP i Þ ¼ ð4Þ
>
>
P i;max P i
if P i;mc 6 P i 6 P i;max
value from 0 to Pi. Allowing the coefficients bi,j to vary >
> P i;max P
>
:
i;mc

over the subperiods accounts for different possible tem- 0 if P i > P i;max
poral distributions over the various subbasins, whereas
The qualitative meaning of the triangular membership
varying the coefficients over subbasins accounts for the
function is the following. The ‘‘true’’ value of precipita-
spatial variations of the rainfall field.
tion, Pi, is certainly included between Pi,min and Pi,max
Note that this methodology is based on the implicit
and is likely to be close to Pi,mc. The key words are ‘‘in-
assumption that the length of a subperiod is larger than
cluded’’ and ‘‘close’’. These words constitute the only
the correlation time of the precipitation signal. If it is
information that one has about the problem [23]. By
not the case, the coefficients bi,j should not be generated
doing so the uncertainty or imprecision associated with
independently from each other. The length of the subpe-
the single value of the forecast precipitation is repre-
riod, however should not be too large, in which case the
sented using a fuzzy number on the basis of the mini-
resulting signals will be close to the average signals over
mum information available about the anticipated
the period. Note that methods exist, that allow signals to
precipitation.
be generated taking into account the signal correlation
[19,24]. In the absence of any information about the cor-
2.3. Algorithm for the propagation of uncertainty
relogram of the rainfall signal, such methods cannot be
used and the coefficients bi,j must be generated independ-
The precipitation represented by a MF and disaggre-
ently from each other. Therefore, for the present imple-
gated into subperiods is propagated using the extension
mentation of the methodology the disaggregated signals
principle of fuzzy set theory. Some definitions on fuzzy
are generated without considering their correlations.
sets are presented in Appendix A. The extension princi-
However, it is important to note that this methodology
ple [31,32] is performesd by the a-cut method [26]. An
for propagating the precipitation uncertainty can still be
example of an a-cut for a MF and corresponding lower
used with correlated signals if available. Also note that
bound (Pi,LB) and upper bound (Pi,UB) are shown in Fig.
the number of subperiods has a direct influence on the
3. The Pi,LB and Pi,UB are the minimum and the maxi-
maximum possible rainfall intensity. This is because
mum values of the precipitation, respectively for the
the maximum possible intensity is obtained when the
given a-cut.
rainfall is concentrated over one subperiod, with a rain-
Let a function f represents the rainfall-runoff-routing
fall intensity nPi/T.
model with the precipitation as input and the runoff, Q,
as output:
2.2. Precipitation uncertainty represented by a member-
ship function
µ (Pi )
The representation of uncertainty using fuzzy set the-
ory on the accumulated precipitation, Pi (i = 1, . . ., m) re- 1 (α ) (α )
quires the membership function (MF) of the Pi to be Pi , LB Pi , UB
α -cut level
known. In the absence of sufficient information about α
the MFs, it is assumed that only the range of uncertainty Pi
Pi,min Pi,mc Pi,max
given by the minimum (Pmin) and maximum (Pmax) val-
ues are available. It is also assumed that the given value Fig. 3. An a-cut level and corresponding upper and lower bounds. The
of Pi is its most credible value represented by Pi,mc. With definitions of Pi,min, Pi,mc and Pi,max are identical to those in Fig. 2.
892 S. Maskey et al. / Advances in Water Resources 27 (2004) 889–898

Q ¼ f ðpi;j ; i ¼ 1; . . . ; m; j ¼ 1; . . . ; nÞ of input sets. Different types of algorithms for the


determination of the minimum and maximum val-
¼ f ½ðp1;1 ; . . . ; p1;n Þ; . . . ; ðpm;1 ; . . . ; pm;n Þ ð5Þ
ues may deal differently for the values of s and r,
A general procedure for the propagation of the three such as using stopping criteria. The algorithm used
forms of precipitation uncertainty through the model in the present study for the determination of maxi-
(Eq. (5)) can be decomposed into the following steps. mum and minimum values is presented in Section
2.4.
(1) Select a value of a 2 [0, 1] (called an a-cut level) for (9) Repeat steps (2)–(8) for as many a-cuts as needed to
the MF of the input. Use the same value of a for the produce a complete MF for the output.
MFs of all inputs.
(2) For the given a-cut, determine for all i = 1,. . .,m the 2.4. Algorithm used for the determination of minimum and
ðaÞ ðaÞ
lower bound, P i;LB , and the upper bound, P i;UB (see maximum
Fig. 3).
ðaÞ ðaÞ
(3) Generate randomly a value P i;ðrÞ between P i;LB and An algorithm is needed for the determination of the
ðaÞ
P i;UB for all i = 1, . . ., m. This value corresponds to maximum and minimum values of the model outputs
the accumulated sum over the period T, i.e. Pi in (Eq. (8)). In the present study a genetic algorithm
Eq. (1). (GA) is used. GAs are search algorithms based on the
(4) Generate the coefficients bi,j,(s) randomly between 0 mechanics of natural selection and natural genetics
and 1. For each i, only n  1 coefficients are gener- [10]. Various applications of GAs are reported elsewhere
ated. If the sum of the first n1 coefficients exceeds [12,17,20]. A simple GA consists of (i) a random selec-
1, reject all the coefficients and generate a new set of tion of the initial population, (ii) a selection for mating,
n  1 coefficients. Repeat this process until the sum (iii) crossover, and (iv) mutation. The version of the GA
is smaller than or equal to 1. The nth coefficient is used in the present study consists of tournament selec-
then determined as follows: tion [9], uniform crossover [21], two children per pair
X
n1 of parents, jump and creep mutations [6], elitism and
bi;n;ðsÞ ¼ 1  bi;k;ðsÞ ð6Þ niching [11].
k¼1 In the present methodology (Section 2.3) the tasks for
ðaÞ steps (3), (4), (7) and (8) are replaced by the GA opera-
(5) Use the value P i;ðrÞ generated in step (3) and the
coefficients bi,j,(s) generated in step (4) in Eq. (2) to tions. The number of parameters (N) for the GA is given
disaggregate the rainfall over the subperiods; i.e. by
ðaÞ ðaÞ N ¼ m þ ðn  1Þm ¼ nm ð9Þ
pi;j;ðrÞ;ðsÞ ¼ P i;ðrÞ bi;j;ðsÞ ð7Þ
ðaÞ where m and n are the number of subbasins and subpe-
(6) Use the pi;j;ðrÞ;ðsÞ ði ¼ 1; . . . ; m; j ¼ 1; . . . ; nÞ as riods respectively, as defined in Section 2.1. Instead of
inputs to the model (Eq. (5)). This leads to a model the parameters s and r (Section 2.3), GA uses initial pop-
ðaÞ
output QðrÞ;ðsÞ . ulation (Np) and maximum number of generation (Ng).
(7) Repeat steps (4)–(6) s times. This means that for In the present study, Np = 50 and Ng = 40 were used
every value of the accumulated precipitation for the example with three subperiods (Section 4.1)
obtained from step (3), s number of disaggregated and Np = 200 and Ng = 50 were used for the example
sets of precipitation are generated. This produces s with six subperiods (Section 4.2).
outputs.
(8) Repeat steps (3)–(7) r times. This produces r · s out- 2.5. Simplification of the methodology
puts. From the r · s outputs determine the mini-
mum and maximum of the outputs. These are An obvious drawback of the above methodology is
assumed to be equal to the lower and upper bounds that the search space becomes larger as the number of
of the output, that is subbasins and subperiods increases, which makes the
9
ðaÞ ðaÞ
QLB ¼ minðQðrÞ;ðsÞ Þ >= determination of the maximum and minimum values
r;s
ð8Þ very time consuming. In an attempt to reduce the search
QUB ¼ maxðQðrÞ;ðsÞ Þ >
ðaÞ ðaÞ
; space, several checks revealed that the minimum and
r;s
maximum values of the model outputs (Qmin and Qmax)
ðaÞ
The values of s and r are primarily governed by the actually correspond to the lower bound ðP i;LB Þ and the
ðaÞ
size of the search space for the determination of upper bound ðP i;UB Þ values (for the given a-cut) respec-
minimum and maximum values of the outputs. tively, of the precipitation determined in step (2) of the
The search space in this example is the function of algorithm. Therefore, instead of looking for all possible
ðaÞ
m and n. Obviously, more number of subbasins values of precipitations between LB and UB, P i;LB is
ðaÞ
and subperiods means more possible combinations used to evaluate Qmin and P i;UB is used to evaluate Qmax.
S. Maskey et al. / Advances in Water Resources 27 (2004) 889–898 893

This means that the model function is assumed mono- SB 1 SB 2


tonic with respect to the quantity of the accumulated
SB 4
precipitation in the forecast period. This simplification
brings the value of r in the algorithm down to 2: one JN 1 SB 3
for the upper bound and one for the lower bound. In SB 7 JN 2
RH 1
RH 3
terms of the number of parameters for the GA (Eq. RH 5 RH 4
SB 5
(9)), N reduces to (n  1)m. This is an important conclu- SB 8
RH 2 Legend
JN 3
sion as it helps to reduce the computational effort, JN 4 RH 6 Subbasin
SB 6
particularly when working on a large catchment with RH 7
SB 9 Junction
many subbasins.
JN 5 Subbasin outlet

Outlet River reach

3. Application example Fig. 4. Schematic diagram of the watershed (SB=subbasin; RH=reach;


JN=junction).
The methodology was applied to a flood forecasting
model of Klodzko valley (Poland) on the river Nysa values as 0.7 and 1.3 times the given precipitation.
Klodzka. The river serves as a highland tributary of Although the minimum and maximum bounds are taken
the upper Odra and the lag time between rainfall and arbitrarily, discussion with experts suggested that the
runoff is very short (see [16,27]). The basin under consid- assumption is not too far from reality. The starting date
eration consists of nine subbasins. The basin covers a for simulation was 4 July 1997 at 06:00 h, while the fore
total area of 1744 km2 with the subbasin areas ranging cast of precipitation started on July 6 1997 at 00:00 h
from 64 to 280 km2. The data observed during the flood (i.e. 42 h from the point of start of simulation).
of July 1997 were used. The town of Klodzko located on Since the interval of the available precipitation data is
the river Nysa Klodzka was virtually ruined by this 3 h the size of the forecast period is also taken as 3 h. The
flood with several casualties and the destruction of very short lag time between the rainfall and runoff justi-
numerous houses [16]. fies the use of the present methodology of uncertainty
The model was built with HEC-1 and HEC-HMS. In assessment with disaggregation. Three subperiods of
particular, HEC-HMS (its Calibration Module) was 1 h each and six subperiods of half an hour each are con-
used for model calibration and HEC-1 was used for sim- sidered for disaggregation. It is to be noted that for each
ulation. The options used in the present application to forecast, only the uncertainty in the forecast precipita-
represent the rainfall-runoff processes are the Soil Con- tion during the same forecast period is considered and
servation Service (SCS) Curve Number method for run- no uncertainty is assumed in the precipitation previous
off volume computation, ClarkÕs unit hydrograph to the current forecast period. For example, to forecast
method for direct runoff computation, the exponential the flow for 3 h, no uncertainty is assumed in the precip-
recession method for baseflow computation and the itation up to the time 0 h; similarly, to forecast the flow
Muskingum method for flow routing. Readers are re- for 6 h no uncertainty is assumed in the precipitation up
ferred to [29] for the details of these methods. The selec- to the time 3 h, and so on. A total of 20 forecasts
tion of the methods in the present model was governed (20 · 3 h=60 h) from 6 July at 00:00 to 8 July at 09:00
particularly by the availability of the data. The analysis were carried out.
of the influence of different methods is beyond the scope
of this study.
A schematic diagram of the model is shown in Fig. 4. 4. Results
Some of the parameters of the model are estimated by
calibration and some are estimated from available data All the results presented and discussed here are based
and engineering judgement. The calibration is per- on the forecast downstream of junction 5 (Fig. 4).
formed using the calibration module of HEC-HMS. In Firstly, the simulation is carried out to compute the dis-
particular, the SCS Curve Number, the time of concen- charge without considering any uncertainty in the pre-
tration, the storage coefficient and the Muskingum K cipitation. The simulated discharge together with the
were determined by calibration. corresponding observed discharge is presented in Fig.
Observed precipitation data (cumulative for every 3 h) 5. The figure also shows the basin average precipitation
were available. Due to the lack of forecasted precipita- on the negative y-axis. It is to be noticed that the simu-
tion data, the observed precipitation was assumed as a lation was carried out taking the accumulated precipita-
forecasted precipitation for the application of the present tion in each subbasin, and not with the basin average
methodology. The MF representing the uncertainty in precipitation.
the quantity of the forecasted precipitation for each Secondly, the methodology is applied to estimate the
subbasin is obtained by taking minimum and maximum uncertainty in the forecasted discharges due to the
894 S. Maskey et al. / Advances in Water Resources 27 (2004) 889–898

Time (hours)
0 24 48 72 96 120 144 168
0.0

Precipitation (mm)
10.0

20.0

30.0
2000
Discharge (m3/s) Simulated
1500
Observed
1000

500

0
0 24 48 72 96 120 144 168
Time (hours)

Fig. 5. Observed and simulated discharges without considering uncertainty in the forecasted precipitation for the period 4 to 11 July 1997. The hour
0 refers to 4 July 1997 at 06:00 h.

uncertainty in the forecasted precipitation. The results without disaggregation. This suggests that the uncer-
are characterised as ‘‘with disaggregation’’ and ‘‘without tainty due to the unknown temporal distribution can
disaggregation’’. Whereas ‘‘with disaggregation’’ consid- be a predominant factor over the uncertainty in the
ers uncertainty due to the unknown temporal distribu- magnitude of the precipitation.
tion of the precipitation by using disaggregated signals The results shown in Figs. 6 and 7 correspond to
as inputs, ‘‘without disaggregation’’ uses a constant sig- a = 0 and 1 only. For a complete membership function
nal, equal to the mean value, over the entire subperiods. (MF) of the output to be constructed, the computation
Since each subbasin precipitation is represented using a needs to be carried out at various a-cuts. Three more
fuzzy membership function for uncertainty, the former a-cuts at a = 0.25, 0.5, and 0.75 were selected. The
considers both temporal and spatial uncertainty and MFs obtained for the forecasts at 81, 84 and 87 h with
the latter considers only the spatial uncertainty. In and without disaggregation are shown in Fig. 8(a–c).
Fig. 5, the simulated discharge is the result of the precip- These figures further illustrate that the uncertainty due
itation for a = 1 without disaggregation. to the unknown temporal distribution is predominant
over the uncertainty due to the quantity of precipitation.
4.1. Results with three subperiods The support of disaggregated discharges is also greatly
increased.
Fig. 6 shows the uncertainty bounds in the forecasted For a = 1, there is no upper and lower bounds of
ða¼1Þ ða¼1Þ ða¼1Þ
discharges, i.e. QUBQLB, with disaggregation for a = 0 the input precipitation, i.e. P i;LB ¼ P i;UB ¼ P i . This
and 1 as a function of the forecast time. The comparison means that for a = 1 without disaggregation there is only
of the uncertainty bounds (QUBQLB) with and without one set of input precipitation resulting in only one out-
disaggregation is presented in Fig. 7. For a = 1 without put. Consequently, there are no upper and lower bounds
disaggregation, there is only one possible value Pi,mc for of the output. In fact, this particular case (i.e. a = 1,
the precipitation. Therefore, there is no uncertainty in without disaggregation) considers neither temporal nor
this case. It is clearly seen from Fig. 7 that in all fore- spatial uncertainty. Whereas, in the case with disaggre-
ða¼1Þ
casts the output uncertainty is much larger with than gation, the one set of input precipitation, i.e. P i , will

150
QUB - QLB (m 3/s)

100 α =1 α=0

50

0
42 54 66 78 90 102
Time (hours)

Fig. 6. Uncertainty range in forecasted discharges (QUBQLB) with disaggregation.


S. Maskey et al. / Advances in Water Resources 27 (2004) 889–898 895

150
Without disaggregation

QUB - Q LB (m /s)
With disaggregation
100

3
50

0
42 54 66 78 90 102
(a) Time (hours)

150
With disaggregation
QUB - Q LB (m /s)
3

100

50

0
42 54 66 78 90 102
(b) Time (hours)

Fig. 7. Uncertainty range in forecasted discharges (QUBQLB): (a) for a-cut level 0, and (b) for a-cut level 1. There are no upper and lower bounds
for a-cut level 1 in the case of without disaggregation.

1.00 of temporal uncertainty. These conclusions are clearly


Membership

seen in Fig. 8. Also of significance are the shapes of


0.50
the MFs with and without disaggregation. The shapes
of the MFs without disaggregation are fairly triangular
showing that if the temporal uncertainty is neglected the
0.00
600 700 800 triangular MFs of the inputs (precipitation) result in a
(a) Discharge (m 3/s)
triangular MF of the output (discharge). Alternatively,
with disaggregation the shapes are rather uniform sug-
1.00 gesting that the additional uncertainty due to the un-
Membership

certainty in the temporal distributions is almost


0.50 independent from the a-cut level.

4.2. Results with three and six subperiods


0.00
800 900 1000
(b) Discharge (m 3/s) A key issue in the application of this methodology is
the selection of an appropriate number of subperiods.
1.00 Increasing the number of subperiods leads to increase
Membership

the number of parameters for the search space and con-


sequently the computational requirement. On the other
0.50
hand, intuitively, increasing the number of subperiods
may widen the uncertainty bounds in the output. To ver-
0.00
1150 1250 1350 ify this, the results with three subperiods are compared
(c) Discharge (m3/s)
with the results with six subperiods. With six subperiods
and nine subbasins, the number of parameters increases
Fig. 8. Membership functions of forecasted discharges without disag- from 18 to 45. In order to reduce the number of param-
gregation (grey line) and with disaggregation (dark line): (a) forecast eters, the temporal pattern is assumed to be identical for
at 81 h, (b) forecast at 84 h, and (c) forecast at 87 h.
all the subbasins. This assumption significantly reduces
the number of parameters from 45 to 5 in the case of
have a number of temporally disaggregated precipita- six subperiods and 18 to 2 in the case of three subperi-
tion sets resulting in a number of output discharges ods. It is however important to check the effect of this
which means that there exist upper and lower bounds assumption. Therefore, the three subperiods case is also
for the outputs even for a = 1 due to the consideration repeated applying this assumption. The comparison of
896 S. Maskey et al. / Advances in Water Resources 27 (2004) 889–898

150
Same temporal pattern for all
subbasins

QUB - Q LB (m /s)
100

3
Varying temporal pattern over
subbasins

50

0
42 54 66 78 90 102

(a) Time (hours)

150
Same temporal pattern for all
subbasins
QUB - Q LB (m /s)
3

100 Varying temporal pattern over


subbasins

50

0
42 54 66 78 90 102
(b) Time (hours)

Fig. 9. Uncertainty range in forecasted discharges (QUBQLB) with disaggregation into three subperiods for: (a) a-cut level 0, and (b) a-cut level 1.

200
3 Subperiods
QUB - Q LB (m /s)

150 6 Subperiods
3

100

50

0
42 54 66 78 90 102

(a) Time (hours)

200
3 Subperiods
Q UB - Q LB (m /s)

150 6 Subperiods
3

100

50

0
42 54 66 78 90 102

(b) Time (hours)

Fig. 10. Uncertainty range in forecasted discharges (QUBQLB) with disaggregation for: (a) a-cut level 0, and (b) a-cut level 1.

this result with the previous result with varying temporal 5. Conclusion
patterns over subbasins is presented in Fig. 9. The differ-
ences are minor except for forecasts at 84 and 87 h. In This paper has presented a methodology for the treat-
this particular example, assuming identical temporal ment of precipitation uncertainty in rainfall-runoff-rout-
patterns over all the subbasins resulted in larger uncer- ing models in the framework of fuzzy set theory assisted
tainty bounds than with the varying patterns. by genetic algorithms. The methodology uses the ran-
Fig. 10 shows the comparison of uncertainty bounds dom disaggregation of precipitation into subperiods to
of cases with three and six subperiods for a = 0 and 1. It take into account uncertainty due to the unknown tem-
can be observed that the increase in the uncertainty poral distribution of the precipitation. The principle of
bound is not very significant compared to the increase disaggregation presented in this paper for uncertainty
in the number of subperiods. This means that the num- assessment can also be used in the framework of the
ber of subperiods can be taken reasonably small without Monte Carlo technique if the forecast of precipitation
underestimating the uncertainty in the output is probabilistic. This methodology is independent of
unacceptably. the structure of the forecasting model. In other words,
S. Maskey et al. / Advances in Water Resources 27 (2004) 889–898 897

it can be used with any rainfall-runoff-routing type of A.1. Fuzzy set and fuzzy number
deterministic model. The results show the great potential
of the fuzzy extension principle combined with a genetic Let X be a universe set of x values (elements). Then A
algorithm for the propagation of uncertainty. The re- is called a fuzzy (sub)set of X, if A is a set of ordered
sults also show that the output uncertainty due to the pairs:
uncertainty in the temporal and spatial distributions A ¼ fðx; lA ðxÞÞ; x 2 X ; lA ðxÞ 2 ½0; 1g ðA:1Þ
can be significantly dominant over the uncertainty due
to the uncertain magnitude of precipitation. This sug- where lA(x) is the grade of membership (or degree of be-
gests that using space- and time-averaged precipitation lief) of x in A. The function lA(x) is called the member-
over the catchment may lead to erroneous forecasts. ship function of A. A fuzzy set is called a fuzzy number if
The estimated uncertainty in the output may seem small it is normal (i.e. the maximum of the memberships is 1)
compared to the magnitude of the flood. This is due to as well as convex.
relatively short forecast period (3 h). Obviously increas-
ing the forecast period significantly increases the uncer- A.2. Fuzzy extension principle
tainty in the forecasted precipitation and thereby
increases the output uncertainty. Moreover, it is to be Consider a function of several uncertain variables:
noted that the estimated uncertainty is only due to the y ¼ f ðx1 ; . . . ; xn Þ ðA:2Þ
uncertainty in the precipitation. It does not include the
parameter and model uncertainty. Let fuzzy sets A1, . . ., An be defined on the universes
This paper has also attempted to answer the question X1, . . ., Xn such that x1 2 X1, . . ., xn 2 Xn. The mapping
concerning an appropriate number of subperiods. While of these input sets can be defined as a fuzzy set B:
the estimated uncertainty without disaggregation and B ¼ f ðA1 ; . . . ; An Þ ðA:3Þ
disaggregated with three subperiods differed signifi-
where the membership function of the image B is given
cantly (Fig. 9a), the results with three and six subperiods
by
showed little difference (Fig. 10). Therefore, in this par- 8
ticular example three subperiods seem enough. In gen- > maxfmin½lA1 ðx1 Þ; . . . ; lAn ðxn Þ;
>
>
eral, the determination of the number of subperiods < y ¼ f ðx ; . . . ; x Þg
1 n
should be governed from the considerations that: (i) lB ðyÞ ¼ ðA:4Þ
>
> 0 if there is no x1 2 X 1 ; . . . ; xn 2 X n
the uncertainty should not be underestimated beyond >
:
a reasonable limit, (ii) the computational requirements such that f ðx1 ; . . . ; xn Þ ¼ y
should not be too large, (iii) the subperiods should be Eq. (A.4) is the mathematical expression for the exten-
large enough for the disaggregated precipitations to re- sion principle of fuzzy sets. The above equation is de-
main realistic, and (iv) the length of the subperiod fined for a discrete-valued function f. If the function f
should be of the same order of magnitude as the corre- is continuous-valued then the max operator is replaced
lation time of the precipitation signal. by the sup (supremum) operator (the supremum is the
least upper bound).

Acknowledgments A.3. Alpha-cut of a fuzzy set

This research was initiated by the EC project Opera- An a-cut (alpha-cut) of a fuzzy set A, denoted as Aa is
tional Solutions for the Management of Inundation the set of elements x of a universe of discourse X for
Risks in the Information Society (OSIRIS), contract which the membership function of A is greater than or
IST-1999-11598. The authors highly appreciate and ex- equal to a. That is,
tend their gratitude to the IMWM, Poland for providing
Aa ¼ fx 2 X ; lA ðxÞ P a; a 2 ½0; 1g ðA:5Þ
the data and the model. The authors are also grateful to
Prof. Mike J. Hall, UNESCO-IHE Institute for Water The a-cut provides a convenient way of performing
Education, The Netherlands for providing valuable arithmetic operations on fuzzy sets and fuzzy numbers
remarks on the manuscript.
µ X (x)

Appendix A. Definitions on fuzzy sets and operations 1


a b α-cut
-cut level
α
The mathematical definition of the extension princi-
ple and some other relevant definitions on fuzzy sets X
x1 x2
and operations are briefly presented here. More detailed
coverage of the topic can be found in [1,8,13,25,33]. Fig. 11. Illustration of an a-cut of a fuzzy set.
898 S. Maskey et al. / Advances in Water Resources 27 (2004) 889–898

including in applying extension principle. Let us con- [14] Krzysztofowicz R. Bayesian theory of probabilistic forecasting via
sider a fuzzy number (a membership function) as shown deterministic hydrologic model. Water Resour Res 1999;35(9):
2739–50.
in Fig. 11. And let a a-cut level intersects at two points a [15] Krzysztofowicz R. The case for probabilistic forecasting in
and b on the membership function. The values of the hydrology. J Hydrol 2001;249:2–9.
variable x corresponding to points a and b are x1 and [16] Kundzewicz ZW, Szamalek K, Kowalczak P. The great flood of
x2 (x1, x2 2 X), respectively. Then the set Aa contains 1997 in Poland. Hydrol Sci J 1999;44(6):855–70.
all possible values of the variable X including and be- [17] Maskey S, Jonoski A, Solomatine DP. Groundwater remediation
strategy using global optimization algorithms. J Water Resour
tween x1 and x2. The x1 and x2 are commonly referred Planning Manage 2002;128(6):431–40.
as lower and upper bounds of the a-cut. [18] Maskey S, Price RK. Uncertainty issues in flood forecasting. In:
Proc Workshop: Flood Events—Are We Prepared? Berlin, 2003.
p. 123–36.
[19] Mantoglou A, Wilson JL. The turning bands method for
References simulation of random fields using line generation by a spectral
method. Water Resour Res 1982;18(5):1379–94.
[1] Bardossy A, Duckstein L. Fuzzy rule-based modelling with [20] Nixon JB, Dandy GC, Simpson AR. A genetic algorithm for
applications to geophysical, biological and engineering sys- optimising off-farm irrigation scheduling. J Hydroinformatics
tems. Boca Raton: CRC Press; 1995. 2001;3:11–22.
[2] Beven K, Binley A. The future of distributed models: model [21] Pawlowsky MA. Crossover operations. In: Chambers L, editor.
calibration and uncertainty prediction. Hydrol Processes Practical handbook of genetic algorithms, applications,
1992;6:279–98. vol. 1. Boca Raton: CRC Press; 1995. p. 101–14.
[3] Blazkova S, Beven K. Flood frequency estimation by continuous [22] Radwan M, Willems P, Berlamont J. Sensitivity and uncertainty
simulation for a catchment treated as ungauged (with uncer- analysis for river water quality modelling. In: Hydroinformatics
tainty). Water Resour Res 2002;38(8):1139. 2002: Proc fifth international conference on hydroinformat-
[4] Borga M. Accuracy of radar rainfall estimates for streamflow ics. London: IWA Publishing; 2002. p. 482–7.
simulation. J Hydrol 2002;267:26–39. [23] Revelli R, Ridolfi L. Fuzzy approach for analysis of pipe network.
[5] Cameron DS, Beven KJ, Tawn J, Blazkova S, Naden P. Flood J Hydraulic Eng 2002;128(1):93–101.
frequency estimation by continuous simulation for a gauged [24] Rodriguez-Iturbe I, Eagleson PS. Mathematical models of rain-
upland catchment (with uncertainty). J Hydrol 1999;219:169–87. storm events in space and time. Water Resour Res 1987;23(1):
[6] Carroll DL. Chemical laser modelling with genetic algorithms. 181–90.
Amer Inst Aeronaut Astronaut (AIAA) J 1996;34(2):338–46. [25] Ross TJ. Fuzzy logic with engineering applications. New
[7] Cullen AC, Frey HC. Probabilistic techniques in exposure York: McGraw-Hill; 1995.
assessment: a handbook for dealing with variability and uncer- [26] Schulz K, Huwe B. Water flow modelling in the unsaturated zone
tainty in models and inputs. New York: Plenum Press; 1999. with imprecise parameters using a fuzzy approach. J Hydrol
[8] Dubois D, Prade H. Possibility theory—an approach to compu- 1997;201:211–29.
terized processing of uncertainty. New York: Plenum Press; [27] Szamalek K. The great flood of 1997 in Poland: the truth and
1988. myth. In: Marsalek J, Watt WE, Zeman E, Sieker F, editors.
[9] Gen M, Cheng R. Genetic algorithms and engineering optimiza- Flood issues in contemporary water management, 2 Environmen-
tion. New York: John Wiley & Sons; 2000. tal security. NATO science series. Dordrecht: Kluwer Academic
[10] Goldberg DE. Genetic algorithms in search, optimization, and Publishers; 2000. p. 67–83.
machine learning. Reading: Addison-Wesley Publishing Com- [28] USACE. HEC-1 flood hydrograph package userÕs manual. Davis,
pany; 1989. CA: Hydrologic Engineering Centre; 1998.
[11] Goldberg DE, Richardson J. Genetic algorithms with sharing for [29] USACE. Hydrologic modelling system HEC-HMS technical
multimodal function optimization. In: Genetic algorithms and reference manual. Davis, CA: Hydrologic Engineering Centre;
their applications: Proc of the second international conference 2000.
on genetic algorithms. 1987. p. 41–9. [30] USACE. Hydrologic modelling system HEC-HMS userÕs man-
[12] Harrouni KEl, Ouazar D, Walters GA, Cheng AH-D. Ground- ual. Davis, CA: Hydrologic Engineering Centre; 2001.
water optimisation and parameter estimation by genetic algorithm [31] Yager RR. A characterization of the extension principle. Fuzzy
and dual reciprocity boundary element method. Eng Anal Sets and Systems 1986;18:205–17.
Boundary Elements 1996;18:287–96. [32] Zadeh LA. The concept of a linguistic variable and its application
[13] Kaufmann A, Gupta MM. Introduction to fuzzy arithmetic, to approximate reasoning, Part I. Inform Sci 1975;8:199–249.
theory and applications. New York: Van Nostrand Reinhold; [33] Zimmermann H-J. Fuzzy set theory and its applications. Dor-
1991. drecht: Kluwer Academic Publishers; 1991.

You might also like