Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
To cite this article: ROKHSARE ROSTAMIAN , AAZAM JALEH , MAJID AFYUNI , SEYED
FARHAD MOUSAVI , MANOUCHEHR HEIDARPOUR , AHMAD JALALIAN & KARIM C.
ABBASPOUR (2008) Application of a SWAT model for estimating runoff and sediment in two
mountainous basins in central Iran, Hydrological Sciences Journal, 53:5, 977-988, DOI: 10.1623/
hysj.53.5.977
Abstract The Soil and Water Assessment Tool (SWAT) was used to model runoff and sediment in the
Beheshtabad (3860 km2) and Vanak (3198 km2) watersheds in the northern Karun catchment in central Iran.
Model calibration and uncertainty analysis were performed with sequential uncertainty fitting (SUFI-2),
which is one of the programs interfaced with SWAT, in the package SWAT-CUP (SWAT Calibration
Uncertainty Programs). Two measures were used to assess the goodness of calibration and uncertainty
analysis: (a) the percentage of data bracketed by the 95% prediction uncertainty (95PPU) (P factor), and
(b) the ratio of average thickness of the 95PPU band to the standard deviation of the corresponding
measured variable (D factor). Ideally, the P factor should tend towards 1 with a D factor close to zero. These
measures together indicate the strength of the calibration-uncertainty analysis. Runoff and sediment data
from four hydrometric stations in each basin were used for calibration and validation. The P factor for
Beheshtabad stations ranged from 0.31 to 0.86, while those for Vanak stations were between 0.71 and 0.80.
The D factor for Beheshtabad ranged from 0.3 to 1.1, and for Vanak it was 0.771.16. These measures
indicate a fair model calibration and accounting of uncertainties. The predicted runoff values were quite
similar to those for discharge.
Key words SWAT-CUP; uncertainty analysis; SUFI-2; sediment; runoff; Iran
Application du modle SWAT pour estimer lcoulement et le transport solide dans deux
bassins versants montagneux du centre de lIran
Rsum Le modle Soil and Water Assessment Tool (SWAT) a t utilis pour modliser lcoulement et le
transport sdimentaire dans les bassins versants de Beheshtabad (3860 km2) et de Vanak (3198 km2) du
bassin de Karun dans le centre de lIran. Le calage du modle et une analyse de sensibilit ont t conduits
avec la procdure dajustement squentiel de lincertitude (SUFI-2), qui est lun des programmes interfacs
avec SWAT dans le package SWAT-CUP (SWAT Calibration Uncertainty Programs). Deux mesures ont t
utilises pour valuer la qualit du calage et de lanalyse de sensibilit: (a) le pourcentage de donnes
cartes par lincertitude de prvision 95% (95PPU) (facteur P), et (b) le rapport entre lpaisseur moyenne
de la bande 95PPU band et lcart type de la variable mesure correspondante (facteur D). Idalement, le
facteur P doit tendre vers 1 avec un facteur D proche de zro. Ces mesures indiquent ensemble la force de
lassociation calage-analyse dincertitude. Les donnes de dbit et de transport sdimentaire de quatre
stations hydromtriques dans chaque bassin versant ont t utilises pour le calage et la validation. Le
facteur P va de 0.31 0.86 pour les stations de Beheshtabad, et de 0.71 0.80 pour les stations de Vanak. Le
facteur D va de 0.3 1.1 pour Beheshtabad, et de 0.77 1.16 pour Vanak. Ces valeurs indiquent un calage et
une prise en compte des incertitudes, du et par le modle, corrects. Les simulations prsentent de meilleurs
rsultats pour lcoulement que pour le transport solide.
Mots clefs SWAT-CUP; analyse dincertitude; SUFI-2; sdiment; coulement; Iran
1 INTRODUCTION
Soil erosion is an important economic, social and environmental problem. According to past
studies, Asia suffers more than other continents from soil erosion, and Iran is one of the worst
affected countries in Asia (Dregne, 1992; FAO, 1994). Based on research by Ahmadi et al. (2003),
the mean annual erosion rate in Iran is estimated to be about 2500 t km-2, which is 4.3 times more
than the mean erosion rate in the world (Ahmadi et al., 2003). Also, available information shows
that 59% of 17 large basins studied in Iran have been severely degraded (Ahmadi et al., 2003). For
the purpose of planning and managing a watershed, information from hydrometric stations is
needed for model calibration. Other data needs for modelling include elevation, soil, land use and
climate. Hydrometric stations are quite limited in the northern Karun catchment in central Iran and
Open for discussion until 1 April 2009 Copyright 2008 IAHS Press
978 Rokhsare Rostamian et al.
many of them have sparse data. Therefore, management plans are difficult to develop due to the
lack of measured data. Hence, watershed modelling plays a crucial role in the proper planning and
development of local resources. In recent years, mathematical models of watershed hydrology and
transport processes have been employed to address a wide spectrum of environmental and water
resources problems.
The Soil and Water Assessment Tool (SWAT) (Arnold et al., 1998) was developed to predict
the effects of different management practices on water quality, sediment yield and pollution loading
in watersheds. Arnold et al. (2000) applied SWAT with the addition of a streamflow filter and reces-
sion methods for regional estimation of baseflow and groundwater recharge in the upper Mississippi
River basin. The results showed a general tendency for SWAT to underpredict spring peaks and to
overestimate autumn streamflow compared to measured monthly data during both calibration and
validation periods. Tolson & Shoemaker (2004) reported application of the SWAT2000 model to
Cannonsville Reservoir, a New York City water supply reservoir, and found it was a valuable tool
that could be used to help identify and quantitatively evaluate the long-term effects of various phos-
phorus management options for mitigating loading to the reservoir. Abbaspour et al. (2007) used
the SWAT model to simulate all related processes affecting water quantity, sediment and nutrient
loads in the Thur watershed in Switzerland. Their study indicated excellent results for discharge
and nitrate, and quite good results for sediment and total phosphorus.
Mountainous regions in Iran are important sources of surface water supply and groundwater
recharge. Our objectives in this investigation were to evaluate SWAT and its applicability in the
Iranian mountainous regions for prediction of streamflow and sediment yield. The two watersheds
chosen, Beheshtabad and Vanak (named after the important stations in these watersheds), are
located in the Karun basin, which is an important source of water for the central and western part
of the country.
As distributed hydrological modelling is subject to large uncertainties, the definition and
quantification of model uncertainty have become the subject of considerable research in recent
years. To fulfil this demand, researchers have developed various uncertainty analysis techniques
for watershed models. These include Bayesian inference methods, such as: the Markov chain
Monte Carlo (MCMC) method (Kuczera & Parent, 1998; Vrugt et al., 2003; Yang et al., 2007);
generalized likelihood uncertainty estimation (GLUE) (Beven & Binley, 1992); parameter solution
(ParaSol) (van Griensven & Meixner, 2006); and sequential uncertainty fitting (SUFI-2)
(Abbaspour, et al., 2007). As no single calibration program can meet the objectives of different
modelling needs, GLUE, ParaSol, SUFI-2, and MCMC were interfaced with SWAT into a single
package, referred to as SWAT-CUP (SWAT Calibration Uncertainty Programs) (Abbaspour,
2007). In the current study, we used SUFI-2 for a combined calibration and uncertainty analysis of
our SWAT models.
Beheshtabad
watershed
Vanak watershed
mean annual temperature is 11C and mean annual precipitation is 600 mm. About 60% of
precipitation in Vanak watershed falls as snow. Approximately 40% of the land within the basin is
used for agricultural activities. Approximately 50% of the watershed is covered by pasture and
10% by rocks. About 56% of agricultural activities are rainfed farming.
& Samani (1985). Water flow is routed through the channel network using the variable storage
routing method or the Muskingum river routing method. Sediment yield in SWAT is estimated
with the modified soil loss equation (MUSLE) developed by Williams & Berndt (1977).
Calibration program
model.in backup
swat2000/2005.exe
swat_extract.exe
SWAT output
model.out
the system. Omitting the identifiers <hydrogrp>, <soltext>, <landuse>, and <subbsn> allows
global assignment of parameters.
The program SWAT-CUP, coupling various programs to SWAT, has the general concept
shown in Fig. 2. The steps are: (a) calibration programs write model parameters in model.in;
(b) swat_edit.exe edits the SWAT text files, inserting the new parameter values; (c) the SWAT
simulator is run; and (d) the swat_extract.exe program extracts the desired variables from the
SWAT output files and writes them to model.out. The procedure continues as required by the
calibration program (see Abbaspour, 2007 for more detail).
The parameterization and the basic data sets required to develop the two projects for this study
in the ArcView interface (Di Luzio et al., 2002) were: topography, soil, land-use, and climatic
data. The data used in modelling are as follows:
(i) a digital elevation model (DEM) taken from the National Cartographic Centre of Iran (grid:
20 m 20 m);
(ii) a digital stream network at the 1:75 000 scale, produced by the National Cartographic Centre
of Iran;
(iii) soil and land-use maps, at a scale of 1:100 000, produced by the Natural Resources
Department of the Cartographic Centre of Iran; and
(iv) climate data records from 17 precipitation gauges and seven air temperature gauges over a
period of 19 years (19852004); data were obtained from the Iran Metrological Organization.
The Beheshtabad watershed was subdivided into 32 sub-basins and 451 HRUs. The soil map
includes 44 types of soil. Soil texture, available water content, hydraulic conductivity, bulk density
and organic carbon content information were available for different layers for each soil type.
Wheat and alfalfa were chosen as representative crops in the Beheshtabad watershed. The
simulation time period was 19922004, where the first four years were used as a warm-up. Data
from the Beheshtabad, Darkeshvarkesh, Kohesokhte and Polkharagy hydrometric stations in the
Beheshtabad watershed were used for calibration and validation. As the available stations did not
have data for every month of the simulation period, it was inevitable to include different time
periods for calibration at different stations. Three-quarters of the available data for the stations at
Beheshtabad watershed were used for calibration and the remainder were used for validation.
The Vanak watershed was subdivided into 37 sub-basins and 247 HRUs. About 41 soil types
were identified for this watershed and wheat was chosen as a representative crop. The simulation
time period was 19852004, where the first four years were also used as a warm-up. Data from the
Vanak, Kasegan, Solegan, and Aghbolag hydrometric stations were used for calibration and
validation. Two-thirds of the available data at each station were used for calibration and the
remainder were used for validation. In both watersheds, the Hargreaves-Samani method was used
to estimate evapotranspiration, and the Muskingum routing method was selected to route water
through the channel network.
The SWAT model was initially calibrated based on the monthly measured discharge data. The
objective function, G, was formulated using the Nash-Sutcliffe coefficient:
(Qm Qs )i2
G = NS = 1 i
(Qm,i Qm )
2
where subscripts m and s represent measured and simulated, respectively, and Qm is the average
of measured data. The model was then recalibrated for both discharge and sediment by including
sediment in the objective function as well:
G = w1 NS Q + w 2 NS S
where subscripts Q and S represent discharge and sediment load, respectively, and w1 and w2 are
weights calculated as 1 / n 2 , where n is the number of data and 2 is the variance of the
respective variables.
Sediment data were based on collected grab samples, which were used to measure suspended
solids. Because these grab samples were the only available data for model calibration, the
sediment calibration should be considered as preliminary in this study.
The results of monthly runoff calibration and validation for Vanak watershed are presented in
Figs 5 and 6, respectively. These results are generally better than Beheshtabad in terms of the
percentage of data being bracketed (P factor), but the uncertainties are larger as expressed by the D
factor. This has resulted in large uncertainty in discharge peaks for all stations. For the validation
period, P factor values for stations in the Vanak watershed ranged from 0.77 to 0.89. The range of
D factor values for these stations is 0.862.04. Although the simulation of monthly runoff for the
Kasegan station was satisfactory during the calibration period, the SWAT model exhibited large
uncertainties for the validation period. The results of the timing for peak flow and the measured
peaks in Table 2 are similar to those of Beheshtabad watershed.
In both watersheds, the model mostly shows large uncertainties at extreme events during the
calibration and validation periods. Both watersheds are mountainous and snowfall is very
important (about 55% of precipitation in Beheshtabad and 60% of precipitation in Vanak falls as
snow). The SWAT model classifies precipitation as rain or snow using the average daily
temperature. Due to an apparent weakness of the model to simulate discharge in MarchMay, it
60
40
40
20
20
0 0
1998/10 1999/08 2000/06 2001/04 2002/02 2002/12 1996/01 1997/01 1998/01 1999/01 2000/01 2001/01 2002/01
Polkhargy - Calibration
Kohesokhte - Calibration 30
60
P-factor = 0.86 P-factor = 0.55 95PPU Observed
95PPU Observed
D-factor = 1.1 D-factor = 0.80
Discharge (m s )
Discharge (m3 s-1)
-1
40 20
3
20 10
0 0
1997/09 1998/07 1999/05 2000/03 2001/01 2001/11 2002/09 1998/10 1999/06 2000/02 2000/10 2001/06 2002/02 2002/10
Month Month
Discharge (m s )
-1
3
40 20
20 10
0 0
2003/04 2003/09 2004/02 2004/07 2002/07 2003/01 2003/07 2004/01 2004/07
Kohesokhte - Validation
30 Polkharagy - Validation
P-factor = 0.80 95 PPU Observed 20
P-factor = 0.33
D-factor = 1.33 95 PPU Observed
D-factor = 1.52
Discharge (m s )
Discharge (m3 s-1)
-1
20
3
10
10
0 0
2002/12 2003/06 2003/12 2004/06 2003/03 2003/07 2003/11 2004/03 2004/07
Month Month
Fig. 4 Monthly runoff validationBeheshtabad watershed.
Discharge (m 3 s -1)
80
40
40
20
0 0
1989/9 1991/5 1993/1 1994/9 1996/5 1998/1 1989/9 1991/5 1993/1 1994/9 1996/5
Aghbolag - Calibration
60 Vanak - Calibration
P-factor = 0.78 95PPU Observed P-factor = 0.80 95PPU Observed
D-factor = 1.14 160
D-factor = 0.77
Discharge (m3 s-1)
40
120
Discharge (m s )
-1
3
80
20
40
0 0
1996/9 1997/7 1998/5 1999/3 2000/1
1995/9 1996/12 1998/3 1999/6 2000/9
Month
Month
Fig. 5 Monthly runoff calibrationVanak watershed.
40 80
20 40
0 0
1998/10 2000/1 2001/4 2002/7 2003/10 1996/10 1997/9 1998/8 1999/7 2000/6 2001/5
Vanak - Validation
Aghbolag - Validation
30 P-factor = 0.89 95PPU Observed
P-factor = 0.81 95PPU observed D-factor = 0.86
D-factor = 0.95 80
-1
Discharge (m s )
Discharge (m3 s-1)
20
40
10
0 0
2000/10 2001/6 2002/2 2002/10 2003/6 2000/10 2001/6 2002/2 2002/10 2003/6
Month Month
Fig. 6 Monthly runoff validationVanak watershed.
seems that we were not able to adequately calibrate SWAT to simulate snowmelt in these moun-
tainous watersheds.
Similar conclusions were reached by Fontaine et al. (2002), who modified the snowmelt
component of the SWAT model to improve its capability to simulate hydrology of a non-
agricultural mountainous region (unfortunately we did not have access to this modification). Their
results indicated that the modified routines improved the models annual streamflow prediction
with an R2 value of 0.86 compared to an initial value of 0.7. Chu & Shirmohammadi (2004)
applied the SWAT model to predict surface and subsurface flow for a 340-ha watershed in the
Piedmont physiographic region of Maryland. Preliminary simulations showed that SWAT
underestimates subsurface flow and total streamflow, especially during wet periods. They reported
that SWAT is unable to simulate the extremely wet hydrological conditions, even after
adjustments to measured data. Tolson & Shoemaker (2004) reported that SWAT is not designed to
simulate such an extreme event and the model usually underpredicts the largest flow events.
The results of daily sediment calibration for Beheshtabad watershed are shown in Fig. 7. The
P factor values range from 0.32 to 0.76, while D factors range from 0.19 to 1.13. In general,
similar to discharge, it is the smaller sediment values that are mostly not bracketed by the pre-
diction band. This is probably due to the weakness of the model to simulate snowmelt. Other
reasons could be poor accuracy of the measured data and the dispersed nature of the data. The
largest error in prediction of sediment, however, was associated with large peak flows. Tolson &
Shoemaker (2004) calibrated and validated SWAT model for prediction of dissolved and
particulate phosphorus transport, and also flow and sediment transport, against a large set of
monitoring data. As in our case, they reported that the largest error in model predictions for
sediment loading was always associated with peak flow prediction errors. As Abbaspour et al.
(2007) discuss, the second storm effect may also be partly responsible for poor sediment results.
Despite these shortcomings we find the results quite useful, given that grab samples were used to
represent the measured data. Validation results for sediment are shown in Fig. 8, where large
uncertainties can be seen. The sediment results for the Vanak watershed were generally worse than
those for Beheshtabad and are not shown.
4 CONCLUSION
The SWAT model was applied to two mountainous watersheds in central Iran to predict runoff and
sediment. As expected, predicted runoff values were much better than those for sediment. The
weakness of the model to simulate runoff for some months was probably due to poor characteriza-
tion of snowmelt processes in these mountainous watersheds, lack of sufficient discharge data, and
40000
40000
20000
20000
0 0
3/2/1996 1/2/1998 11/2/1999 9/2/2001 1/2/1996 9/2/1997 5/2/1999 1/2/2001 9/2/2002
6000
8000
3000
4000
0 0
1/2/1996 9/2/1997 5/2/1999 1/2/2001 10/5/1997 6/5/1999 2/5/2001 10/5/2002
Date Date
Fig. 7 Daily sediment calibrationBeheshtabad.
Darkeshvarkesh - Validation
Beheshtabad- Validation
20000 P-factor = 0.43
P-factor = 0.69 95 ppu observed
95 PPU Observed 12000 D-factor = 0.23
D-factor = 0.29
Daily Sediment (tn)
15000
Daily Sediment (tn)
8000
10000
4000
5000
0 0
11/9/2002 6/9/2003 1/9/2004 8/9/2004 11/7/2002 5/7/2003 11/7/2003 5/7/2004
Kohesokhte - Validation Polkharagy - Validation
4000 1500
P-factor = 0.83 P-factor = 0.57
95PPU Observed 95 PPU Observed
D-factor = 0.43 D-factor = 0.42
3000
Daily Sedim ent (tn)
1000
2000
500
1000
0 0
7/15/2002 3/15/2003 11/15/2003 7/15/2004 6/3/2003 11/3/2003 4/3/2004 9/3/2004
Date Date
Fig. 8 Daily sediment validationBeheshtabad.
lack of input data for simulation of groundwater recharge and groundwaterriver interaction. The
weakness of the model to simulate sediment was due to the improper peak runoff simulation and the
nature and accuracy of the measured sediment data. Prediction of runoff and soil loss is important for
assessing soil erosion hazards, and for determining suitable land uses and soil conservation measures
for a catchment. In turn, this can help in deriving optimum benefit from the use of the land whilst
minimising the negative impacts of land degradation and other environmental problems. As there are
limited data available from the region of study, the model developed herein could help assess
different land management options and in studying the effect of climate change on soil erosion.
Acknowledgements This study was supported partly by the Swiss Federal Institute of Aquatic
Science and Technology (Eawag). The authors would like to express their appreciation for this
cooperation.
REFERENCES
Abbaspour, K. C. (2007) User Manual for SWAT-CUP, SWAT Calibration and Uncertainty Analysis Programs. Swiss Federal
Institute of Aquatic Science and Technology, Eawag, Dbendorf, Switzerland.
http://www.eawag.ch/organisation/abteilungen/siam/software/swat/index_EN [Last accessed August 2008].
Abbaspour, K. C., Yang, J., Maximov, I., Siber, R., Bogner, K., Mieleitner, J., Zobrist, J., Srinivasan, R. & Reichert, P. (2007)
Modelling of hydrology and water quality in the pre-alpine/alpine Thur watershed using SWAT. J. Hydrol. 333, 413430.
Ahmadi Ilkhchi, A., Hajabbassi, M. A. & Jalalian, A. (2003) Effects of converting range to dry-farming land on runoff and soil
loss and quality in Dorahan, Chaharmahal & Bakhtiari Province. J. Sci. & Technol. Agric. & Natural Resour. 6(4),
103114.
Arnold, J. G. & Allen, P. M. (1996) Estimating hydrologic budgets for three Illinois watersheds. J. Hydrol. 176, 5777.
Arnold, J. G., Srinisvan, R., Muttiah, R. S. & Williams, J. R. (1998) Large area hydrologic modeling and assessment. Part I:
Model development. J. Am. Water Resour. Assoc. 34(1), 7389.
Arnold, J. G., Muttiah, R. S., Srinivasan, R. & Allen, P. M. (2000) Regional estimation of base flow and groundwater recharge
in the upper Mississippi River basin. J. Hydrol. 227(1), 2140.
Beven, K. & Binley, A. (1992) The future of distributed modelsmodel calibration and uncertainty prediction. Hydrol.
Processes 6(3), 279298.
Chow, V. T., Maidment, D. R. & Mays, L. W. (eds) (1988) Applied Hydrology. McGraw-Hill Inc., New York, USA.
Chu, T. W. & Shirmohammadi, A. (2004) Evaluation of the SWAT models hydrology component in the piedmont
physiographic region of Maryland. Trans. Am. Soc. Agric. Engrs 47(4), 10571073.
Di Luzio, M., Srinisvasan, R., Arnold, J. G. & Neitsch, S. L. (2002) ArcView Interface for SWAT2000. Blackland Research &
Extension Center, Texas Agricultural Experiment Station and Grassland, Soil and Water Research Laboratory, USDA
Agricultural Research Service.
Dregne, H. E. (1992) Erosion and soil productivity in Asia. J. Soil Water Conserv. 47, 813.
FAO, UNDP and UNEP (1994) Land degradation in south Asia: its severity, causes and effects upon the people. World Soil
Resources Report no. 78, FAO, Rome, Italy.
Fontaine, T. A., Cruickshank, T. S, Arnold, J. G. & Hotchkiss, R. H. (2002) Development of a snowfallsnowmelt routine for
mountainous terrain for the soil water assessment tool (SWAT). J. Hydrol. 262(14), 209223.
Hargreaves, G. & Samani, Z. A. (1985) Reference crop evapotranspiration from temperature. Applied Engng Agric. 1, 9699.
Kuczera, G. & Parent, E. (1998) Monte Carlo assessment of parameter uncertainty in conceptual catchment models: the
Metropolis algorithm. J. Hydrol. 211(14), 6985.
Monteith, J. L. (1965) Evaporation and environment. In: The State and Movement of Water in Living Organisms (ed. by
G. F. Fogg), 205234. Cambridge University Press, Cambridge, UK.
Nash, J. E & Sutcliffe J. V. (1970) River flow forecasting through conceptual models. Part I. A discussion of principles.
J. Hydrol. 10(3), 282290.
Neudecker, H. & Magnus, J. R. (1988) Matrix Differential Calculus with Applications in Statistics and Econometrics, 136. John
Wiley & Sons, New York, USA.
Priestley, C. H. B. & Taylor, R. J. (1972) On the assessment of surface heat flux and evaporation using large-scale parameters,
Monthly Weather Rev. 100, 8192.
Tolson, B. A. & Shoemaker, C. A. (2004) 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, Ithaca, New York, USA.
van Griensven, A. & Meixner, T. (2006) Methods to quantify and identify the sources of uncertainty for river basin water
quality models. Water Sci. Technol. 53(1), 5159.
Vrugt, J. A., Gupta, H. V., Bouten, W. & Sorooshian, S. (2003) A shuffled complex evolution metropolis algorithm for
optimization and uncertainty assessment of hydrologic model parameters. Water Resour. Res. 39(8), art. no. 1201,
doi:10.1029/2002WR001642.
Williams, J. R. & Berndt, H. D. (1977) Sediment yield prediction based on watershed hydrology. Trans. Am. Soc. Agric. Engrs
20(6), 11001104.
Yang, J., Abbaspour, K. C., Reichert, P., Yang, H. & Xia, J. (2008) Comparing uncertainty analysis techniques for a SWAT
application to the Chaohe basin in China. J. Hydrol. 358, 123.