Top Banner

of 14

Developement of Fuzzy Rainfall Runoff...

Feb 20, 2018

Download

Documents

Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    1/14

    Hydrological Sciences-Journal~des Sciences Hydrologiques, 46(3) June 2001

    363

    Development of a fuzzy logic-based rainfall-runoff

    model

    YESHEWATESFA HUNDECHA, ANDRAS BARDOSSY &

    HANS-WERNER THEISEN

    Institut fiir Wasserba u, UniversitdtStuttgart D-70550Stuttgart Germany

    e-mail:[email protected] ;[email protected]

    Abstract Rainfall-runoff models are used to describe the hydrological behaviour of a

    r iver catchment. Many different models exist to simulate the physical processes of the

    relationship between precipitat ion andrunoff. Some of them are based on simple and

    easy-to-handle concepts, others on highly sophisticated physical and mathematical

    approaches that require extreme effort in data input and handling. Recently,

    mathematical methods using l inguist ic variables, rather than conventional numerical

    variables applied extensively in other disciplines, are encroaching in hydrological

    studies. Among these is the application of a fuzzy rule-based modell ing. In this paper

    an attempt was made to develop fuzzy rule-based routines to simulate the different

    processes involved in the generation of runoff from precipitat ion. These routines were

    implemented within a conceptual , modular , and semi-distr ibuted modelthe HBV

    model. The investigation involved determining which modules of this model could be

    replaced by the new approach and the necessary input data were identif ied. A fuzzy

    rule-based routine was then developed for each of the modules selected, and

    application and validation of the model was don e on a rainfall-runoff analysis of the

    Neckar River catchment, in southwest Germany.

    Key words

    rainfall-runoff mod elling; HBV mo del; fuzzy logic; River Neckar, Germany

    Dveloppement d'un modle pluie-dbit base de logique floue

    R sum Les modles pluie-dbit sont uti l iss pour dcrire le comportement

    hydrologique d'un bassin versant. De nombreux modles existent pour simuler les

    processus physiques dterminant la transformation de la pluie en dbit . Certains

    d'entre eux sont bass sur des concept simples et aisment transposables, tandis que

    d'autres s 'appuient sur des approches physiques et mathmatiques trs sophist iques

    qui ncessitent beaucoup d'efforts au niveau de la pr ise en compte et du trai tement des

    donnes . Depuis que lque temps, des mthodes mathmat iques manipulant auss i bien

    des variables alphanumriques que les habituelles variables numriques, ont t

    dveloppes en hydrologie. Parmi celles-ci se trouve la modlisation base de logique

    floue. Dans cet ar t icle nous prsentons une tentative de dveloppement de routines

    base de logique f loue pour simuler les diffrents p roce ssus mis en jeu dans la

    production d 'coulem ent part ir des prcipitat ion s. Ces routines ont t implmen tes

    au sein du modle HBV, conceptuel , modulaire et semi-distr ibu. L 'tude a ncessit

    de dterminer quels modules du modle pouvaient tre remplacs par la nouvelle

    approche d'une part et d' identif ier les donnes ncessaires d'autre part . Une routine

    base de logique f loue a alors t dveloppe pour chacun des modules ainsi identif is,

    et la mise en uvre et la validation du modle global ont t ralises avec les

    donnes de pluie et de dbit du bassin versant de la r ivire Neckar , au Sud-Ouest de

    l 'Al lemagne .

    Motsclefs modles pluie-d bit; modle HBV ; logique floue; Rivire Neckar, Allemagne

    INTRODUCTION

    It is customary to establish a rainfall-runoff relationship in hydrological studies of a

    river basin. Traditionally, this task has been accomplished using methods ranging from

    Open for discussion until I Decem ber 2001

    mailto:[email protected]:[email protected]:[email protected]:[email protected]:[email protected]
  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    2/14

    364 Yeshewatesfa H undecha et al.

    those that give explicit emphasis to the underlying physical processes involved in the

    generation of runoff from rainfall to simple conceptual approaches that treat the

    catchment system in a simple idealized way.

    In using models that are based on descriptions of the physical processes, rigorous

    mathematical equations are often needed to solve the problem at hand. Such models are

    highly demanding in terms of their data requirement and it is often necessary to estimate

    input parameters specific to the catchment being modelled. In many cases, a large number

    of these parameters are involved and there is no way of estimating them uniquely. Instead,

    they are determined subjectively based on the modeller's judgement and the effect is

    normally manifested in the output of the model (Prakash, 1986). Hence, models which are

    easy to handle and have a minimum data requirement are often sought to solve problems

    where data availability is limited and the system is too complicated to be handled by

    physical m odels.

    Recently, mathematical methods using linguistic variables rather than conventional

    numerical variables, which have been applied in other disciplines, are encroaching into

    hydrological studies as well. Among these is the application of a fuzzy rule-based

    approach in modelling processes involved in the hydrologie cycle. A fuzzy rule-based

    modelling is a qualitative modelling scheme by which one describes system behaviour

    using a natural language (Sugeno, 1993). In using a fuzzy logic-based approach in

    modelling cause and effect, relationships are described verbally rather than using known

    governing physical relationships. Some of the causes that are taken into account in the

    physically-based models may be omitted. On the other hand, some of the causes that are

    not considered in idealized types of models, because of the nature of generalization or

    unavailability of known relationships, can be included in a fuzzy logic-based approach.

    Establishing these relationships depends on observing trends between the cause and effect

    and a detailed knowledge of the underlying processes is, therefore, not required. A few

    attempts have been made so far to implement such an approach, in modelling processes

    taking place in a river catchment, and promising results have been obtained. Brdossy &

    Disse (1993) have already demonstrated the applicability of such an approach in modelling

    infiltration. See & Openshaw (1999, 2000) have also worked on the application of entirely

    Artificial Intelligent approaches and a hybrid model constructed by integrating conven

    tional and Artificial Intelligent approaches in river level and flood forecasting. In the area

    of meteorological data management, Abebe et al. (2000) have shown the applicability of

    fuzzy rule-based models for reconstruction of missing precipitation events.

    The purpose of this paper is to demonstrate the applicability of a fuzzy logic-based

    approach to rainfall-runoff mo delling. An attempt was m ade to mod el the individual

    processes involved in a watershed system, and their applicability was investigated by

    incorporating them in a mod ular conceptual model already in use.

    A

    F U Z Z Y L O G I O B A S E D M O D E L L I N G A P P R O A C H

    Quantitative rules pertaining to physical science are normally described by

    mathematical functions which, for every element in the domain, assign a unique output

    value. There are also certain classes of rules applied to linguistic variables, which do

    not have unique numerical values. A very simple example that can demonstrate such

    classes of rules could be drafted as:

    If the air is warm, one has to consume m uch water.

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    3/14

    Developmentof a fuzzyrule-basedrainfall-runoff model 365

    Here , the term wa rm is not a quantity that can be clearly defined. It can have

    values w ithin an arbitrarily chosen range. But all temp erature values w ithin the defined

    range may not be considered equally warm. Similarly, the consequence, i.e.

    consum ption of m uch water is not a quantity that can be assigned a unique value. A

    fuzzy logic-based modelling approach enables one to establish a one-to-one relation

    ship between air temperature and water consumption in a way that is quite different

    from a conve ntional functional form (Zadeh, 1973).

    Fuzzy logic modelling is based on the theory of fuzzy sets (Zadeh, 1965). Unlike

    in an ordinary binary set, in a fuzzy set the boundary is not clearly defined and partial

    membership of elements is possible. Each element of the set is assigned a membership

    value w hich can be betwe en 0 and 1 inclusively. The function that assigns this value is

    referred to as the membership function associated with the fuzzy set. Fuzzy numbers

    are special types of fuzzy sets defined on the set of real numbers. Fuzzy numbers are

    usually defined by using membership functions that have triangular shapes and are

    expressed as

    (a.\,

    02, a

    3

    )

    T

    such thata\ < a

    2

    aj a

    2

    a

    3

    04 x

    Fig. 2 Membership function of a trapezoidal fuzzy number.

    \

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    4/14

    366

    Yeshewatesfa Hundecha et al.

    given set of model variable values depends on the degree to which they fulfil the condi

    tion of the rule. The truth value corresponding to the fulfilment of the conditions of a

    rule for a given set of values of the arguments is referred to as the degree of fulfilment

    (DOF) of the rule and has values in the interval

    [0,1].

    This value is determined based on

    the membership value of each of the arguments and the logical connectors used (Brdossy

    &Duckstein, 1995).

    Normally, several rules are partially satisfied for a given set of model variables

    and hence there are several associated fuzzy consequences, which are then combined

    into an overall fuzzy consequence using different techniques of rule combination. The

    combination method used in this paper is the maximum combination technique in

    which the membership function of the overall consequence is determined as the maxi

    mum of the product of the degree of fulfilment of each of the rules and the mem ber

    ship function of their corresponding rule consequence. The combined rule consequence

    1

    lnput-1

    Fuzzification

    Xfj

    Input-N

    Fuzzification

    Defuzzification

    Crisp output

    (y)

    Fig. 3 Schematic representation of a fuzzy logic-based modelling procedure.

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    5/14

    Developmentof a fuzzyrule-basedrainfall-runoff model 367

    is then conve rted into a crisp real num ber using defuzzification techniqu es. T he

    defuzzification technique com mo nly used is the mean defuzzification in which the

    centroid of the overall fuzzy consequence is taken as the crisp output of the fuzzy rule

    system (Brdossy & Duckstein, 1995). Figure 3 shows the schematic representation of

    the modelling procedure using a fuzzy rule-based approach.

    DESCR I PTI ON OF M O DEL APPLI CATION AR EA

    Application and validation of the model was carried out on the catchment of the River

    Neckar in southwest Germany, most of which is in the state of Baden-Wurttemberg, It

    has a total dr aina ge area of 13 957 km

    2

    and its complex topography ranges from

    moderate hills to plains with erratic geological formations. The elevation of the land

    surface varies from 100 to 1000 m a.s.l. A variety of land-use patterns ranging from

    residential areas to forests characterize the basin. The outlet of the basin is at the point

    of confluence of the River N eckar with the River R hine.

    The entire basin was subdivided into 41 sub-basins so that each sub-basin could be

    modelled separately. Each sub-basin was further subdivided into up to ten different

    elevation classes. This was performed based on a 100 m elevation difference.

    Daily time series of precipitation from 1980 to 1995 and daily time series of mean

    temperature data from 1960 to 1998, observed at all meteorological stations within the

    basin, were used to run the model. For each elevation zone in each sub-basin, the

    corresponding daily time series of temperature and precipitation were computed using

    geostatistical methods from the station data. Daily discharge values from 1980 to 1996

    were obtained from different gauging stations on the River Neckar and many of its

    tributaries. Fro m th ese, data from 15 stations were chosen to calibrate the m odel.

    F O R M U L A T I O N O F M O D E L C O M P O N E N T S

    Four different processes taking place in a watershed system were identified and a fuzzy

    logic-based routine was formulated for each of these processes. The modules identified

    are: snowmelt, vapotranspiration, runoff, and basin response. A fuzzy rule-based

    routine was formulated for each of the modules independently of the others.

    Snowmelt

    The dynamics of snowmelt depends on the energy balance of the accumulated snow,

    which in turn depends on temperature, net short-wave radiation, net long-wave

    radiation, and additional heat energy input due to incoming rainfall. In utilizing a fuzzy

    rule-based routine, air temperature, accumulated snow depth and magnitude of daily

    precipitation were considered as factors that influence the amount of snowmelt. Since

    the short-wave and long-wave radiations largely depend on the air temperature, they

    were omitted. Although the effect of temperature on the amount of snowmelt is

    apparent, the effect of the magnitude of precipitation and the amount of accumulated

    snow is indirect. The magnitude of rainfall plays a role in increasing the amount of

    snowmelt by providing additional energy input to the snowpack. This is especially true

    if the temperature of the incoming rainfall is higher.

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    6/14

    368

    Yeshewatesfa Hundecha

    et al.

    The depth of the accumulated snow also affects the amount of snowmelt. Only the

    upper few centimetres of the snowpack are under the direct influence of the atmos

    phere. Temperature of this layer depends on the atmospheric temperature. Snowmelt

    only occurs from this layer when its temperature reaches 0C and the net energy

    balance is positive. If the temperature of the lower layer is below 0C, part or all of the

    snowmelt is refrozen. The heat released in the process is used to satisfy part or the

    entire heat deficit of the lower snowpack. The amount of heat energy needed to raise

    the temperature of the lower layer from some negative value to 0C depends on the

    volume of this part of the snowpack. When the temperature of the lower snowpack

    reaches 0C, any additional snowmelt is used to satisfy the free water holding capacity

    of the snowpack. Only snowmelt in excess of this free water holding capacity becomes

    available for infiltration and surface runoff (Anderson, 1968).

    Two systems of rules were established to determine the proportion of precipitation

    that is in the form of snow and the amount of snowmelt, respectively. The arguments

    to be used in each sy stem of rules were identified and were div ided into different fuzzy

    classes.

    In the first system of rules where the proportion of precipitation in snow form is

    determined, the only argument is temperature and the consequence is proportion of

    solid precipitation. In the second system of rules, temperature, accumulated snow

    depth, and magnitude of daily precipitation are used as arguments and the consequence

    is the daily amount of snowmelt.

    The temperature values were divided into five fuzzy classes. The magnitude of

    precipitation and accumulated snow were also partitioned into three fuzzy classes. The

    consequences of the rule systems were also fuzzified by dividing the amount of snow

    melt into four fuzzy classes and the proportion of solid precipitation in the form of

    snow into three fuzzy classes. These are summ arized in Tables 1-5.

    Table 1Fuzzy classes of temperature for snowmelt computation.

    Class of temperature

    Cold

    About zero

    Cool

    Warm

    Table 2Fuzzy classes of accumulated snow depth.

    Accumulated snow depth

    Low

    Moderate

    High

    Table3 Fuzzy classes of daily precipitation.

    Daily precipitation

    Low

    Moderate

    High

    Fuzzy number representation

    (C)

    ( _ o o , _ l , 0 ) j -

    ( - 1 , 0 . 1 . 2 ) ,

    ( 1 , 2 , 3 , 4 ) ,

    (3 ,

    4,

    +

    ~ )

    r

    Fuzzy number representation

    (mm water equivalent)

    (0 ,

    20,

    35)

    r

    (20,35,

    45)

    T

    (35,45,

    + ~ )

    r

    Fuzzy number representation

    (mm)

    (0 , 10, 15)j-

    (10, 15 ,20)

    r

    (15 ,20 , + )j-

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    7/14

    Development

    of a

    fuzzy rule-based rainfall-runoff model

    369

    ble4Fuzzy classesofamountofsnowmelt.

    Amountofsnowmelt

    Fuzzy number representation

    (mm day ')

    Low

    Moderate

    High

    Extreme

    (0,

    4, 8)

    r

    (4,8, 15)

    7

    -

    (8,

    15,

    20)

    r

    (15,20,55)7-

    ble5Fuzzy classesofpercentageofprecipitationinsnowform.

    Proportionofprecipitationin

    snow form

    Fuzzy number representation

    ( )

    Nosnow

    Half

    All

    (0,25,5 0)

    r

    (25,50,

    7 5)

    T

    (50,

    75, 100)

    r

    ble6Asystemofrules describing proportionofprecipitationinsnowform.

    Ruleno.

    Argument (temperature)

    Proportionofprecipitationinsnow form

    1

    2

    3

    4

    Cold

    About zero

    Cool

    Warm

    All

    Half

    No

    snow

    Nosnow

    ble7Asystemofrulesfor thedeterminationofsnowmelt.

    Rule no.

    1

    2

    3

    4

    5

    6

    7

    8

    9

    10

    11

    12

    13

    14

    15

    16

    17

    18

    19

    20

    21

    22

    23

    24

    25

    Argument:

    Snow depth

    L ow

    L ow

    L ow

    L ow

    L ow

    L ow

    L ow

    Moderate

    Moderate

    Moderate

    Moderate

    Moderate

    Moderate

    Moderate

    Moderate

    Moderate

    High

    High

    High

    High

    High

    High

    High

    High

    High

    Tempe

    About

    About

    About

    Cool

    Cool

    Cool

    Warm

    About

    About.

    About.

    Cool

    Cool

    Cool

    Warm

    Warm

    Warm

    About:

    About;

    About

    :

    Cool

    Cool

    Cool

    Warm

    Warm

    Warm

    rature

    zero

    zero

    zero

    zero

    zero

    zero

    zero

    zero

    zero

    Precipitation

    High

    Moderate

    L ow

    High

    Moderate

    Low

    -

    High

    Moderate

    Low

    High

    Moderate

    L ow

    High

    Moderate

    L ow

    High

    Moderate

    L ow

    High

    Moderate

    L ow

    High

    Moderate

    L ow

    Amount of snowmelt

    High

    Moderate

    L o w

    High

    Moderate

    L o w

    Extreme

    High

    Moderate

    L o w

    High

    Moderate

    L o w

    Extreme

    Extreme

    High

    Moderate

    L ow

    L o w

    High

    Moderate

    L ow

    Extreme

    Extreme

    High

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    8/14

    370

    Yeshewatesfa Hundechaet al.

    In the first system of rules, for each fuzzy class of temperature values, a

    corresponding fuzzy class of proportion of precipitation in snow form was assigned

    (Table 6). The consequence of each rule was computed for a given temperature value

    and the crisp output of the system of rules was computed by combining the fuzzy

    consequences of each rule using the maximum combination technique and defuzzi-

    fying the com bined fuzzy c onsequenc e using the mean defuzzification techniqu e. The

    same techniques were used in all rule systems in the other modules as well.

    In the second system of rules, the arguments, namely the fuzzy classes of

    temperature, magnitude of precipitation, and accumulated snow, were connected using

    the A N D operator. The complete system of rules is shown in Table 7. No te that the

    increase in temperature and the magnitude of precipitation lead to the increase in the

    amount of snowmelt. The consequence of increase in the amount of accumulated snow

    is to decrease the potential for snowmelt as explained earlier in this section.

    Evapotranspiration

    The fuzzy logic routine for this process was formulated based on the long-term series

    of observed mean monthly temperature and the corresponding mean monthly vapo

    transpiration values. A set of rules was developed in which the incremental tempera

    ture value above the long-term monthly mean value (T - T

    m

    ) is used as an argument

    and the ratio of the actual to the long-term mean monthly vapotranspiration value

    (PEJPEm) is the consequence. The incremental temperature value above the long term

    monthly mean value was divided into seven fuzzy classes and the corresponding

    classification of the consequence was m ade. These are summ arized in Table 8.

    Table 8Summary of the fuzzy rule system used to describe vapotranspiration.

    r-r,(c) PE

    a

    /PE

    m

    (-2,0,2)

    r

    (0.8,1.0,1.2)7-

    (1,3,5),- (1.1, 1.2, 1.3)

    r

    (3,5,7)

    T

    (1.2, 1.4,

    1,

    6)

    r

    (5,7,+oo)

    r

    (1.4, 1.6, 1.8)

    r

    ( -5 , -3 , - l )

    r

    (0.3,0.6,0.8)7-

    (-7,

    -5 ,

    - 3 )

    r

    (0.2,0.3,0.4 , 0.6)*

    ( -00, -7, -5 )7- (0 .1 ,0 ,2 ,0 .3)7-

    Calculation of runoff

    This mod ule com putes the proportion of rain or snowm elt that is converted to runoff at

    a given soil moisture deficit. The proportion of rain or snowmelt that contributes to

    runoff depends on the relative soil moisture. The relative soil moisture is the ratio of

    the actual soil moisture to the field capacity of the soil. The proportion of rain or

    snowmelt that contributes to runoff increases with the relative soil moisture. In the

    HB V m odel, their relationship is expre ssed by an exponential function.

    The fuzzy logic based routine was formulated by using the relative soil moisture as

    the rule argument and the proportion of rain or snowmelt that contributes to runoff as

    the consequence of the rule system. The field capacity and the permanent wilting

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    9/14

    Developmento f a fuzzyrule-basedrainfall-runoff model

    371

    Table 9A rule system for soil water accounting.

    Relative soil moisture (ratio of actual to the

    maximum soil moisture storage)

    (0,0, 0.2, 0.3)

    R

    (0.2,0.3,0.4,0.5)*

    (0.4,

    0.5, 0.6, 0.7)

    R

    (0.6,0.7, 0.8, 0.9)

    R

    (0.8,0.9, 1.0, 1.0)

    R

    Percentage of rain or snow that goes to runoff

    (0,2,4)

    T

    (10,

    15,20)

    r

    (20,

    30,4 0)

    r

    (60,

    70 ,80)

    T

    (75, 90, 100, 100)R

    points for the different zones were defined based on the type of soil in each zone. Five

    fuzzy classes of the relative soil moisture values were established and the

    corresponding classes of percentage of rain or snowmelt that contributes to runoff were

    defined as shown in Table 9.

    The basin response

    This module describes the dynamics of the generated

    runoff,

    and thus its distribution in

    time once the water balance is set by the modules for snowmelt and

    runoff.

    The

    conceptual system used in the HBV model (Bergstrm, 1995) was adapted to use a

    fuzzy logic-ba sed rou tine. The proce ss is conc eptua lized by a fictitious system of two

    reservoirs arranged one over the other in which the outflows from the upper and lower

    reservoirs simulate the direct runoff and the base flow respectively at the outlet of the

    basin. These reservoirs are conceptually defined to simulate movement of the runoff

    within the basin before reaching the outlet of the basin. The infiltration excess water

    computed in the runoff module is input to the upper reservoir. The lower reservoir is

    replenished by percolation from the upper reservoir.

    In formulating the fuzzy logic-based routine of this module, three systems of fuzzy

    rules were utilized to determine the outflows from the two reservoirs and the

    percolation from the upper reservoir to the lower one respectively. The volume of

    water in the upper reservoir was used as an argument in the rule systems established to

    determine the outflow from the upper reservoir and the percolation from the upper to

    the lower reservoir. In the rule system used to determine the outflow from the lower

    reservoir, the volume of water was used as an argument. The volume of water in the

    upper reservoir was classified into 10 fuzzy sets while that in the lower reservoir was

    classified into 15 fuzzy sets to establish the rules.

    Because of the erratic nature of the entire basin, the rules that apply for one part of

    the basin would not apply to other parts of the basin. This would necessitate the

    formulation of different sets of rules for each sub-basin, which would be difficult to

    handle. For this reason, rules were established for three different types of upper and

    three different types of lower reservoirs (Tables 10-12). For each sub-basin, a pair of

    reservoir types was selected and the rules corresponding to the chosen pair were used.

    RESULTS AND DI SCUSSI ON

    Before testing the applicability of the fuzzy logic-based routines for the different

    catchment processes, the HBV model was calibrated and used for the basin.

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    10/14

    37 2 Yeshewatesfa Hundechaet al.

    Table 10A rule system to predict outflow from the upper reservoir.

    Volume of water in the

    reservoir (mm)

    ( 0 , 1 ,

    2)

    T

    ( 1 , 2 , 4 , 8 ) *

    (4 , 8, 10)

    r

    (8 ,

    10, 15)

    T

    ( 1 0 , 1 5 , 2 0 ) *

    ( 1 5 , 2 0 , 2 5 )

    r

    ( 2 0 , 3 0 , 4 0 )

    r

    (30, 40, 45 )

    T

    (40, 45, 55)

    r

    (45 , 55, =o)j.

    Reservoir outf low (mm

    Res. Type 1

    (0.0,0 .17,0.33)7-

    (0.17,0.33,0.5)7-

    (0.33,

    0.5, 0 .67)

    r

    ( 0 . 5 , 0 . 6 7 , 1 . 0 )

    7

    (0.67, 1.0,

    1.33)

    r

    (1 .0 ,1 .33,2.0)7-

    (3 .0 , 3 .33, 3 .67)T-

    (3 .33 ,

    3.67, 4.33)

    T

    (3.67, 6.67,

    10)T-

    (10.0, 13.33, 15.0)7-

    day

    1

    ):

    Res. Type 2

    (0.0, 0.25, 0.5)

    r

    (0 .25,0.83,1.5)7-

    (0.83,1.5 ,2 .17)7-

    ( 1 .5 ,2 .17 , 3 .0 )

    r

    (2 .17,3.0 ,4 .33)7-

    (3 . 0 , 4 . 3 3 ,

    6 . 0 )

    r

    (5 .0 ,8 .33,11.67)7-

    (8.33, 11.83, 15.0)

    r

    ( 11 .83 ,15 .0 ,18 .5 ) 7

    (15.0, 18.5, 21.67)

    r

    Res. Type 3

    (0.0, 0.17, 0.33) ,-

    (0 .17,0.33,0.67)7-

    (0 .33 , 0 .67 .1 .33)

    r

    ( 0 . 6 7 , 1 . 3 3 , 2 . 0 )

    r

    (1 .33 , 2.0, 2.67)7-

    (2.0 , 3 .33, 5 .67)

    r

    (6.67,

    9.0, 10.0)T-

    (9 .33 ,

    10.67, 12.33)7-

    (11.67,12.67, 14.0)7-

    (13.0, 15.0, 16.0)7-

    Table 11A rule system to predict percolation from the upper to the lower reservoir.

    Volume of water in the

    reservoir (mm)

    ( 0 , l , 2 )

    r

    ( 1 , 2 , 4 , 8 ) *

    (4 , 8, 10)

    r

    (8 ,

    10,

    15)T-

    (10, 15, 20)

    r

    (15,20,25)7-

    (20, 30, 40)

    r

    (30, 40, 45)

    r

    (40, 45, 55)

    7

    (45 , 55, )

    r

    Percolation rate (mm

    Res. Type 1

    (0.0, 0 .28, 0 .56)

    r

    (0 .28 ,

    0.56, 0.83)7-

    ( 0 . 5 6 , 0 . 8 3 ,

    1,11)

    T

    ( 0 . 8 3 , 0 . 9 5 , l . l l )

    r

    (0 .95,1.11,1.22)7-

    (1 .11,

    1.39,

    1.67)

    r

    (1.38,

    1.67,

    1.95)

    T

    (1.67, 1.95, 2.22)

    r

    ( 1 . 9 5 , 2 . 2 2 , 2 . 5 )

    r

    (2.22, 2.5, 2.78)T-

    . d a y

    1

    ) :

    Res. Type 2

    (0.0, 0 .43, 0 .86)

    r

    (0.43, 0.86, 1.24)T-

    (0.86,

    1.24, 1.67)T-

    ( 1 . 2 4 , 1 . 6 7 , 2 . 1 5 )

    r

    (1 .67, 2 .15, 2 .54)

    r

    ( 2 . 1 5 , 2 . 5 4 , 3 . 2 7 )

    r

    (2.54, 3.27, 3.89)

    r

    ( 3 .27 , 3 .89 ,4 .17)

    r

    (3 .89,4.17,4.72)7-

    (4.17,4.72,5.17)7-

    Res . Type 3

    (0.0, 0 .3 , 0 .44)

    r

    (0 .3 ,

    0 .5 , 0 .86)

    r

    ( 0 . 5 , 0 . 8 6 ,

    1.06)

    r

    (0.86 , 1.06, 1.3)

    r

    (1.06, 1.3,

    1.47)

    r

    (1 .3 ,

    1.58,

    1.78)

    r

    (1 .58,1.78,2.03)7-

    (1.8 ,2 .03,2.26)7-

    (2 .03 , 2 .26 , 2 .6 )

    r

    (2.26, 2.6, 2.78)7-

    Table 12A rule system to predict outflow from the lower reservoir.

    Volume of water in the

    reservoir (mm)

    (0.0, 5.0, 10.0)7-

    (5.0 ,10.0,20.0)7-

    ( 10 .0 ,20 .0 ,40 .0 ) r

    (20.0, 40.0, 60.0)

    r

    (40.0, 60.0, 80.0)T-

    (60.0, 80.0, 100.0)7-

    (80.0, 100.0, 120.0)7-

    (100.0, 120.0, 140.0)7-

    (120.0, 140.0, 160.0)

    r

    (140.0, 160.0, 180.0)7-

    (160.0, 180.0, 200.0)

    r

    (180.0, 200.0, 220,0)

    r

    (200.0, 220.0, 240.0)

    r

    (220.0, 240.0, 260.0)

    r

    (240.0, 260.0, ~ )

    r

    Outflow rate (mm day

    1

    )

    Res . Type 1

    (0.0,0 .05,0.1)7-

    ( 0 . 0 5 , 0 . 1 , 0 . 1 5 )

    r

    (0 .1 ,0 .15,0.2)7-

    ( 0 . 1 5 , 0 . 2 , 0 . 2 5 )

    7

    (0.2, 0.25, 0 .6)

    r

    (0 .25 , 0.3, 0 .35)

    T

    (0 .3 ,

    0.35, 0.43)r

    (0 .35 ,

    0.43, 0.5)7-

    (0.43, 0.5, 0.58)T-

    (0 .5 , 0 .58 , 0 .65)

    r

    (0 .58 ,

    0 .65 , 0 .73)

    r

    (0 .65 ,

    0 .73 , 0 .80)

    r

    (0.70, 0.80, 0.90)7-

    ( 0 . 8 0 , 0 . 9 0 , 1 . 0 )

    r

    (0.90,

    1.0, 1.10)T-

    1:

    Res. Type 2

    (0.0,0 .08,0.15)7-

    (0 .08 ,

    0.15, 0.25)7-

    (0.15,0.25,0.35)7-

    (0 .25 ,

    0.35, 0 .45)

    T

    (0 .35 , 0.45, 0 .53)

    7

    (0 .45 ,0.5 3, 0.60)j-

    (0 .53 , 0.6, 0 .68)

    r

    (0.6,

    0.68, 0 .75 )

    r

    (0 .68 , 0 .75 , 0 .83)

    r

    (0 .75 , 0 .83 , 0 .90)

    r

    (0 .83 ,

    0.90, 0 .98)

    r

    ( 0 . 9 0 , 0 . 9 8 ,

    1.05)7-

    ( 0 .98 ,1 .05 ,1 .13) 7

    (1 .05 ,

    1.13,

    1.2)7-

    ( l . l , 1 . 2 , 1 . 3 )

    r

    Res. Type 3

    ( 0 . 0 , 0 . 0 5 , 0 . 1 )

    r

    (0 .05 ,

    0.10, 0 .15)

    r

    (0 .1 ,0 .15,0.20)7-

    ( 0 . 1 5 , 0 . 2 0 , 0 . 2 5 )

    r

    (0 .20, 0 .25, 0 .30)

    7

    (0 .25 , 0.30, 0.35)7-

    (0.30, 0.35, 0.40)2-

    (0 .35 ,

    0 .40 , 0 .45)

    r

    (0.40, 0.48, 0.55)7-

    (0 .48 , 0.55, 0.63)7-

    (0 .55 ,

    0 .63 , 0 .70)

    r

    (0 .63 ,

    0 .70 , 0 .78)

    r

    (0 .70, 0 .78, 0 ,85)

    T

    (0 . 7 8 ,

    0.85, 0 .90)

    T

    (0 .85 ,

    0 .90 , 0 .95)

    r

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    11/14

    Development of a fuzzy rule-based rainfall-runoff model

    373

    Applicability of the fuzzy logic-based routines was investigated by incorporating only

    one of the fuzzy rule-based modules at a time while retaining the HBV versions for the

    other modules. Calibration of the modules was performed by using a multi-objective

    calibration procedure (Gupta & Sorooshian, 1998). Three different measures of model

    performance were used to calibrate the modules. These measures are the correlation

    coefficient of the modelled and the observed discharges (RCOEF), the Nash-Sutcliffe

    measure (NS), and the bias (the average of the difference between the modelled and the

    observed discharges). The consequences of each rule system were manually changed

    until acceptable values of all the three measures of model performance were obtained.

    The model performance values were calculated based on simulation of daily discharge

    for a period of 10 years, i.e. N =365 2. Fina lly, all the inde pen den tly ad justed fuzzy

    logic-based routines were coupled together to model the catchment. All computations

    were carried out for a time step of two hours.

    Performance of each of the fuzzy rule-based modules was found to be nearly the

    same as that of the corresponding HBV modules. Generally, the fuzzy rule-based

    routine for snowmelt was observed to show the best performance (Table 13).

    Estimation of the winter high flows by the HBV model was improved a little at some

    observation points by introducing this module (Fig. 4). This is because, in the HBV

    model, the snowmelt was represented by a simple degree-day method in which only

    temperature is considered a driving force for snowmelt. In the fuzzy rule-based

    routine, additional factors that were not considered in the HBV routine were

    incorporated, leading to an improvement in the estimation of the amount of snowmelt.

    On the other hand, the fuzzy rule-based routine for the basin response module was

    found to be the most difficult to handle. Assignment of the combination of upper and

    lower reservoir types to the sub-catchments was undertaken through a kind of trial and

    160.00-]

    140.00 -

    HBVModel

    120.00 - O bserved

    ^ Snow-Fuzzy Model

    ^S 100.00 -

    ff 80.00-

    a

    g 60.00-

    Date

    Fig. 4 Com par ison be tween the HB V and the Sno w-F uzzy models a t one of the

    observa t ion nodes .

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    12/14

    Yeshewatesfa Hundecha

    et al.

    >>

    3

    &.

    >

    N

    s

    tu

    i

    O

    &.

    J

    S

    es

    CO

    >

    N

    S

    P-

    J+-

    o

    c

    3

    C

    > 1

    N

    -3

    &.

    1

    H

    a

    > 1

    N

    N

    3

    eu

    O

    W

    l/i

    >

    S

    s

    o

    ^

    -s:

    oa

    S:

    b .

    b j

    O

    o

    as

    2

    :

    K5

    co

    *

    k,

    fc)

    O

    O

    as

    co

    ^

    sa

    S

    b .

    fc)

    O

    O

    as

    co

    -

    Q

    ^

    %

    k,

    k j

    O

    o

    as

    -o

    oc o

    N

    r~ i r ^ r o o c o

    T t t o c o o r - o o o r ^ o o

    O O h C O C O C O M O O O O O O C ^

    d c d d d o d d c d d

    ON

    (N

    t ~~ o "t r-

    w * o \o h

    co

    *

    d d d d d p C f N

    r-avoo h n ^ c o o e c f o ^ t

    C C h O O C O C C C O O O O C C C C O C O

    n - i - c ^ M c o m C s n

    o o o o - h c o i n o i u w L J i -

    C O C O O O O O M O X O C j & O C O

    d d d d d d d d d d d d d d

    r - - o o \ o i n t n r - s >

    m c o T t m m O c N

    m

    h *

    - 1 1 -

    -*tf- fN ' T f - "

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    13/14

    Development

    of a fuzzy

    rule-based

    rainfall-runoff model

    375

    Feb/85 Feb/

    85

    Mar/

    85

    Apr/

    85

    Apr/85 May/85 Jim/85 Jun/85

    Date

    Fig. 5 Comparison between the HBV and the All-Fu zzy models at one of the

    observation nodes.

    error procedure and the associated large number of rules pertaining to the outflows

    from the reservoirs resulted in difficulty in adjustment of the rales.

    The entirely fuzzy logic-based model was found to reproduce the observed

    discharges well. Generally, under low and normal flow conditions the model

    performed well and no noticeable difference was observed between the HBV and the

    fuzzy rule-based models. The fuzzy rule-based model was observed to overestimate

    the peak flows (Fig. 5).

    C O N C L U D I N G R E M A R K S

    The applicability of a fuzzy logic-based approach to modelling catchment processes

    has been illustrated. When using such an approach, a sound knowledge of the

    underlying physical processes is not prerequisite.. Only knowledge of the factors that

    influence a process and a qualitative relationship between the cause and effect is

    required. Since the mathematical relationship between the cause and effect is not

    necessary, quantities that are not explicitly included in other conceptual or physically

    based models because of the nature of idealization of the process in the models or

    unavailability of a known relationship can easily be included as rule arguments in a

    fuzzy rule-based modelling.

    Parameters are normally required to be determined by field measurements or

    estimated through model calibration in other types of models. Each parameter can have

    different values for different zones of the model area. No model parameters are

  • 7/24/2019 Developement of Fuzzy Rainfall Runoff...

    14/14

    376

    Yeshewatesfa H undecha et al.

    considered in a fuzzy rule-based modelling approach. This makes the approach easier

    and faster to work with.

    REFERENCES

    Abebe, A. I , Solomatine, D. P. & Venneker, R. G. W. (2000) Application of adaptive fuzzy rule-based models for

    reconstruction of missing precipitation events.

    Hydrol. Sci. J.

    45(3) , 425-436 .

    Anderson, E. A. (1968) D evelopment and testing of snowpack energy balance equations.

    Wat. Resour. Res.

    4(1), 19-38.

    Brdossy, A. & Disse, M. (1993) Fuzzy rule-based models for infiltration. Wat. Resour. Res. 29, 3 7 3 -3 8 2 .

    Brdossy, A. & Duckstein, L. (1995)

    Fuzzy Rule-Base d Modeling w ith Applications to Geoph ysical, Biological, an d

    Engineering Systems.

    CRC Press , Inc., Boca Raton, Florida, USA.

    Bergstrom, S. (1995) The HBV model. In:

    Computer Models of Watershed Hydrology

    (ed. by V. P. Singh), 443^176.

    Water Resources Publications, Litt leton, Colorado, USA .

    Gupta, H. V. & Sorooshian S. (1998) Toward improved calibration of hydrologie models: multiple and noncomm ensurable

    measures of information.

    Wat. Resour. Res.

    34(4), 751-763.

    Prakash, A. (1986) Current state of hydrologie modeling and its limitations. In: Proc. Int. Symp. on Flood Frequency and

    Risk Analysis

    (14-1 7 May 1986). Louisiana State University, Baton Rouge, Louis iana /US A.

    See,

    L. & Openshaw, S. (1999) Applying soft computing approaches to river level forecasting.

    Hydrol. Sci. J.

    44(5), 7 6 3 -

    778.

    See,

    L. & Openshaw, S. (2000) A hybrid multi-model approach to river level forecasting.

    Hydrol. Sci. J. 45(4),

    523-536 .

    Sugeno, M. & Y asukawa, T. (1993) A fuzzy-logic-based approach to qualitative modeling. IEEE Trans, on Fuzzy Systems

    1(1),

    7 - 3 1 .

    Welstead, S. T. (1994)

    Neural Network and Fuzzy Logic Applications in C/C++.

    John Wiley & Sons Inc., New York,

    USA.

    Zadeh, L. A. (1965) Fuzzy sets.

    Information and Control

    8(3), 338-353 .

    Zadeh, L. A. (1973) Outline of a new approach to the analysis of complex systems and decision processes.

    IEEE Trans.

    Systems, Man, and Cybernetics 3 , 28-44 .

    Received

    6 June 2000;

    accep t ed

    20 December 2000