You are on page 1of 29

Elsevier Editorial System(tm) for International Journal of Engineering Science

Manuscript Draft
Manuscript Number:
Title: The advantage of sensitivity analysis for global parameters applying to spatially distributed flood
forecasting model
Article Type: Full Length Article
Keywords: Sensitivity analysis; Global parameter; Flood forecasting; WetSpa extension model
Corresponding Author: Mrs. Chi Thi Phuong Pham, M.D.
Corresponding Author's Institution: Geosphere Research Institute
First Author: Chi Thi Phuong Pham, M.D.
Order of Authors: Chi Thi Phuong Pham, M.D.; Chiaki T Oguchi, Dr.
Abstract: Flood prediction models combined with simulation and GIS such as WetSpa series are useful
and have been improved for some decades. However, to predict flooding phenomenon in time is a
challenge for scientists because of necessity of tremendous parameters involving not only hydrological
parameters but also soil physical properties as well as land use conditions In order to evaluate the
importance of parameters for input to simulation model, this study focus on creating new combined
simulating model for flood discharge by the WetSpa extension model by using the global technique
Morris's sensitivity analysis. Hydrological parameters were collected from An Chi station, which
locates in the upstream part of the Ve river basin, the largest rainy region of Quang Ngai province in
the central of Vietnam. The application for flood forecasting purpose of the model shows good, by
evaluating using Nash-Sutcliffe criteria of 0.76. However, the logarithmic version of Nash-Sutcliffe
efficiency for low flow is 0.78 and the adapted version of Nash-Sutcliffe efficiency for high flow
evaluation is 0.74 showed the poor ability of reproducing the time evolution of flows due to the semiopen basin characteristics.

Acknowledgements

Acknowledgements
We would like to thank Dr. Nguyen T.G. and Ms. Nguyen T.T. for academic advice and other
support. We also want to say thank to The Ministry of Education, Culture, Sports, Science and
Technology, Japan for the fund for this research.

Title page with author information

The advantage of sensitivity analysis for global parameters applying


to spatially distributed flood forecasting model
Chi PHAM1 and Chiaki T. OGUCHI2
1

PhD student, Graduate School of Science and Engineering, Saitama University, 255 Shimo-okubo,
Sakura-ku, Saitama 338-0825, Japan

Associate Professor, Geosphere Research Institute, Saitama University, 255 Shimo-okubo, Sakuraku, Saitama 338-0825, Japan

Corresponding author. Tel (81)8044528686; Email: Chitrum@ymail.com

*Manuscript without author details


Click here to view linked References

The advantage of sensitivity analysis for global parameters

applying to spatially distributed flood forecasting model

3
4

ABSTRACT

Flood prediction models combined with simulation and GIS such as WetSpa series are

useful and have been improved for some decades. However, to predict flooding phenomenon in

time is a challenge for scientists because of necessity of tremendous parameters involving not

only hydrological parameters but also soil physical properties as well as land use conditions In

order to evaluate the importance of parameters for input to simulation model, this study focus

10

on creating new combined simulating model for flood discharge by the WetSpa extension model

11

by using the global technique Morriss sensitivity analysis. Hydrological parameters were

12

collected from An Chi station, which locates in the upstream part of the Ve river basin, the

13

largest rainy region of Quang Ngai province in the central of Vietnam. The application for flood

14

forecasting purpose of the model shows good, by evaluating using Nash-Sutcliffe criteria of

15

0.76. However, the logarithmic version of Nash-Sutcliffe efficiency for low flow is 0.78 and the

16

adapted version of Nash-Sutcliffe efficiency for high flow evaluation is 0.74 showed the poor

17

ability of reproducing the time evolution of flows due to the semi-open basin characteristics.

18

Keywords: Sensitivity analysis; Global parameter; Flood forecasting; WetSpa extension model

19

1. Introduction

20

Flood prediction using simulation of discharge is important to reduce the unpleasant effects

21

of this hazard causing on the nature and human livings. Usually this progress of simulation

22

meets difficulties due to the limitation of data as well as the available technology. For some

23

decades, people tend to use the Geographical Information Systems (GIS) hydrological models to

24

imitate flood hydrographs due to their easy-to-use characteristics. However, to predict flooding

25

phenomenon in time is a challenge for scientists because many input parameters are necessary

26

for each model. The value of parameters must be estimated from the topography of the basin,

27

physical properties of soil type, aquifer zones or land use conditions and so on. These factors

28

cannot be measured easily, especially for large areas. For this reason, optimization using a

29

certain initial values, after that, adjusted parameters are useful for higher efficiency of models.

30

This optimization steps take time, so that sensitivity analyze is necessary to find the important

31

parameters in calibration processes. The meaning of doing sensitivity analysis is to be

32

understood the sense of each parameter used in the models, which are the most important ones

33

for preliminary calibration of models. It can assess the effects of inputs on outputs of the model

34

by investigating only several parameters. Definition of sensitive parameters or reducing the

35

number of adjusted inputs leads to better estimation of their values and to shorten the operating

36

time for modeling. From that, better predicting water discharge and managing watershed can

37

minimize the damages of flood phenomenon.

38

There have been done the sensitivity analysis for the WetSpa (Water and Energy Transfer

39

between Soil, Plants and Atmosphere) extension model. Liu and De Smedt [1] described the

40

order of parameter priority and gave relative sensitivity of the variables. The authors not only

41

suggested the order in which parameters are typically adjusted, but also showed the aspect of the

42

output is primarily affected. Bahremand and De Smedt [2] used a model-independent nonlinear

43

parameter optimization for a local analysis. The noticeable recommendation for further studies

44

is using a global sensitivity analysis instead of the local parameter estimation (PEST) method.

45

Because the PEST design uses only local information that might not represent the entire

46

parameter space, the results yield only local sensitivities or uncertainties.

47

The aim of this research is using the Morris [3] method for global sensitivity analysis of

48

parameters in the WetSpa Extension model, and then, applying the results in simulating flood

49

hydrographs for the Ve catchment in central Vietnam.

50

2. Study area

51

The Ve river basin is located in Quang Ngai Province in the central region of Vietnam. The

52

basin has a catchment area of 1300 km2 at outlet and the total length of main stream from

53

upstream to the coast is 91 km [4]. The river flows in a generally north-east direction towards

54

the coast. The upper reaches of the catchment is steep mountainous region with the highest point

55

having an elevation of over 1100 m. The floodplains near the coast are flat and the mouth of the

56

river joins the mouth of the Tra Khuc River through a coastal water course and a network of

57

minor channels. The inundated areas of the Ve basin join the floodplain of the Tra Khuc basin in

58

the north and also extend along the coast towards the south. These flow patterns in the

59

floodplain are complex with a network of channels and branches in the downstream. The

60

National Highway and Railway Line both cross the river in the lower section at approximately

61

10 km upstream of the mouth of the river.

62

An Chi station, which is one of the main stations in the province is located where the

63

irrigation channel crosses the river in a siphon. The gauge is also located at a point immediately

64

upstream of where there is an extensive area of floodplain inundation, while the flow is

65

relatively constricted at the station. In major floods with significant downstream inundation,

66

there is a backwater effect from the downstream floodplain. Within this research, the basin of

67

the upstream area of An Chi station was taken into account. This surface area of the basin is 757

68

km2. The terrain of steep slope in a small area makes flood process complicated. Furthermore,

69

the location of the catchment corresponds to heavy rainfall area in the region, so that flash flood

70

often occurs, causing serious damages to the human community. The problems need to be

71

solved here is to raise the degree of accuracy in flow forecasting and to extent the foresee time

72

of predicting water level in order to prevent and reduce damages caused by flood.

73

3. Flood forecasting model

74

There have been a large number of flood forecasting models used in Vietnam. Among of

75

those, several classical models should be mentioned as for not only the purpose of research as

76

well as the practical forecasting; e.g., the HEC model system of United States Army Corps of

77

Engineers or MIKE of Denmark Hydraulic Institute, and BASIN model of United States

78

Environmental Protection Agency. Recently, the new model WetSpa Extension has been tested

79

in the karstic Suoi Muoi region for modeling flood processes by Liu et al [5]. This model is

80

suitable for flood prediction and also able to simulate the spatial variation of other hydrological

81

characteristics such as ground water and water level at each time step. The same model has been

82

used for flood simulating of Ca river basin in Vietnam [6]. For further applying, it should be

83

continued the research on the applicable of this model for more case study in order to access the

84

reliability for Vietnamese catchment.

85

The WetSpa extension model [1] was an extension of the original WetSpa by Wang and

86

Batelaan [7]. Another extension WetSpass also has been developed by Batelaan and De Smedt

87

[8] for regional groundwater flow systems while the extension WetSpa was proposed for flood

88

prediction as well as water balance simulation. This model is a kind of GIS-based models and

89

suitable on catchment scale. It accounts for spatially distributed hydrological and geophysical

90

characteristics of a catchment with simulating runoff and other hydrological data at certain

91

points in the river networks and their distribution in grid cell scale. Further, the model can

92

provide easy-operating interface, which is suitable for studying the impact of input change on

93

the hydrological behaviors of a basin.

94

4. Sensitivity analysis

95

There are different methods for sensitivity analysis applying to refine the parameter

96

proceeding to calibrate. Werner et al [9] used Generalized Likelihood Uncertainty Estimation

97

(GLUE) to assess the uncertainty value of the land using distribution in the interaction 1D and

98

2D hydrodynamic models in Meuse river basin in the Netherlands. Bahremand and De Smedt

99

[1] validated and performed sensitivity analysis for some global parameters using a

100

model-independent Parameter Estimator (PEST) with WetSpa model for the Torysa basin which

101

has a surface area of 1297 km2 in Slovakia. Fedak [10] has studied the influence of grid cell size

102

in the two models of HEC-1 and TopModel for the Back Creek watershed in America. There are

103

some other researches were reported (e.g. Iman and Helton [11], Campolongo and Saltelli [12],

104

Nguyen and De Kov [13])

105

Selecting the sensitivity analysis methods is usually based on the complexity of the model

106

and analyze target. Morgan and Henrion [14] gave four selection criteria as follows: 1)

107

uncertainty in the model form, 2) the essence of that model, 3) analyzing requirements, and 4)

108

the base conditions. Based on these criteria, a new method developed by Morris was used in this

109

research. This design of Morris has been used widely and proved to be an effective global

110

sensitivity analysis tool in previous studies (e.g. Morris [3], Saltelli et al [15], Nguyen and De

111

Kok [13]).

112

The original purpose of Morris method is to improve the economy of a sensitivity analysis

113

basing on the construction of an input factor value j by k matrix with rows that represent runs

114

and columns stands for parameters. The progress would be repeated many times, it takes only

115

one parameter in each run, and then the number of steps is the linear function of the number of

116

parameters. During the simulation, the corresponding experiment provides k elementary effects

117

from (k+1) running steps. The economy of a design will be defined to be the number of

118

elementary effects divided by the number of experimental runs. Therefor this method has an

119

economy of k/(k+1). The main advantage of this design is low computing cost.

120
121

To define the elementary effect dj which is the change in output caused by the change in
input factor, first we suppose that the output function is as below:

122

y = f(x1,x2,...,xk)
123
124

(1)

The values of the elementary effects could be computed as below.

dj

y ( x1 , x 2 ,...x j 1 ,...x k ) y ( x1 , x 2 ,...x j ,...x k )


x j 1 x j

(2)

125

Where, j indicates the jth input. For more explanation, readers are recommended to reference

126

from Morris [3].

127

The sensitivity analysis to find out the most influential inputs can be done by calculating

128

the standard deviation and mean of the elementary effect value series. The high mean value

129

shows the important global impact when large standard deviation corresponds to the interaction

130

with other factors or nonlinear impact.

131

5. Global parameters for sensitivity analysis

132

There are many parameters included in the WetSpa extension model; however, this study

133

put only seven parameters into sensitivity analysis applied for the Ve catchment. The brief

134

description about these seven parameters will be given as below. These global parameters have

135

physical interpretations that are important in controlling runoff production and hydrographs at

136

the basin outlet. Nevertheless, it is difficult to assign them properly over a grid scale. Therefore,

137

it is preferable to calibrate these parameters against observed runoff data in addition to the

138

adjustment of the spatial distributed model parameters. The initial values were determined based

139

upon known values in previous simulation studies of Bissen Catchment in Luxembourg (Liu

140

and De Smedt [9]). Then the calibration progress was done by the trial simulation. The outputs

141

from the model were compared with observed stream flows, and evaluated by assessment

142

criteria to achieve fit of runoff volumes, peak discharge, delay time and fit of hydrograph shape

143

as well as to refine hydrograph fit. Based upon those comparisons and evaluations, parameter

144

adjustments are made to improve the performance of the model.

145

5.1. Evaporation recession coefficient (Kr)

146

Potential evapotranspiration represents the evapotranspiration rate of a uniform-height

147

green crop and with adequate water status in the soil profile. It reflects the energy available for

148

evaporation and the ability of the lower atmosphere to transport evaporated moisture away from

149

the land surface. Actual evapotranspiration is regarded as potential evapotranspiration when

150

there is sufficient water. Potential evapotranspiration is usually measured indirectly, from other

151

climatic factors, although it depends on local factors such as existence of free water, soil or

152

vegetation types and elevation. For applying this global parameter to WetSpa extension, the

153

parameter value is necessary to be estimated from data from several meteorological station

154

distributed in the target catchment. To account for variation of the values, a correction factor is

155

required in the computed potential evapotranspiration. It is normally close to 1, and can be

156

calibrated by the model through a long-term water balance simulation.

157

In case of lacking of the evapotranspiration data, the correction factor for potential

158

evapotranspiration would be calculated using the below equation representing a part of

159

precipitation (%):

160

xinput x K r

(3)

161

Where, xinput is the input precipitation, x is the measured precipitation, Kr is the evaporation

162

recession coefficient (%).

163

5.2. Scaling factor for interflow computation (Ki)

164

Interflow or subsurface runoff is an essential runoff component for the humid temperate

165

region especially for the sloping areas with dense vegetation. It is assumed to occur when soil

166

moisture exceeds the field capacity and there is sufficient hydraulic gradient to move the water.

167

Even though a uniform soil matrix is considered in the model, but in fact, the porosity and

168

permeability of soil tend to decrease with depth due to overlying soils and the translocation of

169

material in percolating water to lateral subsurface flow. Quick movement of soil water to a

170

stream through root canals, animal tunnels, or pipes produced by subsurface erosion may also

171

become a critical component of peak flow. Taking these effects into account, Ki which is

172

generally greater than 1, should be calibrated by comparing the recession part of computed

173

flood hydrographs with the observed hydrographs.

174

5.3. Groundwater recession coefficient (Kg)

175

The groundwater recession coefficient (Kg) reflects the storage characteristics of the

176

sub-watershed and, therefore, is the same for all hydrographs at a given location. It will remain

177

constant if storage and discharge volumes are divided by area and expressed as depth in mm.

178

This is under the condition that groundwater flow for each sub-catchment has the same

179

recession constant, and total groundwater at the outlet of the river is only a time-shifted

180

superposition of partial groundwater flow from each sub-catchment.

181

The general groundwater flow equation can be expressed as

QGs t K g SGs t / 1000

182

(4)

183

where QGs is the average groundwater flow at the sub-catchment outlet (m3/s), SGs (mm) is the

184

groundwater storage of the sub-catchment at time t, the exponent m is set to 1 for linear

185

reservoir, and to 2 for non-linear reservoir, Kg is a groundwater recession coefficient taking the

186

sub-catchment area into account, has a dimension of (m2/s) for linear reservoir and (m-1s-1) for

187

non-linear reservoir, which is dependent upon area, shape, pore volume and transmissivity of

188

the sub-catchment.

189

In real river basins, base flow recession coefficient for each sub-catchment may not be the

190

same, and may have a considerable deviation from the theoretical constant. For model

191

simplification, a general value of groundwater flow recession coefficient is determined at the

192

basin outlet.

193
194

The equation can be expressed as:

K g ,s K g

Ss
Ws
S

(5)

195

Where Kg,s and Kg (m2/s) are groundwater recession coefficient of the sub-catchment and the

196

entire basin, Ss and S are average slope of the sub-catchment and the entire basin, and Ws is the

197

areal weight of the sub-catchment. Calibration of this parameter is necessary by comparing the

198

computed and observed low flow hydrographs.

199

5.4. Initial soil moisture (Kss)

200

Soil moisture content is a key element in the model controlling the hydrological processes

201

of surface runoff production, evapotranspiration, percolation and interflow. Proper initial soil

202

moisture (Kss) condition may provide a much more realistic starting point for predictions.

203

However, for a long-term flow simulation in a watershed, the initial soil moisture condition is

204

less important, as it affects the hydrological processes only in the initial part of the simulation.

205

An assumption of uniform initial moisture distribution can be made in this case with modelling

206

purpose of flood prediction under present condition. A ratio against field capacity is then

207

defined in the input parameter file for setting up the initial soil moisture conditions. This value

208

can be adjusted during calibration by analysis of water balance output and comparison between

209

the computed and observed hydrographs for the initial phase.

210

5.5. Initial groundwater storage

211

Groundwater balance is maintained on sub-catchment scale and for the active groundwater

212

storage, which is that part of storage in perched or shallow aquifers that contribute to the surface

213

stream flow. Water percolating from the root zone storage may flow to active groundwater

214

storage or may be lost by deep percolation. Active groundwater eventually reappears as base

215

flow, but deep percolation is considered lost from the simulated system. A value of initial

216

groundwater storage (G0) in depth (mm) is set up in the input parameter file for all

217

sub-catchment. This value can be adjusted during calibration by comparing the computed and

218

observed low flows for the initial phase.

219

5.6. Surface runoff exponent for a near zero rainfall intensity (Krun)

220

Rainfall intensity has a big influence in controlling the proportion of surface runoff and

221

infiltration. For a near zero rainfall intensity, this surface runoff exponent (Krun) is assumed to

222

be a variable starting from a higher value, and changing linearly up to 1 along with the rainfall

223

intensity, when the predetermined maximum rainfall intensity is reached. If an exponent value 1

224

is given, the actual runoff coefficient is then a linear function of the relative soil moisture

225

content, and the effect of rainfall intensity on the runoff coefficient is not taken into account.

226

5.7. Rainfall intensity corresponding to a surface runoff exponent of 1 (Pmax)

227

This rainfall intensity (Pmax) corresponds to a threshold rainfall intensity in unit of mm/h of

228

the model simulation at the surface runoff exponent of 1. The actual runoff coefficient becomes

229

a linear function of the relative soil moisture content. A constant value is assumed in the current

230

model for simplification. This parameter is in fact spatially distributed, depending upon the cell

231

characteristics, such as soil type, land use, and slope, etc. Calibration of this parameter can be

232

performed by comparison of the observed and computed surface runoff volume and the peak

233

discharge for high floods.

234

6. Sensitivity of parameters

235

Initial calibration was performed to find variation limits of the parameters by using the two

236

methods of Random sampling and Latin hypercube sampling by Doldersum [16]. From these

237

calibration processes, minimum and maximum values are obtained (Table 1).

238

The sensitivity analysis was done by Morris [3] method for the two output factors peak

239

discharge and the total discharge volume of flow data series at An Chi station. The below results

240

were achieved based on the data collected in the two storms of November and December 1999.

241

Figure 2 presents the sensitivity of the model parameters regarding to the highest point of

242

each flood showing by arrow in Fig. 1. Parameters Kg and Krun have high standard deviation,

243

indicating strong interaction with other parameters. Kr, Ki and Pmax have relatively large

244

standard deviations that demonstrate the ability to interact with each other and with other

245

parameters. Kr, Ki and Kg have high mean values, which represents the influence on peak flow

246

values. Pmax have relatively large mean value, corresponding to significant influence to the

247

output. The two factors Kss and G0 are not sensitive to the peak discharge of stream flow.

248

The results of sensitivity analysis for the total value showed clearly that parameters Kg have

249

a very high standard deviation, indicating a strong interaction with other parameters. Parameter

250

Kr, Ki and Krun have relatively large standard deviations demonstrate their interaction ability.

251

Parameter Kr, Ki and Kg have a high average value shows the influence on flooding volume. The

252

remaining factors are not sensitive in this case.

253

From the above results, apparently that the groundwater recession coefficient Kg is the most

254

sensitive one for both the flood peak and the total volume of flooding water amount. Beside of

255

that is the interaction with other parameters in the model. This is the most important factor

256

worth to notice in the calibration period.

257
258
259

The standard deviation of Krun which represents the surface runoff exponent for a near zero
rainfall intensity is also rather high, reflecting the ability to interact with other parameters.
The evaporation recession coefficient Kr and the interflow coefficient K3 has strongly

260

influence on the peak as well as the total amount stream flow.

261

7. Flood forecasting

262

7.1. Input data

263

Spatial data for the WetSpa extension includes three base digital maps: digital elevation

264

map (DEM), soil type map and land use map. In addition, to compare and calculate basin

265

characteristics, the river networks as well as the hydrological station network maps are also

266

needed. All these maps have the grid cell size of 90 m.

267

Precipitation data at 4 stations An Chi, Son Giang, Gia Vuc and Ba To were used to

268

calculate the stream flow in the basin. Among of those, Ba To and An Chi stations are located

269

inside the basin while the two other gauges are outside. The hourly rainfall data calculated from

270

the 6h-data measured in these 4 stations were used to draw the Thiessen polygons and

271

interpolate data over the entire basin. For the demand of continuous input data for running the

272

model, the interpolating method was also used for some missing data during the executed time.

273

The discharge data of seven years from 1999 to 2005 was used for model calibration. The

274

observed data for each six hours during flood season from September to December was

275

interpolated into hourly values.

276
277

Flow data series at An Chi station was used to compare to output results from WetSpa
model. The data was collected in the two storms of October and November 2008.

278

7.2. Simulating the output flow

279

The suggested progress for calibration of twelve global parameters in the WetSpa extension

280

model is basing on the evaluation results and the visual comparison between observed and

281

calculated hydrographs, readjusting global parameters in the input file, then running the model

282

with a semi-distributed approach. These steps would be repeated until a good match is reached

283

(Liu and De Smedt [9]).

284

Taking advantage of results from the sensitivity analysis for the flood forecasting, in order

285

to shorten the calibrating time, the authors focus on the values of four parameters Kg, Krun, Kr,

286

and Ki but not all the global parameters.

287

Automatic operation of the model was carried out based on the modified source code of the

288

original model. Then instead of calculating the output volume for each certain set of parameters

289

one-by-one, the model can execute with all the set in one-time operation. The concerning output

290

was runoff data at downstream An Chi station corresponding to each set of parameters.

291

In this study, one hundred values randomly chosen from the variation range in Table 1 for

292

each factor were executed. Therefore, in total there were four hundred parameter sets

293

corresponding to four hundred running steps. This progress saved nearly a half of consuming

294

time when compare to operating without sensitivity analysis.

295

The obtained parameter set are shown as below in Table 2.

296

The discharge graph simulated from the model performs the good agreement with the

297

measured data. The model bias is 0.045 for the simulation period. Model determination

298

coefficient is 0.86 and the flow efficiency coefficient is 0.76. Adapted version of Nash-Sutcliffe

299

efficiency for high flow and low flow evaluations are 0.74 and 0.78, respectively. These poor

300

performances may due to the semi-open basin condition. Under normal weather conditions it is

301

a closed basin, but in extreme circumstances water can flow in and out through the boundary.

302

These evaluations show the ability of producing a good flow hydrographs for the study area of

303

Ve catchment, however the accuracy of peak discharge prediction needs to be improved.

304

8. Conclusions

305

This study focus on combining flood discharge prediction by the WetSpa extension model

306

and sensitivity analysis by the global technique Morriss [3]. When using WetSpa extension for

307

flood forecasting, it is need to focus on adjusting values of four global parameters: the

308

groundwater recession coefficient, the surface runoff when the rainfall intensity is very small,

309

the evaporation recession coefficient, and the scaling factor for interflow computation.

310

Basing on the model evaluations, the model is available to application for flood prediction

311

in Ve catchment. The achieved Nash efficiency for reproducing the river discharges of 76%

312

shows good agreement with observed hydrographs. However, to apply in flood forecasting, it is

313

needed to test the model for some other catchment areas.

314

References

315

[1].

Liu Y.B., De Smedt F., 2004. Documentation and User Manual WetSpa Extension; A GIS

316

based Hydrologic Model for Flood Prediction and Watershed Management, Vrije

317

Universiteit Brussel; Department of Hydrology and Hydraulic Engineering.

318

[2].

319
320

Analysis in Torysa Watershed, Slovakia, Water Resources Management, 22, 393-408.


[3].

321
322

Bahremand A., De Smedt F., 2008. Distributed Hydrological Modeling and Sensitvity

Morris D.M., 1991. Factorial sampling plans for preliminary computational experiments,
Technometrics, 33, 161-174.

[4].

Nguyen T.S, 2008. Simulating the rainfall-runoff process for rational water resources and

323

land use in some upstream watersheds in central region of Vietnam. PhD thesis, Hanoi

324

University of Science, Vietnam.

325

[5].

Liu Y.B, Batelaan O., Huong N.T., Tam, V.T., De Smedt F., 2004. Flood prediction in the

326

Karstic

327

Transdisciplinary Conference on Development and Conservation of Karst Regions,

328

Hanoi, Vietnam.

329

[6].

330
331

Suoi

Muoi

catchment,

Vietnam,

Proceedings

of

the

International

Nguyen T.T, 2008. Using the WetSpa extension model for flood simulating of Ca river
basin. Bachelor thesis, Hanoi University of Science, Vietnam.

[7].

Wang, Z.M., Batelaan, O. and De Smedt, F., 1996. A distributed model for water and

332

energy transfer between soil, plants and atmosphere (WetSpa), Physics and Chemistry of

333

the Earth, 21(3), 189-193.

334

[8].

Batelaan, O., De Smedt, F., 2001. WetSpass: a flexible, GIS based, distributed recharge

335

methodology for regional groundwater modelling, In: Gehrels, H., Peters, N.E., Hoehn,

336

E., Karsten, J., Leibundgut, C., Griffioen, J., Webb, B. & Zaadnoordijk, W.J. (eds),

337

Impact of human activity on groundwater dynamics, IAHS Publ. 269, 11-17.

338
339
340
341

[9].

Werner M.G.F., Hunter N.M, Bates P.D., 2005. Identifiability of distributed floodplain
roughness values in flood extent estimation, Journal of Hydrology, 314, 139157.

[10]. Fedak R., 1999. Effect of Spatial Scale on Hydrologic Modeling in a Headwater
Catchment, Master thesis, State University, The United States of America.

342
343

[11]. Iman R.L., Helton J.C., 1988. An investigation of uncertainty and sensitivity analysis
techniques for computer models, Risk Analysis, 8, 71-90.

344

[12]. Campolongo F., Saltelli A., 1997. Sensitivity analysis of an environmental model: an

345

application of different analysis methods, Reliability Engineering & System Safety, 57,

346

49-69.

347

[13]. Nguyen, T.G., De Kok J., 2006. Systematic testing of an integrated systems model for

348

coastal zone management using sensitivity and uncertainty analyses, Environmental

349

Modeling & Software, 22, 1572-1587.

350

[14]. Morgan, M.G., Henrion, M., 1990. Uncertainty: A Guide to Dealing with Uncertainty in

351

Quantitative Risk and Policy Analysis. Cambridge University Press, The United States of

352

America.

353
354
355
356

[15]. Saltelli, A., Chan, K., Scott, E., 2000. Sensitivity Analysis, Chichester: John Wiley and
Sons Ltd.
[16]. Doldersum T, 2009. Global sensitivity analysis of the WetSpa model, Bachelor thesis,
Twente University, Enschede, The Netherlands.

Figure1

Water discharge (m3/s)


3500

Peak discharge

3000
2500
2000
1500
1000
500
0

Fig. 1. Flow data series at An Chi station on Ve River

Figure2

s
600
Kg

Krun

500
400
300

Ki
200
Kr

Pmax

100

Kss

G0

0
-400

-200

m
200

400

600

800

Fig. 2. Sensitive distribution of parameters to the peak discharge

Figure3

s
70
Kg

60
50
40
30

Krun

20

Ki

10

Kss
Pmax

G0

0
-20

-10

Kr

10

20

30

40

50

60

70

80

Fig. 3. Sensitive distribution of parameters to the total discharge volume

Figure4

Discharge (m3/s)
1600
1400
1200

Qsimulated

1000

Qobserved

800
600
400
200
0

Fig. 4. Flood hydrographs of October-November 2008

Table1

Table 1. Variation range of the parameters


Parameter
Minimum value
Maximum value

Kr
0.9
1.1

Ki
2
11

Kg
0.002
0.06

Kss
0
1.5

G0
0
50

Krun
0
10

Pmax
0
500

Table2

Table 2. Parameter for flood simulation


Parameter

Kr

Ki

Kg

Kss

G0

Krun

Pmax

Value

1.1

5.0

0.04

0.0

16.67

3.33

333.33

You might also like