CORE Metadata, citation and similar papers at core.ac.

uk
Provided by Elsevier - Publisher Connector

Available online at www.sciencedirect.com

Procedia
Environmental
Procedia Environmental
Procedia Sciences
Environmental 8 (2011)
Sciences 1786–1796
13 (2012) 1760 – 1770
Sciences
www.elsevier.com/locate/procedia

The 18th Biennial Conference of International Society for Ecological Modelling

Model calibration and uncertainty analysis for runoff in the

Chao River Basin using sequential uncertainty fitting
F.F. Tanga, H.S. Xua,b, Z.X. Xua*
a
Key laboratory of Water and Sediment Sciences, Ministry of Education; College of Water Sciences, Beijing Normal University,
Beijing 100875, China
b
Institute of Resources and Environment, Henan Polytechnic University, Jiaozuo 454003, China

Abstract

The Chao River is one of the most important surface water sources for drinking water in Beijing. Due to the impacts
of human activities and climate change, the Chao River basin is facing water scarcity. Therefore, it is very important
to effectively manage water resources, while the distributed watershed model is the useful and effective tool. Soil and
Water Assessment Tool (SWAT) was selected to set up hydrological model in the Chao River basin. Model
calibration and uncertainty analysis were performed with Sequential Uncertainty Fitting (SUFI-2), which is one of the
programs integrated with SWAT in the package SWAT-CUP (SWAT Calibration and Uncertainty Programs). Results
showed that the p-factor was 0.85 and the r-factor was 1.12 in calibration period (1995-1999) while the p-factor was
0.83 and the r-factor was 2.15 in validation period (2000-2002). When values of p-factor and r-factor are accepted,
further goodness of fit can be quantified by the coefficient of determination (R 2) and Nash–Sutcliffe coefficient (NS)
between the observed and the final best simulated data. The results indicated that R2 was 0.90 and NS was 0.88 in
calibration period, while R2 was 0.77 and NS was 0.74 in validation period. The results of calibration and uncertainty
analysis were satisfactory.
© 2011 Published by Elsevier B.V. Selection and/or peer-review under responsibility of School of
© 2011 Published by Elsevier Ltd.
Environment, Beijing Normal University. Open access under CC BY-NC-ND license.
Keywords: SWAT; Chao River; Runoff; SUFI-2; p-factor; r-factor

* Corresponding author. Tel.: +0-086-010-58801136; fax: +0-086-010-58802739.

E-mail address: [email protected].

1878-0296 © 2011 Published by Elsevier B.V. Selection and/or peer-review under responsibility of School of Environment, Beijing Normal University.
Open access under CC BY-NC-ND license. doi:10.1016/j.proenv.2012.01.170
 F.F. Tang
F.F. et al.
Tang / Procedia
et al./ Environmental
Procedia Sciences
Environmental 13 8(2012)
Sciences (2011)1760 – 1770
1786–1796 1787 1761

1. Introduction

The Chao River basin is one of the surface water sources for drinking water in Beijing and the
drainage watershed of Miyun reservoir. In recent years, the Chao River basin is facing water scarcity due
to the impact of human activities and climate change. Therefore, it is very important to effectively
manage water resources. Distributed watershed model is a useful and effective tool to manage water
resources. In recent years, many hydrological models have been developed such as AGNPS (Agricultural
None Point Source model) [1], SWAT (Soil and Water Assessment Tool) [2] and HSPF (Hydrologic
Simulation Program – Fortran) [3]. These watershed models can be used in several areas including
integrated watershed management, peak flow forecasting, test of the effectiveness of measures for the
reduction of non-point source pollution, soil loss prediction, assessment of the effect of land use change,
analysis of causes of nutrient loss, and climate change impact assessment. SWAT has extensive
application all over the world because of easy availability and friendly interface [4]. For this reason,
SWAT model was selected to estimate runoff in the Chao River basin.
As distributed watershed models are increasingly being used to support decisions on alternative
management strategies, it is very important for these models to carry out a careful calibration and
uncertainty analysis. However, calibration of watershed models is a challenging and time consuming task
because of input, model structure, parameter, and output uncertainty. The definition and quantification of
model uncertainty had become the subject of considerable research in recent years because distributed
hydrological modelling is subject to large uncertainties. Up to now, researchers have developed various
uncertainty analysis techniques for watershed models, such as, Markov chain Monte Carlo (MCMC)
method [5-7], generalized likelihood uncertainty estimation (GLUE) [8], parameter solution (ParaSol) [9],
and sequential uncertainty fitting (SUFI-2) [10]. SWAT-CUP (SWAT-Calibration and Uncertainty
Programs) links GLUE, Parasol, SUFI-2 and MCMC procedures to SWAT. It enables sensitivity analysis,
calibration, validation and uncertainty analysis of SWAT models. SUFI-2 is the more frequency used and
calibration and uncertainty analysis method [7, 10-12].
Against this background, the main objective of this study was first to simulate hydrological process of
the Chao River basin using SWAT. Secondly, model calibration and uncertainty analysis of SWAT model
were performed using SUFI-2.

2. Materials and methods

2.1. Study area and input data

The Chao River basin is located in North China with a drainage area of 6277 km2 (Fig. 1). The Chao
River is one of two tributaries flowing into the Miyun reservoir, which is an important drinking water
reservoir for Beijing City and provides nearly half of the city’s water supply. The Chao River originates
from the northern part of Huangqi Town located in Fengning Manchu Autonomous County in Hebei
Province, flows through Luanping County, then runs down to Gubeikou Town in Miyun County and
empties into Miyun reservoir near Xiahui Village. The main tributaries in upstream of the Chao River
include Andamu River and Xiaotang River. There are another two tributaries. One directly flows into
Maoniu River, another runs down to Qingshui River going through Xinglong County and Miyun County.
The characteristic of the climate is temperate continental and semi-arid and semi-humid. Mean air
temperature is 7.3℃-10.3℃. Mean annual precipitation is 470-731 mm. Mean precipitation is 415 mm in
the northwest part while the mean precipitation is 719.1 mm in the southeast part. Precipitation mainly
concentrated in flood seasons from June to September, accounting for 64%-89% of the total annual
1762
1788 F.F.F.F.
TangTang et al.
et al./ / Procedia
Procedia Environmental
Environmental Sciences
Sciences 13 (2012)
8 (2011) 1760 – 1770
1786–1796

precipitation. Maximum runoff is mainly concentrated in late July to mid-August, and the runoff from
June to September accounts for 70% of the total annual runoff.

Fig.1 The map showing precipitation stations and hydrological stations in the Chao River basin

Spatial data used in the study included a digital elevation model (DEM), land use and land cover, soil
type. Additionally, meteorological input data including precipitation, temperature, and solar radiation,
discharge data, nonpoint sources data and point sources data, etc, are also required. The detailed
information about the database was listed in table1.

Table 1 Input data required in SWAT

 F.F. Tang et al. /etProcedia
F.F. Tang Environmental
al./ Procedia Sciences
Environmental 13 (2012)
Sciences 17601786–1796
8 (2011) – 1770 1789 1763

Data type Data description Sources

DEM 1:2500000; ESRI grid

Land use and land cover
1:1000000; Arc/Info coverage Data Center for Resources and Environmental
map
Sciences Chinese Academy of Sciences (RESDC)
Soil type map 1:10000000; Arc/Info coverage
Digital stream network 1:2500000; Arc/Info coverage
Climatic stations, precipitation, temperature, solar China Meteorological Administration National
Meteorological data
radiation and relative humidity data Meteorological Center
Discharges data Monthly average discharge and from 1995 to 2002 Hydrology statistical yearbooks
1764
1790 F.F.F.F.
TangTang et al.
et al./ / Procedia
Procedia Environmental
Environmental Sciences
Sciences 13 (2012)
8 (2011) 1760 – 1770
1786–1796

2.2. SWAT model

SWAT model is a basin-scale, continuous time model that operates on a daily time step and evaluates
the impact of management practices on water, sediment and agricultural chemical yields in ungauged
basins [2]. The model is physically based, computationally efficient, and capable of continuous
simulation over long time periods. The model’s major components include weather, hydrology, erosion,
soil temperature and properties, plant growth, nutrients, pesticides, land management [13-14].
In SWAT model, a watershed is divided into multiple subwatersheds, which are then further
subdivided into hydrologic response units (HRUs) that consist of homogeneous land use, management
and soil characteristics. For each HRU, water balance was simulated for four storage volumes: snow, soil
profile, shallow aquifer, and deep aquifer.
SWAT provides two methods for estimating surface runoff: SCS curve number procedure and the
Green & Ampt infiltration method. Numerous methods have been developed to estimate PET. Three of
these methods have been incorporated into SWAT: Hargreaves method [15], Priestley-Taylor method [16]
and Penman-Monteith method [17-18]. The model can also read in daily PET values if the user prefers to
apply a different potential evapotranspiration method. Groundwater flow contribution to total streamflow
is simulated by creating shallow aquifer storage [19]. Percolation from the bottom of the root zone is
considered as recharge to the shallow aquifer. More information about SWAT model can be found in Soil
and Water Assessment Tool Theoretical Documentation or link http://swatmodel.tamu.edu/swat/.

2.3. SUFI-2 algorithm

The program SUFI-2 was used for calibration and uncertainty analysis. In SUFI-2, parameter
uncertainty accounts for all sources of uncertainties such as uncertainty in driving variables (e.g., rainfall),
conceptual model, parameters, and measured data. The 95PPU is calculated at the 2.5% and 97.5% levels
of the cumulative distribution of an output variable obtained through Latin hypercube sampling.
A short step-by-step description of SUFI-2 algorithm [20] is as follows:
 (1) In the first step an objective function is defined.
 (2) The second step the physically meaningful absolute minimum and maximum ranges for the
parameters being optimized are identified. Because the absolute parameter ranges play a constraining
role, they should be as large as possible, yet physically meaningful.
 (3) This step involves an optional, yet highly recommended “absolute sensitivity analysis” for all
parameters in the early stages of calibration.
 (4) Initial uncertainty ranges are assigned to parameters for the first round of Latin hypercube
sampling. In general, the above ranges are smaller than the absolute ranges, and depend upon
experience. The sensitivity analysis in step 3 can provide a valuable guide for selecting appropriate
ranges.
 (5) Latin Hypercube sampling is carried out next; leading to n parameter combinations, where n is the
number of desired simulations.
 (6) As a first step in assessing the simulations, the objective function is calculated.
 (7) Then a Latin Hypercube sampling is carried out in the hypercube, the corresponding objective
functions are evaluated, and the sensitivity matrix J and the parameter covariance matrix C are
calculated according to:
g
J ij  i i  1...C2n , j  1...m (1)
ji
 F.F. Tang
F.F. et al.
Tang / Procedia
et al./ Environmental
Procedia Sciences
Environmental 13 8(2012)
Sciences (2011)1760 – 1770
1786–1796 1791 1765

where C 2n is the number of rows in the sensitivity matrix (equal to all possible combinations of two
simulations), and j is the number of columns (number of parameters).

C  s g2 ( J T J ) 1 (2)

where sg2 is the variance of the objective function values resulting from m model runs.
 (8)A 95% predictive interval of a parameter bj is computed as follows:

b j ,lower  b *j  tV , 0.025 s j (3)

b j ,lower  b *j  tV , 0.025 s j (4)

where b *j is the parameter bj for the best estimates, and v is the degrees of freedom (m-n).
 (9) The 95PPU is calculated. And then two indices, i.e., the p-factor (the percent of observations
bracketed by the 95PPU) and the r-factor, are calculated:
dX
r  factor  (5)
X
where  X is the standard deviation of the measured variable X. A value of less than 1 is a desirable

measure for the r-factor.

1

k
dX  ( X U  X L )l (6)
k l 1

where k is the number of observed data points; and XL (2.5th) and XU (97.5th) represent the lower and
upper boundary of the 95PPU.
 (10) Because parameter uncertainties are initially large, the value of d tends to be quite large during the
first sampling round. Hence, further sampling rounds are needed with updated parameter ranges
calculated from:
 b  b 'j ,min   b 'j ,max  b 'j ,upper  
b 'j ,min  b j ,lower  Max  j ,lower , 


 (7)
 2 2
  

 b  b 'j ,min   b 'j ,max  b 'j ,upper  

b 'j ,max  b j ,upper  Max  j ,lower , 


 (8)
 2 2
  
where b′ indicate updated values. Parameters of the best simulation are used to calculate bj,lower and
bj,upper. The above criteria, while producing narrower parameter ranges for subsequent iterations, ensure
that the updated parameter ranges are always centered on the best estimates.
In addition, SUFI-2 is linked to SWAT (in the SWATCUP software) through an interface that includes
the programs GLUE, ParaSol, MCMC algorithm. SWAT-CUP was developed by Swiss Federal Institute
of Aquatic and Technology, Naprash Company and Texas A&M University.
1766
1792 F.F.F.F.
TangTang et al.
et al./ / Procedia
Procedia Environmental
Environmental Sciences
Sciences 13 (2012)
8 (2011) 1760 – 1770
1786–1796

3. Results and analysis

3.1. Sensitivity analysis

Sensitivity analysis was performed to identify key parameters in the Chao River basin, as hydrology
component of SWAT involve a large number of parameters. For the sensitivity analysis, 20 parameters
integrally related to stream flow were initially selected. After one iteration run, 12 parameters such as
CN2, ALPHA_BF, ESCO, SOL_AWC, SOL_K, CANMX, EPCO, et al., were found more sensitive. The
ranking of 12 parameters are listed in Table 2. The sensitivity analysis results obtained from this study are
very similar to previous studies [21]. CN2 is the most sensitive as expected.

Table 2 Ranking of SWAT input parameters related to runoff

Ranking Name Definition

1 CN2.mgt SCS runoff curve number for moisture condition II
2 ALPHA_BF.gw Base flow alpha factor (days)
3 ESCO.hru Soil evaporation compensation factor
4 SOL_AWC.sol Soil available water storage capacity (mm H2O/mm soil)
5 SOL_K.sol Soil conductivity (mm hr-1)
6 CANMX.hru Maximum canopy storage (mm H2O)
7 EPCO.hru Plant uptake compensation factor
8 SOL_Z.sol Depth from soil surface to bottom of layer (mm)
9 CH_K2.rte Effective hydraulic conductivity in the main channel (mm hr-1)
10 CH_N2.rte Manning’s n value for main channel
11 GW_DELAY.gw Groundwater delay time (days)
12 GW_REVAP.gw Groundwater revap. coefficient

3.2. Model parameters calibration and uncertainty analysis

In SUFI-2, a p-factor of 1 and r-factor of zero is a simulation that exactly corresponds to measured
data. The degree to which are away from these values can be used to judge the strength of calibration. The
results of monthly runoff calibration and uncertainty for the Chao River basin were presented in Figure 2.
The p-factor was 0.85 and the r-factor was 1.12 in calibration period (1995-1999) (Figure 2a) while the p-
factor was 0.83 and the r-factor was 2.15 in validation period (2000-2002) (Figure 2b). The percentage of
data being bracketed by 95PPU (p-factor) was high both in calibration and validation periods, e.g. 0.85
and 0.83. Additionally, some observed data were not bracketed by the prediction band and occurred at the
beginning stages of calibration and validation periods. This maybe related to the fact that SWAT model in
the Chao River basin was not warmed up in this study. Warming up can define more real initial soil
moisture. If SWAT model was warmed up, the results would be better.
On the other hand, a careful examination of the calibration results showed that the observed data
unbracketed by the prediction band were fallen in the baseflow part. This was maybe caused by a
limitation of SWAT that it does not rigorously simulate groundwater flow [22]. Groundwater recharge is
important in the Chao River basin. If baseflow were better simulated, a larger p- factor as well as a
 F.F. Tang
F.F. et al.
Tang / Procedia
et al./ Environmental
Procedia Sciences
Environmental 13 8(2012)
Sciences (2011)1760 – 1770
1786–1796 1793 1767

smaller r-factor could be achieved for a better simulation result. Therefore, parameters such as
groundwater recharge and groundwater–river interaction were important in the hydrology processes.

120
95PPU
100
observed
80
Runoff（m 3 /s）

60

40

20

0
Jan-95 Jul-95 Jan-96 Jul-96 Jan-97 Jul-97 Jan-98 Jul-98 Jan-99 Jul-99
Month

a
50
95PPU
40 observed
Runoff（m 3 /s）

30

20

10

0
Jan-00 Jul-00 Jan-01 Jul-01 Jan-02 Jul-02
Month

b
Fig.2 Monthly runoff simulation at Xiahui station (a. calibration period; b. validation period)

Figure 2 showed that the r-factor was 1.12 in calibration period, while the r-factor was 2.15 in
validation period. Results showed that the SWAT model revealed large uncertainties for validation period
which were in wet years, although the simulation of monthly runoff for the Xiahui station was satisfactory
during the calibration period. This maybe resulted from the fact that SWAT can not well simulate extreme
event. Tolson & Shoemaker reported that SWAT is not designed to simulate an extreme event and the
model usually underpredicts the largest flow events [23].
When values of p-factor and r-factor are accepted, then the parameter uncertainties are the desired
parameter ranges. Further goodness of fit can be quantified by the coefficient of determination (R 2) and
Nash–Sutcliffe coefficient (NS) [24] between the observed and the final simulated data. For R2, the closer
to 1 R2 is, the better the simulation result is. For NS, the simulation results are good when NS is larger
1768
1794 F.F.F.F.
TangTang et al.
et al./ / Procedia
Procedia Environmental
Environmental Sciences
Sciences 13 (2012)
8 (2011) 1760 – 1770
1786–1796

than 0.75; the simulation results are satisfactory when NS is larger than 0.36 and smaller than 0.75;
simulation results are satisfactory when NS is larger than 0.36 and smaller than 0.75; simulation results
are not good when NS is smaller than 0.36. The results indicated that R2 was 0.90 and NS was 0.88 in
calibration period (Figure 3a), and R2 was 0.77 and NS was 0.74 in validation period (Figure 3b). The
simulation results were satisfactory. It should be noted that it is not necessary to seek the “best
simulation” as in such a stochastic procedure the “best solution” is actually the final parameter ranges.

Precipitation Observed data Simulated data

200 0

Precipitation（mm）
100
Runoff（m 3 /S）

150
200
100
300
50
400

0 500
Jan-95 Jul-95 Jan-96 Jul-96 Jan-97 Jul-97 Jan-98 Jul-98 Jan-99 Jul-99

Month

Precipitation Observed data Simulated data

50 0
50

Precipitaion（mm）
40
Runoff（m 3 /S）

100
30
150
20
200
10 250
0 300
Jan-00 Jul-00 Jan-01 Jul-01 Jan-02 Jul-02

Month

b
Fig.3 Simulated and observed hydrograph at Xiahui station in the Chao River basin (a. calibration period; b. validation period)

4. Conclusions

SWAT model was applied in the Chao River basin to simulate runoff from 1995 to 2002 in this study.
The year of 1995-1999 was selected as the calibration period and the year of 2000-2002 was selected as
the validation period. Then sensitivity analysis, model calibration and uncertainty analysis were
performed using SUFI-2 algorithm integrated with SWAT. The following conclusions were obtained.
 F.F. Tang
F.F. et al.
Tang / Procedia
et al./ Environmental
Procedia Sciences
Environmental 138(2012)
Sciences (2011)1760 – 1770
1786–1796 1795 1769

 (1) Results of sensitivity analysis showed that 12 parameters such as CN2, ALPHA_BF, ESCO,
SOL_AWC, SOL_K, CANMX, EPCO, et al., were found more sensitive.
 (2) Results of uncertainty analysis indicated that SWAT model had large uncertainties for validation
period, although the simulation of monthly runoff for the Xiahui station was satisfactory during the
calibration period.
 (3) Calibration and validation results showed that R2 was 0.90 and NS wad 0.88 in calibration period,
and R2 was 0.77 and NS was 0.74 in validation period. The simulation results were satisfactory.

References

[1] Young RA, Onstad CA, Bosch DD, Anderson WP. AGNPS – A nonpoint-source pollution model for evaluating agricultural
watersheds. J. Soil Wat. Conserv 1989;44:168-73.
[2] Arnold JG, Srinivasan R, Muttiah RS, Williams JR. Large area hydrologic modeling and assessment–Part 1: Model
development. J Am Water Resour As 1998;34:73-89.
[3] Bicknell BR, Imhoff J, Kittle J, Jobes T, Donigian AS. Hydrological Simulation Program–Fortran User’s Manual. USEPA
2000.
[4] Gassman PW, Reyes MR, Green CH, Arnold JG. The Soil and Water Assessment Tool: Historical Development,
Applications, and Future Research Directions. Transactions of the ASABE 2007;50:1211-50.
[5] Kuczera G, Parent E. Monte Carlo assessment of parameter uncertainty in conceptual catchment models: the Metropolis
algorithm. J. Hydrol 1998;211:69-85.
[6] Vrugt JA, Gupta HV, Bouten W, Sorooshian S. A Shuffled Complex Evolution Metropolis algorithm for optimization and
uncertainty assessment of hydrologic model parameters. Water Resour Res 2003;39:1201.
[7] Yang J, Reicher P, Abbaspour K C, Xia J, Yang H. Comparing uncertainty analysis techniques for a SWAT application to
the Chao he Basin in China. J. Hydrol 2008;358:1-23.
[8] Beven K., Binley A. The future of distributed models –model calibration and uncertainty prediction. Hydrol.Process
1992;6:279-98.
[9] van Griensven A, Meixner T. Methods to quantify and identify the sources of uncertainty for river basin water quality
models. Water Sci Technol 2006;53:51-9.
[10] Abbaspour KC, Yang J, Ivan M, Siber R, Bogner K, Mieleitner J, Zobrist J, Srinivasan R. Modelling hydrology and water
quality in the pre-alpine/alpine Thur watershed using SWAT. J. Hydrol 2007;333:413-30.
[11] Faramarzi M, Abbaspour KC, Schulin R, Yang H. Modelling blue and green water resources availability in Iran.
Hydrol.Process 2009;23:486-501.
[12] Schuola J, Karim CA, Srinivasan R, Yang H. Estimation of freshwater availability in the West African sub-continent using
the SWAT hydrologic model. J. Hydrol 2008;352:30-49.
[13] Arnold JG, Fohrer N. SWAT2000: current capabilities and research opportunities in applied watershed modelling. Hydrol.
Proces 2005;19:563-72.
[14] Behera S, Panda, RK. Evaluation of management alternatives for an agricultural watershed in a sub-humid subtropical
region using a physical process based model. Agric. Ecosyst. Environ 2006;113:62-72.
[15] Hargreaves GH, Samni ZA. Reference crop evapotranspiration from temperature. Appl Eng Agric 1985;1:96-9.
[16] Priestley CHB, Tylor RJ. On the assessment of surface heat flux and evaporation using large-scale parameters. Mon
Weather Rev 1972;100:81-92.
[17] Monteith JL. Evaporation and environment, the state and movement of water in living organisms. In: 19th symp Soc Excp
Biol., Academic press, New York; 1964.
[18] Allen RG. A Penman for all seasons. J Irrig Drain E-ASCE 1986;112:348-68.
[19] Arnold JG, Allen PM. Estimating hydrologic budgets for three Illinois watersheds. J. Hydro 1996;176:57-77.
1770
1796 F.F. F.F.
Tang Tang et al.
et al./ / Procedia
Procedia Environmental
Environmental Sciences
Sciences 13 (2012)
8 (2011) 1760 – 1770
1786–1796

[20] Abbaspour KC, Johnson C A and van Genuchten MTh. Estimating Uncertain Flow and Transport Parameters Using a
Sequential Uncertainty Fitting Procedure. Vadose Zone J 2004;3:1340-52.
[21] Lenhart T, Kckhardt K, Fohrer, N, Frede HG. Comparison of two different approaches of sensitivity analysis. Phys Chem
Earth 2002;27:645-54.
[22] Rostamian R, Jaleh A, Afyuni M, Mousavi S F, Heidarpour M, Jalalian A, Abbaspour KC. Application of a SWAT model
for estimating runoff and sediment in two mountainous basins in central Iran. Hydrolog Sci J 2008;53:977-88.
[23] Tolson BA, Shoemaker CA. Watershed modeling of the Cannonsville basin using SWAT2000: model development,
calibration and validation for the prediction of flow, sediment and phosphorus transport to the Cannonsville Reservoir. Technical
Report, School of Civil and Environmental Engineering, Cornell University, New York, USA; 2004.
[24] Nash JE, Sutcliffe JV. River flow forecasting through conceptual models, Part 1- a discussion of principles. J. Hydrol
1970;10:282-90.