Isfahan University of Technology, College of Agriculture, Department of Water Engineering, Isfahan 84156-83111, IranIsfahan University of Technology, College of Agriculture, Department of Soil Science, Isfahan 84156-83111, IranEawag, Swiss Federal Institute of Aquatic Science and Technology, Ueberlandstr 133, 8600 Dübendorf, Switzerland
r t i c l e i n f o
rticle history:eceived 26 February 2010eceived in revised form 24 October 2010ccepted 26 October 2010vailable online 20 November 2010
Application of large amounts of mineral and organic fertilizers in intensive agricultural regions ofHamadan–Bahar watershed in western Iran contributes to excessive nutrient loads in soils and groundwa-ter bodies. Groundwater supplies approximately 88% of the water consumed in Hamadan. The objectiveof this study was to investigate the temporal and spatial variability of nitrate leaching in Hamadan–Baharwatershed. We employed the Soil and Water Assessment Tool (SWAT) to model the amount and dynam-ics of nitrate leaching from a typical crop rotation in this watershed. The SWAT model was calibratedand validated with uncertainty analysis using SUFI-2 (Sequential Uncertainty Fitting, ver. 2) based onmeasured daily discharge data from 7 hydrometric stations, wheat and potato yield, and measured dailynitrate at the outlet of the watershed. The calibration using crop yield increases the confidence on soilmoisture and evapotranspiration. The calibration (R2 = 0.83, NS = 0.77) and validation (R2 = 0.70, NS = 0.70)results were quite satisfactory for the outlet of watershed. Spatial variations in nitrate leaching were also
found to agree reasonably well with measured nitrate concentrations in groundwater (73% overlap basedon a defined criterion). Also, nitrate leaching was found to be more significant under potato (Solanumtuberosum L.) rotation (254–361 kg N ha−1 year−1), representing 30–42% of nitrogen applied to the soil.About 36% of Hamadan–Bahar aquifer has a nitrate leaching rate higher than 100 kg N ha−1 year−1. Thepresented model and its results have the potential to provide a strong base for considering differentscenarios to reduce nitrate leaching and suggest a BMP (best management practice) in Hamadan–Bahar watershed.
Abbreviations: 95PPU, 95% prediction uncertainty; ˚, efficiency criterion basedn modified R2; �2, variance; b, slope of the regression line; CN, curve number;, objective function; HI, Harvest Index; HRU, hydrologic response unit; HRWA,amadan Regional Water Authority; LAI, leaf area index; MCMC, Monte Carloarkov Chain; MOJAH, Ministry of Jahade-Agriculture of Hamadan; MSE, mean
quare error (range of values −∞ to +∞); n, number of observations; NLoad , monthlyitrate loads carried by surface runoff (kg N month−1); NO−
3 , nitrate concentrationmg l−1); NS, Nasch–Sutcliffe factor (range of value ≤1); P-factor, %data brack-ted by the 95PPU (range of value 0–1); p-value, p-value shows the significancef the sensitivity (range of value ≥0); ParaSol, Parameter Solution; PET, poten-ial evapotranspiration; q, discharge in the river reach (m3 s−1); R2, coefficient ofetermination; RMSE, root mean square error; R-factor, the average thickness ofhe 95PPU band divided by the standard deviation of the measured data (rangef value 0–1); SDNCO, denitrification threshold water content; SUFI-2, Sequentialncertainty Fitting, ver. 2; SWAT, Soil and Water Assessment Tool; SWAT-CUP,WAT Calibration Uncertainty Procedures; Yobs , observed crop yield (ton ha−1); Ysim ,imulated crop yield (ton ha).∗ Corresponding author. Tel.: +41 44 823 5359.
In Hamadan–Bahar watershed, water scarcity has become anincreasingly serious problem. Groundwater is the major source ofwater supply for drinking, domestic, industrial, and agriculturalsectors in this region. One of the problems affecting the qualityof groundwater is leaching nutrients from the soil, which is espe-cially evident in agricultural dominated watersheds (Jalali, 2005).Application of large amounts of nitrogen fertilizers, at higher thancrop uptake requirement rate, in intensive agricultural regions ofHamadan–Bahar plain contributes to excessive nitrate accumula-tion in soils and leaching into groundwater bodies (Jalali, 2005;Nadafian, 2007; Rahmani, 2003).
Nitrogen leaching from agricultural lands is a widespread global
problem. In areas where nonpoint source pollution is dominant,regional models are often the only practical way to examine theimpacts of changing landuse on the concentration of nitrate pol-lution. Hydrological models that are able to calculate the nitrogenfate and transport are useful tools to determine the probable effects
676 S. Akhavan et al. / Agriculture, Ecosystems and Environment 139 (2010) 675–688
F this fis
oioeqscTim
g(rfi
is
ig. 1. Study area of Hamadan–Bahar watershed in Hamadan province, Iran. Inub-basins are shown.
f agricultural activities on local hydrology and aqueous geochem-stry. Direct measurement of the impact of agricultural practicesn groundwater quality is tedious and usually too expensive. Forxample, soil sampling is time consuming and sampling wateruality at field level, using suction cups or lysimeters is expen-ive and impractical (Lord and Shepherd, 1993). Such approachesannot directly help in making a general decision in large scales.hese issues demand integrated management of water resourcesn watersheds (Pohlert et al., 2007), which rely mostly now on
odeling.Nonpoint source nitrate loading has substantially impacted
roundwater nitrate concentrations in Hamadan–Bahar aquiferJalali, 2005; Nadafian, 2007; Rahmani, 2003). These wateresources are located in the vicinity of drinking water wells. There-
ore, it is essential to determine how management practices willmpact groundwater nitrate concentrations.
With this background in mind, the objectives of this studyncluded: (1) using the SWAT program to model the temporal andpatial variability of nitrate leaching dynamics for the present agri-
gure, river network, main aquifer, meteorological stations and SWAT-delineated
cultural activities at hydrologic response unit (HRU) level withmonthly time steps, (2) to calibrate, validate and perform uncer-tainty analysis for the SWAT hydrologic model for Hamadan–Baharwatershed based on river discharge and nitrate data, and then (3) tocalibrate, validate and perform uncertainty analysis for the SWATcrop yield model using the main regional crops, which are potato(Solanum tuberosum L.) and irrigated and rainfed wheat (Triticumasestivum L.).
2. Materials and methods
2.1. Description of the study area
The Hamadan–Bahar watershed with an area of 2460 km2 is sit-
uated between longitudes of 48◦7′E and 48◦52′E and latitudes of34◦35′N and 35◦12′N, in western Iran (Fig. 1). In this watershed,most of the rivers originate from southern heights (Alvand Moun-tains). All of the rivers merge in the central plain and form theSiminehrud (Fig. 1). The outlet of watershed is Koshkabad in the
Crop yield 2 cities (annual yield) Organization of Jahade-Agricultureof Hamadan
Historical annual yield from 1992to 2008b
Nitrate 1 station (daily) Sampled by authors During fall 2007 to spring 2008b
ns2zia
TM
TM
Nitrate well 30 wells (monthly)
a These are input data for SWAT model.b These data are used for calibration and validation of SWAT model.
orth-east. Mean elevation of the watershed is 2038 m above meanea level. The average daily discharge at Koshkabad station was
.5 m3 s−1 for the period of 1992–2008, with a minimum value ofero and a maximum value of 90.4 m3 s−1. The climate of the regions semiarid with mean annual precipitation of 324.5 mm and meannnual temperature of 11.3 ◦C.
able 3anagement practices (tillage, planting, and fertilizer application) for rainfed wheat in H
Year Operation type Date
1 Tillage 4 June1 Tillage 10 June1 Planting 11 October1 Fertilizer 11 October2 Fertilizer 20 April (122 Harvest 5 July (21 Ju
a The dates in parenthesis are for Bahar region.
able 4anagement practices (tillage, planting, and fertilizer application) for irrigated wheat in
Sampled by authors From October 2007 to September2008b
In the Hamadan–Bahar watershed, groundwater is the onlyavailable and widely used source of drinking water for rural and
urban communities and also for irrigation. Groundwater suppliesapproximately 88% of the water consumed in Hamadan. Geo-logically, Hamadan–Bahar aquifer is located on Sanandaj-Sirjanmetamorphic zone (Hamadan Regional Water Authority, HRWA).
amadan.
Data description/properties
Mould board ploughLeveller
(22 September)a –(30 September) 18-46-00, 100 kg ha−1
April) Urea, 100 kg ha−1
ne) –
Hamadan.
Data description/properties
r Mould board ploughr Leveller(3 October)a –(3 October) 18-46-00, 150 kg ha−1
Water stress factor was set to 0.99Urea, 250 kg ha−1
ly) –
678 S. Akhavan et al. / Agriculture, Ecosystems and Environment 139 (2010) 675–688
Table 5Management practices (tillage, planting, and fertilizer application) for potato in Hamadan.
Year Operation type Date Data description/properties
1 Tillage 22 September (10 October)a Mould board plough1 Fertilizer 30 September (12 October) Hen manureb
2 Planting 17 March (6 April) –2 Fertilizer 17 March (6 April) 18-46-00, 250 kg ha−1
2 Autoirrigation 8 April Water stress factor was set to 0.992 Fertilizer 9 May (15 May) Ureab
gust (
TgJbmmH
TD
2 Harvest 21 Au
a The dates in parenthesis are for Bahar region.b Different amounts of fertilizer was applied during simulation period.
he parent rocks are mainly limestone, calcareous shale andranitic material. The oldest deposits contain slate and schist from
urassic age that outcrop in the eastern and southern parts of theasin. The Cretaceous deposits consist of the carbonate series. Theain portion of the study area is covered by Quaternary sedi-ents and consists mainly of recent alluvium and conglomerate.amadan–Bahar alluvial aquifers consist mainly of gravel, sand,
able 6escription of SWAT input parameters included in the calibration process and their sensi
Variable Parametera Definitio
Parameter sensitive to discharge r CN2.mgt SCS runoconditio
Parameter sensitive to crop yield v HI.mgt Harvestv HEAT UNITS.mgt Total he
Parameter sensitive to nitrate v CDN.bsn Denitrifiv SDNCO.bsn Denitrifiv RCN.bsn Concentv NPERCO.bsn Nitrogenv FRT SURFACE.mgt Fraction
soilv SHALLST N.gw Initial N
(mg N L−
v N UPDIS.bsn Nitrogenv SOL NO3.chm Initial N
(mg kg−
v ERORGN.hru Organicv SOL ORGN.chm Initial or
(mg kg−
a v: parameter value is substituted by a value from the given range; r: parameter valueb t-Value shows a measure of sensitivity: the larger t-value are more sensitive.c p-Value shows the significance of the sensitivity: the smaller the p-value, the less cha
15 September) –
silt and clay. The alluvial sediment thickness varies from 25 m inthe borders to 75 m in the centre of the plain. Transmissivity of
2 −1
the Hamadan–Bahar watershed ranges from 100 to 1750 m day ,while specific yield of the aquifer is about 5–10%.
The dominant soil texture in Hamadan–Bahar watershed (48%of the plains) is clay loam. They are mainly deep and moderatelywell-drained. Also, about 15.3% of Hamadan–Bahar plain’s soil is
tivity statistics.
n t-Valueb p-Valuec
ff curve number for moisturen II
21.53 0
l temperature (◦C) 13.77 0lt base temperature (◦C) 9.34 0slope length (m) 6.53 0ck temperature lag factor 4.36 0m melt rate for snow during the year1 day−1)
3.73 0
alpha factor for bank storage (days) 2.74 0.01ductivity (mm h−1) 2.28 0.02
hydraulic conductivity in the main(mm h−1)
2.19 0.03
n parameter 2.06 0.04m melt rate for snow during the year1 day−1)
1.99 0.05
oration compensation factor 1.36 0.17ld depth of water in the shallowor ‘revap’ to occur (mm)
1.29 0.20
take compensation factor 1.25 0.21g’s n value for overland flow 0.99 0.32g’s n value for main channel 0.96 0.34uifer percolation fraction 0.50 0.62density (g cm−3) 0.42 0.67
lable water storage capacityO/mm soil)
0.40 0.69
ater revap. coefficient 0.29 0.77il albedo 0.26 0.79ld depth of water in the shallowequired for return flow to occur (mm)
0.15 0.88
alpha factor (days) 0.13 0.89ater delay time (days) 0.01 0.99
Index – –at unit for crop to reach maturity – –cation exponential rate coefficient 10.36 0cation threshold water content 4.48 0ration of nitrogen in rainfall (mg N L−1) 2.24 0.02
percolation coefficient 2.23 0.02of fertilizer applied to top 10 mm of 1.32 0.19
O3concentration in shallow aquifer1)
1.18 0.24
uptake distribution parameter 1.04 0.3O3 concentration in the soil layer1)
0.55 0.58
N enrichment ratio 0.18 0.85ganic N concentration in the soil layer
1)0.12 0.90
is multiplied by (1 + a given value) (Abbaspour et al., 2007).
nce of a parameter being by chance assigned as sensitive.
a Numbers in parentheses are validation result.b ˚ is objective function.
oam and sandy loam. They are deep and well-drained soils too.hese soils are mostly under potato crop. Generally, the electricalonductivity and pH of Hamadan–Bahar watershed soils vary fromto 2 dS m−1 and from 7.5 to 8, respectively (Information Center ofinistry of Jahade-Agriculture of Hamadan, MOJAH).Landuse in Hamadan–Bahar plain is predominantly agricultural,
ith major crops being wheat and potato. Landuse in the areaverlying the unconfined Hamadan–Bahar aquifer has historicallyncluded wheat and potato. About 52.5% of the main aquifer is usedor irrigation. Potato cultivation in this region has a relatively highrrigation and nitrogen fertilizer demand. Table 1 summarizes the
nnual amount of N-fertilizers applied in Hamadan and Bahar agri-ultural lands for potato according to the data from MOJAH. Forainfed and irrigated wheat, an average of 100 and 250 kg ha−1 ureas applied in this region, respectively.
able 9MSE (ton ha−1) and R-factor for calibration (validation) results of simulated crop yield in
The SWAT model was chosen because of its ability to predict theimpact of land management practices on water and soil in large andcomplex watersheds, over long periods of time (Arnold et al., 1998;Neitsch et al., 2005). On the other hand, SWAT model can simu-late water quantity, water quality and crop growth simultaneously.However, limited attempts have been made to simulate nitrateleaching at watershed scale resulting from agricultural activitieswith the SWAT model (see e.g. Bouraoui and Grizzetti, 2008). Afew recent applications (e.g. Abbaspour et al., 2007; Grizzetti et al.,
2003; Saleh and Du, 2004; Santhi et al., 2001; Volk et al., 2009) haveshown the ability of SWAT to simulate runoff and nutrient at thesubbasin and watershed scales.
680 S. Akhavan et al. / Agriculture, Ecosystems and Environment 139 (2010) 675–688
Koshkabad
0
10
20
30
40
50
60
70
80
Sep
-00
Sep
-01
Sep
-02
Sep
-03
Sep
-04
Sep
-05
Sep
-06
Sep
-07
Sep
-08
Riv
er d
isch
arg
e (m
s3-1
)R
iver
dis
char
ge
(m s3
-1)
Riv
er d
isch
arg
e (m
s3-1
)
Riv
er d
isch
arg
e (m
s3-1
)R
iver
dis
char
ge
(m s3
-1)
Riv
er d
isch
arg
e (m
s3-1
)
Koshkabad
0
10
20
30
40
50
60
70
80
Jan-9
2
Jan-9
3
Jan-9
4
Jan-9
5
Jan-9
6
Jan-9
7
Jan-9
8
Jan-9
9
Yalfan
0123456789
10
Jan-00
Jan-01
Jan-02
Jan-03
Jan-04
Jan-05
Jan-06
Jan-07
Jan-08
Yalfan
0123456789
10
Jan-92
Jan-93
Jan-94
Jan-95
Jan-96
Jan-97
Jan-98
Jan-99
Abbasabad
0
1
2
3
4
5
6
Jan-0
0
Jan-0
1
Jan-0
2
Jan-0
3
Jan-0
4
Jan-0
5
Jan-0
6
Jan-0
7
Jan-0
8
Abbasabad
0
1
2
3
4
5
6
Jan-92
Jan-93
Jan-94
Jan-95
Jan-96
Jan-97
Jan-98
Jan-99
c d
e f
a b
F resulta rtaintr
tia2saetm
wgsa
saiwftf
method, which only requires minimum and maximum tempera-ture. The daily value of leaf area index (LAI) was applied to partitionthe PET into potential soil evaporation and potential plant transpi-ration. LAI and root development were estimated using the crop
Table 10Nitrate leaching classes and areas of Hamadan–Bahar watershed (aquifer) in eachclass of nitrate leaching.
Nitrate leachingrange(kg N ha−1 year−1)
Area of watershed(aquifer) (km2)
Area of watershed(aquifer) (%)
0–25 2067 (249)a 86.7 (51)25–50 61 (28) 2.6 (6)
ig. 2. Calibration (left graph, 2000–2008) and validation (right graph, 1992–1999)nd best simulation, respectively, and gray band expresses the 95% prediction unceeader is referred to the web version of the article.)
SWAT is a basin-scale, distributed, and continuous-time modelhat operates on a daily time step. It was designed to simulate thempact of various management practices on water, sediment, andgricultural chemical yields in ungauged watersheds (Neitsch et al.,005). Main model components consist of weather, hydrology,oil temperature, plant growth, nutrients, pesticides, land man-gement, bacteria and pathogens (Arnold et al., 1998; Gassmant al., 2007; Neitsch et al., 2005). In this study, we focus only onhe hydrologic, plant growth, nutrients (nitrate), and agricultural
anagement components of the SWAT model.In SWAT, a watershed is divided into multiple sub-watersheds,
hich are then further subdivided into HRUs that include homo-eneous slope, landuse, and soil characteristics. Calculated flow,ediment yield, and nutrient loading obtained for each sub-basinre then routed through the river system.
The water in each HRU is stored in four storage volumes: snow,oil profile (0–2 m), shallow aquifer (typically 2–20 m), and deepquifer. Surface runoff from daily rainfall is calculated using a mod-
fied SCS curve number method. Downward flow happens when
ater content exceeds field capacity for each layer. Percolationrom the bottom of soil profile recharges the shallow aquifer. If theemperature in a particular layer is ≤0 ◦C, no percolation is allowedrom that layer. Groundwater flow contribution to total stream flow
s for discharge at three hydrometric stations. The red and blue lines are observationy (95PPU). (For interpretation of the references to colour in this figure legend, the
is estimated by routing a shallow aquifer storage component to thestream (Arnold and Allen, 1996).
Depending on data availability, potential evapotranspiration(PET) can be calculated using different approaches. In this study,potential evapotranspiration was calculated using the Hargreaves
ig. 3. Management map of the Hamadan–Bahar watershed. The map shows loca-ion of groundwater artificial recharges, check dams and agricultural activities.
rowth component of SWAT. This element indicates the relation-hip between vegetation and hydrologic budget (Faramarzi et al.,009, 2010; Luo et al., 2008). The crop growth model in SWAT is aimplification of the EPIC crop model (Williams et al., 1984). SWATimulates crop yield as a product of Harvest Index (HI) and above-round biomass. For the majority of crops, the Harvest Index variesetween zero and 1.0. But, plants whose roots are harvested, such asweet potatoes, may have a Harvest Index greater than 1.0 (Neitscht al., 2005).
The irrigation applications can be simulated for specific datesr with an auto-irrigation routine, which triggers irrigation eventsccording to a water stress threshold. We selected automatic irri-ation, because it is difficult to know when and how much thearmers apply irrigation during simulation periods. In this optionWAT irrigates to field capacity and thus does not generate anyercolation. But in reality, farmers irrigate more than field capacitynd this is how percolation water carrying nitrate contaminateshe groundwater. To accommodate this, we changed the sourceode and introduced additional parameters so that percolation canccur. A second problem with the code was that, with auto irri-ation option (based on the soil water content), the code irrigatedven after the harvest. We modified this part of the code also torevent post-harvest irrigation.
Nitrogen and phosphorus applications can be estimated in theorm of inorganic fertilizer and/or manure inputs. The transfor-
ation and movement of nitrogen within an HRU is calculated inWAT as a function of nutrient cycles. Losses of N from the soil sys-em in SWAT happen by crop uptake, surface runoff, lateral flow,ercolation and erosion. A more comprehensive description of theodel is given by Neitsch et al. (2005).
.3. Model inputs and setup
In this study, we applied Arc-SWAT version 2.1.4 (Winchell et al.,007) in the ArcGIS (version 9.2) environment. Recently availableIS maps for topography, landuse, and soils of the study area were
nd Environment 139 (2010) 675–688 681
used. Table 2 gives an overview of the input data. Typical manage-ment data such as crops grown, fertilizer application, and tillageoperations for different landuses were collected from the state agri-cultural statistics, the statistical yearbook of Information Center ofMOJAH and personal communication with some farmers.
Management operations used for simulation are shown inTables 3–5, for rainfed wheat, irrigated wheat and potato, respec-tively. The management practices (i.e. planting, harvesting andfertilizer application date and amount) used in this study are basedon the average long-term data from MOJAH.
Hen manure is the most used organic fertilizer in the region.As there was no information on hen manure in the SWAT database,we sampled this manure from three different locations in the studyarea and analyzed them. The kg min-N/kg fertilizer, kg org-N/kg fer-tilizer and kg NH3-N/kg min-N of hen manure were 0.0025, 0.0311,and 0.0023, respectively.
On the basis of the DEM and stream network, a drainage areaof 1200 ha was used to discretize the Hamadan–Bahar watershedinto 64 sub-basins. Also, based on the landuse and soil classes, thewatershed was subdivided into 140 HRUs (To define HRU, 10% and30% was selected as threshold for landuse and soil, respectively).
For better simulation of hydrology, the daily operation ofEkbatan dam, and constant monthly loading for some springs wereincorporated into the model.
To adapt SWAT to local conditions we adjusted the curve num-ber (CN) based on the slope using the following equations:
where CN1 is the moisture condition I curve number for the default5% slope, CN2 is the moisture condition II curve number for thedefault 5% slope, CN3 is the moisture condition III curve numberfor the default 5% slope, CN2S is the moisture condition II curvenumber adjusted for slope, and slp is the average fraction slopeof the subbasin (Neitsch et al., 2005). Moisture condition I curvenumber is the lowest value a daily curve number can assume indry conditions. More information on this topic can be found in theSWAT theoretical manual (Neitsch et al., 2005).
Furthermore, to adjust for the effect of elevation on temperatureand rainfall, we applied elevation bands to sub-basins 42, 45, 47, 49,52, 55, 56, 58, 61, 62, and 64 that were located at higher elevations(Fig. 1). The lapse rates of 1 mm km−1 and 6 ◦C km−1 were appliedto rainfall and temperature, respectively.
The simulation period for calibration of discharge and crop yieldwas 1997–2008; the first 3 years were used as warm-up period tomitigate the unknown initial conditions and were excluded fromthe analysis. The validation of discharge and crop yield period was1989–1999, also 3 years of warm-up period was applied. For nitratecalibration, the simulation period for calibration was from Novem-ber 2007 to March 2008. Lack of sufficient nitrate data for validationwas a challenge in this study and should be addressed further infuture studies.
2.4. Model calibration procedures
Model calibration and validation was based on river dischargedata from 7 gauging stations, nitrate data from 1 gauging station,
rainfed wheat yield, irrigated wheat yield, and potato yield data.As crop yield is directly proportional to actual evapotranspiration(FAO, 1986; Jensen, 1968), model calibration using crop yield givesmore confidence on the partitioning of water between soil storage,actual evapotranspiration, aquifer recharge (Faramarzi et al., 2009,
682 S. Akhavan et al. / Agriculture, Ecosystems and Environment 139 (2010) 675–688
0.00
0.20
0.40
0.60
0.80
1.00
1.20
Bahar Hamadan
Ra
infe
d w
he
at
(to
n h
a-1
)
0.00
0.20
0.40
0.60
0.80
1.00
1.20
Bahar Hamadan
Ra
infe
d w
he
at
(to
n h
a-1
)
0.00
1.00
2.00
3.00
4.00
5.00
Bahar Hamadan
Irri
ga
ted
Wh
ea
t Y
ield
( t
on
ha
-1)
0.00
1.00
2.00
3.00
4.00
5.00
Bahar Hamadan
Irri
ga
ted
Wh
ea
t Y
ield
( t
on
ha
-1)
0
5
10
15
20
25
30
35
40
45
Bahar Hamadan
Po
tato
Yie
ld (
ton
ha
-1)
0
5
10
15
20
25
30
35
40
45
Bahar Hamadan
Po
tato
Yie
ld (
ton
ha-1
)
F e (200i are op is figu
2wm
(yutPm(Mia2(
tc
ig. 4. Calibration (left graph) and validation (right graph) results of annual averagrrigated wheat and potato in Bahar and Hamadan regions. The red and blue pointsrediction uncertainty (95PPU). (For interpretation of the references to colour in th
010), and nitrogen uptake by plants. Nitrate data from 30 wellsere used for a qualitative assessment of SWAT-predicted leachingap of the region.In this study, SUFI-2 (Sequential Uncertainty Fitting, ver. 2)
Abbaspour, 2007) was used for calibration and uncertainty anal-sis. The program was applied for a combined calibration andncertainty analysis. This program is currently linked to SWAT inhe calibration package SWAT-CUP (SWAT Calibration Uncertaintyrocedures) and contains Generalized Likelihood Uncertainty Esti-ation (GLUE) (Beven and Binley, 1992), Parameter Solution
ParaSol) (Van Griensven and Meixner, 2006), and a Monte Carloarkov Chain (MCMC) (Vrugt et al., 2003) algorithm. Previous stud-
es have shown that SUFI-2 program is very efficient in calibrationnd uncertainty quantification of large watersheds (Faramarzi et al.,009; Schuol et al., 2008a,b; Yang et al., 2008) and small watersheds
Abbaspour et al., 2007; Rostamian et al., 2008).
In SUFI-2 parameter uncertainty represents all sources of uncer-ainties such as uncertainty in the driving variables (e.g. rainfall),onceptual model, parameters, and measured data. The degree to
0–2008 for calibration and 1992–1999 for validation) crop yield for rainfed wheat,bservation and best simulation, respectively, and the gray band expresses the 95%re legend, the reader is referred to the web version of the article.)
which all uncertainties are recognized is measured by an indexreferred to as the P-factor, which is the percentage of measured databracketed by the 95% prediction uncertainty (95PPU). The 95PPU iscalculated at a rate of 2.5% and 97.5% levels of cumulative distribu-tion of output variable obtained by Latin hypercube sampling. As allactual processes are reflected in the measurements (e.g, discharge),to the degree that the 95PPU brackets the measured data it accountsfor the uncertainties. Hence, the P-factor is used as a measure of theuncertainties captured by the parameters ranges. Another measureto quantify the strength of a calibration/uncertainty analysis is theR-factor, which is the average thickness of the 95PPU band dividedby the standard deviation of measured data. SUFI-2, searches tobracket most of the measured data (P-factor approaching the max-imum value of 100%) with the smallest possible uncertainty band(R-factor approaching the minimum value of zero) (Abbaspour,
2007).
To assess the monthly measured and simulated discharges andnitrate, a weighted version of the coefficient of determination,slightly modified from Krause et al. (2005), was chosen as efficiency
Fig. 5. Calibration results for monthly nitrate loads carried by the river at the water-shed outlet. The red line is the observed load, the blue lines is the best simulation,ttov
c
˚
wibbiiwos
a
g
w
t
M
wvcsM
w8(
2
ao
he gray band expresses the 95% prediction uncertainty, and the yellow lines showhe range of assumed error (10%) around the measured nitrate. (For interpretationf the references to colour in this figure legend, the reader is referred to the webersion of the article.)
riteria, ˚:
={
|b|R2 if|b| ≤ 1|b|−1R2 if|b| > 1
(4)
here the coefficient of determination (R2) represents the dynam-cs of discharge (nitrate), and b is the slope of the regression lineetween the monthly observed and simulated variables. Includingguarantees that variable under or over-estimates are also taken
nto account. The main advantage of this efficiency criterion is thatt ranges from 0 to 1, which compared to Nasch–Sutcliff coefficient
ith a range from −∞ to 1 ensures that in a multi-site calibrationbjective function is not control by one or a few poorly simulatedtations (Faramarzi et al., 2009; Schuol et al., 2008b).
For multiple discharge stations, the objective function was anverage of ˚ for all stations within the study area:
= 1n
n∑i=1
˚i (5)
here n is the number of stations.To compare the measured and simulated crop yield (using his-
orical crop yields data), the mean squared error was used:
SE = 1n�2
n∑i=1
(Yobs − Ysim)2i (6)
here n is the number of observations, �2 is variance of the obser-ations, Yobs is actual crop yield (ton ha−1), and Ysim is simulatedrop yield by SWAT model (ton ha−1). In the result and discussionection, the root mean square error (RMSE) is reported instead ofSE for a better comparison of the results with observation.In the SWAT model, the harvested yield is reported as dry
eight. Therefore for calibration we modified the SWAT yield for0% moisture in potato (Akhavan et al., 2005) and 14% in wheatNour-Mohamadi et al., 1997).
.5. Sampling and analysis
Monthly groundwater samples were collected from 30 wellscross the region of study (one instantaneous sample at the middlef each month from October 2007 to September 2008). The water
nd Environment 139 (2010) 675–688 683
table was 2–50 m in the sampled wells, but in most of the wells,the water table was greater than 10 m. Daily instantaneous sampleswere collected at the outlet of the watershed (Koshkabad station)during fall 2007 to spring 2008. The samples were placed in plas-tic containers and kept in the refrigerator prior to analysis. Thesamples were analyzed for NO3
− by using a DR/4000 Hach spec-trophotometer. Also, for calibration of nitrate at the outlet of thewatershed, the following equation was applied to calculate nitrateload (Neitsch et al., 2005):
NLoad = 570.24NO−3 × q (7)
where NLoad is the monthly nitrate loads carried by surface runoff(kg N month−1), 570.24 is conversion factor, NO−
3 is nitrate concen-tration (mg l−1), and q is discharge in the river reach (m3 s−1).
3. Results and discussion
3.1. Calibration and uncertainty analysis
Table 6 has a listing of the SWAT model parameters included inthe calibration process and their sensitivity statistics. The resultsof the sensitivity analysis indicated that parameters selected forstream flow were sensitive parameters. The t-value (Table 6) pro-vides a measure of sensitivity (larger values are more sensitive) andp-values determine the significance of the p-value (the smaller, themore significant) (Abbaspour, 2007).
Curve number was the most sensitive parameter for streamflow, also reported by Faramarzi et al. (2009). After curve number,snow parameters (snowfall temperature, snowmelt base temper-ature, snow pack temperature lag factor and minimum melt ratefor snow during the year) were the most sensitive parameters. Thereason for the larger sensitivities of the snow parameters is thatHamadan–Bahar watershed is mountainous and snowmelt con-trols much of the stream flows. To consider the sensitivity of nitrateparameters, first the model was calibrated for hydrology and cropyield, then these parameters were fixed and the model calibratedfor nitrate. The sensitivity results are shown in Table 6.
The parameters given in Table 6 were further parameterizedbased on the different soils, landuses, and sub-basins. This resultedin a total of 164 parameters. This option in SWAT-CUP provides theanalyst with more flexibility in parameterizing the watershed (seealso Abbaspour, 2007; Faramarzi et al., 2009; Schuol et al., 2008b).The initial and final range of sensitive parameters is indicated inTable 7. In this table, minimum and maximum range of each param-eter is reported for the whole watershed, and not for each of the164 parameters, which are differentiated based on different soils,landuses, and sub-basins.
The calibration statistics for the seven discharge stations isshown in Table 8. The R-factor of less than 1 generally shows agood calibration result. This is obvious for all seven stations. But,the P-factor for all seven stations is too small indicating that theactual uncertainty is likely larger. This could however be improvedat the expense of a larger R-factor.
The 95PPUs are the integrated result of uncertainties in theconceptual model, the parameters and also the input data. Unlessotherwise indicated, all ranges reported hereafter are the 95% pre-diction uncertainties. As reported by Schuol et al. (2008a), eachhydrological model suffers from uncertainties of conceptual mod-els and this is especially true for models of watershed where many
processes (natural or artificial) may not be represented in the mod-els.
In Fig. 2, calibration and validation results are shown for three(typical) discharge stations. The 95PPU together with the observeddischarges are illustrated.
In the first calibration of Abbasabad hydrometric station theodel could not predict the base flow, because in this station the
ase flow of the river is supplied by springs. In the second attemptFig. 2e), we imported the spring in this sub-basin as point source.he P-factor value increased from 18% to 57%.
The reason for small value of the P-factor (large uncertainty) ishe unaccounted for human activities affecting natural hydrologyuring the period of study. Based on the management map illus-rated in Fig. 3, most parts of the watershed are under intensivegriculture, constructed artificial recharge sites for groundwater,nd check dams for irrigation, that affects calibration and vali-ation results. Therefore, a careful examination of the calibrationesults for Koshkabad station indicates that a large number ofn-bracketed data fall in the region of the base flow because of
nsufficient accounting of agricultural water use in the model.On the other hand, as shown in Fig. 2, the flow dynamics
s simulated quite well for Koshkabad (R2 = 0.83) and AbbasabadR2 = 0.63) stations and relatively good for Yalfan station (R2 = 0.50).ut there are large uncertainties on the peak values in several occa-ions (Fig. 2). In Abbasbbad and Yalfan stations that are located inountainous part of the watershed, underestimation of discharge,
specially the peak values, is quite large, although we accounted forainfall and temperature changes due to elevation. One reason forhis underestimation could be that SWAT classifies precipitation asain or snow based on the average daily temperature for the entireatershed and snow parameters are not spatially defined. Second
ause could be that SCS method cannot simulate runoff from melt-ng snow and on frozen ground (Maidment, 1992). Due to obviousimitations of the model to simulate discharge in March, April and
ay, it seems that we were not able to adequately calibrate SWATo simulate the snowmelt in the study area. This has also been thexperience of Fontaine et al. (2002) and Rostamian et al. (2008).
Validation results for the hydrologic model of Hamadan–Baharatershed are shown in Table 8 and Fig. 2b, d and f. The P-factor
nd R-factor statistics are similar to calibration results indicatingonsistency in model simulation for the calibration and validationeriods.
Crop yield was simulated for wheat and potatoes. Calibration ofrop yield increases model reliability for simulation of crop growth,oil moisture, evapotranspiration, water percolation from the rootone, and plant uptake of nitrogen. Table 7 shows a list of croparameter ranges in the first and final calibration results for rainfedheat, irrigated wheat, and potato. The final ranges of crop param-
ters are smaller than the initial values showing the significance ofalibration to decrease the uncertainty band.
The RMSE and R-factor of calibration and validation of crop yieldre reported in Table 9. For calibration period, the minimum andaximum RMSE are 0.08 and 1.69 ton ha−1 for rainfed wheat and
otato in Hamadan area, respectively. For validation period, rain-ed wheat and potato in Hamadan area have a minimum RMSEf 0.12 ton ha−1 and a maximum RMSE of 4.22 ton ha−1. The largealue of RMSE for potato could be due to the lack of accountingf management practices in the region such as tillage operation,ertilizer application, irrigation operation and planting date as alsoeported by Faramarzi et al. (2010). For example, planting date forotato is reported to be during March 6th to April 20th. Not know-
ng the exact date, we assumed 17th of March as planting of potaton Hamadan area.
In the calibration period, the R-factors of 1.19 and 1.2 werebtained for rainfed wheat, 1.36 and 2.60 for irrigated wheat, and.10 and 0.96 for potato in Bahar and Hamadan regions, respec-
ively. According to this criterion, simulation of irrigated wheatf Hamadan indicated large uncertainty, but the uncertainty ofodel simulation is relatively small for rainfed wheat and potato.
he results of calibration and validation for average annual rainfedheat, irrigated wheat and potato in Hamadan and Bahar regions
Fig. 7. Average (2000–2008) monthly 95PPU ranges of nitrate leaching, percola-tion and irrigation with average (2000–2008) monthly precipitation for the entireHamadan–Bahar watershed.
are shown in Fig. 4. As shown in this figure, the uncertaintiesfor wheat are rather larger than those for potatoes, although theobserved potato yield in the Hamedan region during validation fallsoutside of the 95PPU band. This large uncertainty of yield is mostlydue to the differences in modeled and actual management prac-tices. It is noteworthy to mention that natural calamities such asreduced yield due to pest in not accounted for in the SWAT model,which also could add to the model uncertainty.
The calibration result of nitrate loads carried by the river at thewatershed outlet is shown in Fig. 5. Similar to the discharge atthe outlet of the watershed, nitrate simulation is reasonable, butwith large uncertainty (R-factor = 3.89), while bracketing 100% ofthe data (P-factor = 1). As nitrate transport is largely controlled byflow, and discharge uncertainty at the watershed outlet is large,hence, the nitrate uncertainty is also large. There were not enoughdata to validate the nitrate model at the watershed outlet.
The SWAT model in this study is calibrated simultaneously fordischarge, nitrate, and wheat and potato yields. For this reason theuncertainties are larger than if the model was calibrated only forone variable, as was also noted by Abbaspour et al. (2007).
3.2. Quantification of nitrate leaching
As reported by Pohlert et al. (2007), nitrate leaching and denitri-fication are two highly competitive processes. Due to the cascademodel of percolation, the water percolates into a layer of underly-ing soil, when field capacity is exceeded. Therefore, denitrificationtakes place before water begins flowing, which leads to unusualhigh N losses and a whole depletion of nitrate in the pools of eachsoil layer. In this study, denitrification threshold water content(SDNCO in basins.bsn file) was selected for calibration of nitrate.The best value for this parameter was obtained as 1 and denitrifi-cation value was zero. This value for denitrification is acceptablein Hamadan–Bahar watershed, because denitrification becomesmeaningful when the soil is waterlogged for 36 h or more and suchcondition hardly occurs in the region of study as the soils are well-drained (Killpack and Buchholz, 1993). Fertilization, crop uptake,nitrate leaching, and mineralization were the most important com-ponents of simulated nitrogen balance. Denitrification indicateda small contribution to the overall simulated nitrogen balance(Peralta and Stockle, 2001). Peralta and Stockle (2001) showed that
simulated denitrification was less than 1% of applied nitrogen forpotato and maize.
To show the spatial distribution of nitrate leaching, in Fig. 6we plotted the leaching results along with water percolationand applied nitrogen. As the model results are generated with
686 S. Akhavan et al. / Agriculture, Ecosystems and Environment 139 (2010) 675–688
0
5
10
15
20
25
30
35
40
45
Jan
Feb
Mar
Ap
r
May
Jun
Jul
Au
g
Sep
Oct
No
v
Dec
Nit
rate
Lea
chin
g (
kg N
ha-1
)P
erco
lati
on
(m
m)
0
5
10
15
20
25
30
35
40
45
Gro
un
dw
ater
Nit
rate
(m
g l-1
)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Jan
Feb
Mar
Ap
r
May
Jun
Jul
Au
g
Sep
Oct
No
v
Dec
Jan
Feb
Mar
Ap
r
May
Jun
Jul
Au
g
Sep
Oct
No
v
Dec
Jan
Feb
Mar
Ap
r
May
Jun
Jul
Au
g
Sep
Oct
No
v
Dec
Jan
Feb
Mar
Ap
r
May
Jun
Jul
Au
g
Sep
Oct
No
v
Dec
Jan
Feb
Mar
Ap
r
May
Jun
Jul
Au
g
Sep
Oct
No
v
Dec
Jan
Feb
Mar
Ap
r
May
Jun
Jul
Au
g
Sep
Oct
No
v
Dec
Jan
Feb
Mar
Ap
r
May
Jun
Jul
Au
g
Sep
Oct
No
v
Dec
Nit
rate
Lea
chin
g (
kgN
ha-
1)
0
2
4
6
8
10
12
14
Gro
un
dw
ater
Nit
rate
(m
g l-1
)
0
20
40
60
80
100
120
140
Nit
rate
Lea
chin
g (
kg N
ha-1
)P
erco
lati
on
(m
m)
0
20
40
60
80
100
120
140
Gro
un
dw
ater
Nit
rate
(m
g l-1
)
0
20
40
60
80100
120
140
160
180
200
Nit
rate
Lea
chin
g (
kg N
ha-1
)0
20
40
60
80100
120
140
160
180
200
Gro
un
dw
ater
Nit
rate
(m
g l-1
)
20
40
60
80
100
120
140
160
Nit
rate
Lea
chin
g (
kg N
ha-1
)P
erco
lati
on
(m
m)
20
40
60
80
100
120
140
160
Gro
un
dw
ater
Nit
rate
(m
g l-1
)
0
20
40
60
80
100
120
140
Nit
rate
Lea
chin
g (
kgN
ha-1
)
0
20
40
60
80
100
120
140
Gro
un
dw
ater
Nit
rate
(m
g l-1
)
01020304050
60708090
100
Nit
rate
Lea
chin
g (
kg N
ha-1
)P
erco
lati
on
(m
m)
Per
cola
tio
n (
mm
)P
erco
lati
on
(m
m)
Per
cola
tio
n (
mm
)P
erco
lati
on
(m
m)
01020304050
60708090100
Gro
un
dw
ater
Nit
rate
(m
g l-1
)
0
10
20
30
40
50
60
70
80
Nit
rate
Lea
chin
g (
kgN
ha-1
)
0
10
20
30
40
50
60
70
80
Gro
un
dw
ater
Nit
rate
(m
g l-1
)
a
c d
b
e f
g h
Fig. 8. Average (2000–2008) monthly 95PPU ranges of nitrate leaching (green band) and percolation (blue band) in comparison with nitrate concentration in groundwater(red points) for rainfed wheat rotation (a and b), irrigated wheat rotation (c and d), and potato rotation (e–g) of Hamadan–Bahar watershed. (For interpretation of ther f the
utLpH07f
eferences to colour in this figure legend, the reader is referred to the web version o
ncertainties, we plotted the map at the 50% probability level ofhe cumulative distribution of nitrate leaching obtained throughatin hypercube sampling for the period of 2000–2008. The
redicted nitrate leaching demonstrates high variability acrossamadan–Bahar watershed. The simulated rates varied fromto 595 kg N ha−1 year−1 and percolation ranges from 0 to
75 mm year−1 depending on crop rotation and amount of appliedertilizer.
article.)
Fig. 6a shows a clear spatial pattern for N leaching, with thegreatest leaching taking place in the south-west and central partsof the plain, corresponding to the areas where high N-demanding
irrigated crops are often produced (Fig. 6c). Fig. 6b shows that thelargest percolation occurs in the north. Hence, the largest nitro-gen leaching occurs in the areas that received the largest amountof nitrogen fertilizer, and not where the most percolation occurs.Thus, less fertilization is the most effective approach to reduce
ems a
grS
(fio1prritrcg(mN
nspApuhi1shvti
bnonm
3c
aswcctauongtcAcligig
S. Akhavan et al. / Agriculture, Ecosyst
roundwater nitrogen pollution and minimize the long-term envi-onmental impact of excess nitrate (Karlen et al., 1998; Peralta andtockle, 2001).
Based on the works of De Paz et al. (2008) and Delgado et al.2006), the nitrate leaching rate in the study area was divided intove classes (Table 10). According to this classification, about 36%f Hamadan–Bahar aquifer has a nitrate leaching rate higher than00 kg N ha−1 year−1. The average nitrate leaching rate for potato,otato rotation with wheat, irrigated wheat, and rainfed wheatotations was 254–361, 63–125, 14–24 and 2–6 kg N ha−1 year−1,espectively. Potato farms show the largest nitrate leaching ratesn the region. About 36% (30–42%) and 13% (11–16%) of nitrogen fer-ilizer added for potato and wheat are leached from the soil profile,espectively. According to previous studies, nitrate leaching lossesan change from zero to 60% of the applied N, but losses from usualrain crop vary from 10% to 30% of N applied in fertilizer or manurePratt, 1984; Randall and Iragavarapu, 1995). Delgado (2001) esti-
ated that nitrate leaching loss for small grains was 10% of applied(146 kg N ha−1).The relationship between average monthly percolation and
itrate leaching (2000–2008 period) for Hamadan–Bahar water-hed is shown in Fig. 7. The uncertainty ranges of average monthlyercolation and nitrate leaching are quite large for March and April.s it was mentioned before, there is a large uncertainty interval ateaks in March and April. Therefore, those are affecting predictedncertainty of the percolation and nitrate leaching. On the otherand, according to Tables 3–5, some fertilizers were applied dur-
ng the fall season, especially large amount of hen manure (about5 ton ha−1) for potato crop. These fertilizers provide significantource of nitrogen during March wet month. Following March, weave large uncertainty and large value for percolation in April, butalue of nitrate leaching is small. The main reason is that most ofhe nitrates in the soil layers are leached during March and the rests taken up by crops.
As shown in Fig. 7, the percolation and nitrate leaching is causedy irrigation for months with no rain (June–September). The largeritrate leaching is likely related to greater percolation in areasf high fertilizer application. The correlation coefficient betweenitrate leaching and percolation is around 0.88 for the summeronths.
.3. Comparison of simulated nitrate leaching with nitrateoncentrations in groundwater
Prior to this study, nitrate leaching and associated uncertaintynalysis has not been estimated with such a fine temporal andpatial resolution for Hamadan–Bahar watershed. For this reason,e could not find any previous field data for this watershed for
omparison. As reported by De Paz et al. (2008), it is difficult toorrelate the leaching of nitrates in the root zone with the concen-ration of nitrates in the groundwater, because other factors suchs groundwater depth, lateral flow and denitrification in the unsat-rated zone also play a role. But in order to have some measuref correlation, we defined a classification criterion to compare theitrate leaching with annual average nitrate concentration in theroundwater. Based on this classification a leaching rate of morehan 100 kg N ha−1 year−1 was classified as high leaching, also aoncentration of more than 45 mg l−1 was classified as high value.ccording to this criterion, there was 73% overlap between nitrateoncentration in groundwater (30 wells) and simulated nitrate
eaching. This result demonstrates that simulation of nitrate leach-ng by SWAT model for Hamadan–Bahar watershed matches theroundwater concentration quite well. In summary, nitrate leach-ng values of more than 100 kg N ha−1 year−1 result in a high risk toroundwater nitrate pollution. Similar results were reported by De
nd Environment 139 (2010) 675–688 687
Paz et al. (2008). For a more detail comparison of nitrate leachingsimulated by SWAT model, and nitrate concentration in groundwa-ter, we show monthly variation of nitrate leaching and percolationfor rainfed wheat, irrigated wheat, and potato in Fig. 8.
Fig. 8a–d shows that no significant nitrate leaching eventsoccurred under wheat during the study period. Both wells in rain-fed (Fig. 8a and b) and irrigated (Fig. 8c and d) regions show smallgroundwater nitrate concentration as well.
Fig. 8e–h, however, show large variations in nitrate leach-ing from the bottom of the root zone as well as groundwaternitrate concentration. In these highly fertilized regions both nitrateleaching and groundwater concentrations were high and oftenabove the standard concentration. Observed groundwater nitratevariability is substantially less than nitrate leaching, which is prob-ably explained by the dilution of leached nitrate entering deepaquifer.
Up to now we discussed much about the uncertainties associ-ated with model prediction for different variables, and they werenot small. A question may arise as to the ability of the distributedmodels to suggest real management options for the mitigation ofthe existing problems in the watershed. As a general note on theusually large predicted model uncertainty, it could be argued that“large uncertainty” is not equivalent to “unpredictability”. But ithas been shown (Reichert and Borsuk, 2005) that uncertainty in thedifference of model predictions corresponding to different policiesmay be significantly smaller than the uncertainty in the predic-tions themselves. The shown uncertainties in Figs. 4, 5, 7, and 8are convoluted showing an integration of all kinds of uncertain-ties, including natural variability. For a practical application, it ispossible to decrease this uncertainty by accounting only for someselected uncertainty sources of interest, and collection of better andmore relevant data from the watershed.
4. Conclusions
The SWAT model was successfully employed to simulate runoff,crop yield and river nitrate for Hamadan–Bahar watershed withuncertainty analysis. Two important modifications in SWAT setupwere the implementation in the irrigation module to estimate per-colation and the consideration of spring discharge to predict baseflow. One of the difficulties and limitations within this study wasthe lack of data on the amount of irrigation water that is withdrawnfrom rivers or creeks. However given the complexities of this water-shed and the large number of interactive processes happening, theresults of calibrated SWAT model were quite satisfactory (R2 = 0.83,NS = 0.77) at the outlet of the watershed.
Model validation was also carried out for the hydrologic modelwith results similar to calibration. After validation for hydrology,the model was used to examine the spatial and temporal leach-ing of nitrate from the root zone. Spatial and temporal nitrateleaching simulation was qualitatively validated with the existinggroundwater nitrate data. The result of this validation was alsoquite satisfactory as there where 73% overlap in the criterion wehad defined. In a follow up study we will look at an applicationof the model developed here to determine a best managementpractice to decrease nitrate leaching while maintaining agriculturalprofitability.
Acknowledgements
The authors wish to acknowledge Hamadan Regional WaterAuthority, Water and Wastewater Co. of Hamadan, Isfahan Univer-sity of Technology, the Swiss Federal Institute for Aquatic scienceand Technology (Eawag), and Jahade-Agriculture of Hamadan forproviding assistance to conduct this study.
bbaspour, K.C., 2007. User Manual for SWAT-CUP SWAT Calibration and Uncer-tainty Analysis Programs. Swiss Federal Institute of Aquatic Science and Tech-nology, Eawag, Dübendorf, Switzerland, <http://www.eawag.ch/organisation/abteilungen/siam/software/swat/index EN> (Last Accessed January 2010).
bbaspour, K.C., Yang, J., Maximov, I., Siber, R., Bogner, K., Mieleitner, J., Zobrist,J., Srinivasan, R., 2007. Modelling hydrology and water quality in the pre-alpine/alpine Thur watershed using SWAT. J. Hydrol. 333 (2–4), 413–430.
khavan, S., Mostafazadeh, B., Mousavi, S.F., Ghadami Firoz-Abadi, A., Bahrami, B.,2005. Effect of irrigation amount and method on yield, yield components andquality of potato. Agric. Res. 5 (2), 41–52 (in Persian).
rnold, J.G., Allen, P.M., 1996. Estimating hydrologic budgets for three Illinois water-sheds. J. Hydrol. 176, 57–77.
rnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologicmodeling and assessment—Part 1. Model development. J. Am. Water Resour.Assoc. 34, 73–89.
even, K., Binley, A., 1992. The future of distributed models—model calibration anduncertainty prediction. Hydrol. Process. 6, 279–298.
ouraoui, F., Grizzetti, B., 2008. An integrated modelling framework to estimate thefate of nutrients: application to the Loire (France). Ecol. Model. 212, 450–459.
elgado, J.A., 2001. Use of simulations for evaluation of best management practiceson irrigated cropping systems. In: Shaffer, M.J., Hansen, L., Ma, S. (Eds.), ModelingCarbon and Nitrogen Dynamics for Soil Management. Lewis Publishers, BocaRaton, FL, pp. 355–381.
elgado, J.A., Shaffer, M., Hu, C., Lavado, R.S., Cueto-Wong, J., Joosse, P., Li, X., Rimski-Korsakov, H., Follett, R., Colon, W., Sotomayor, D., 2006. A decade of change innutrient management: a new nitrogen index. J. Soil Water Conserv. 61, 66A–75A.
e Paz, J.M., Delgado, J.A., Ramos, C., Shaffer, M.J., Barbarick, K.K., 2008. Use of a newGIS nitrogen index assessment tool for evaluation of nitrate leaching across aMediterranean region. J. Hydrol. 365 (3–4), 183–194.
aramarzi, M., Abbaspour, K.C., Schulin, R., Yang, H., 2009. Modelling blue and greenwater resources availability in Iran. Hydrol. Process. 23, 486–501.
aramarzi, M., Yang, H., Schulin, R., Abbaspour, K.C., 2010. Modeling wheat yield andcrop water productivity in Iran: implications of agricultural water managementfor wheat production. Agric. Water Manage. 97, 1861–1875.
ontaine, T.A., Cruickshank, T.S., Arnold, J.G., Hotchkiss, R.H., 2002. Development ofa snowfall–snowmelt routine for mountainous terrain for the soil water assess-ment tool (SWAT). J. Hydrol. 262 (1–4), 209–223.
AO, Food and Agriculture Organization, 1986. Yield response to water. In: Irrigationand Drainage Paper 33. FAO, Rome, Italy.
assman, P.W., Reyes, M.R., Green, C.H., Arnold, J.G., 2007. The soil and water assess-ment tool historical development, applications, and future research directions.Trans. ASABE 50 (4), 1211–1250.
rizzetti, B., Bouraoui, F., Granlund, K., Rekolainen, S., Bidoglio, G., 2003. Mod-elling diffuse emission and retention of nutrients in the Vantaanjoki watershed(Finland) using the SWAT model. Ecol. Model. 169, 25–38.
alali, M., 2005. Nitrates leaching from agricultural land in Hamadan, western Iran.Agric. Ecosyst. Environ. 110, 210–218.
ensen, M.E., 1968. Water Consumption by Agricultural Plants. In Water Deficits inPlant Growth (1). Academic Press, New York, pp. 1–22.
arlen, D.L., Kramer, L.A., Logsdon, S.D., 1998. Field-scale nitrogen balances associ-ated with long-term continuous corn production. Agron. J. 90, 644–650.
illpack, S.C., Buchholz, D., 1993. Nitrogen in the Environment: Denitrification.
rause, P., Boyle, D.P., Base, F., 2005. Comparison of different efficiency criteria forhydrological model assessment. Adv. Geosci. 5, 89–97.
uo, Y., He, C., Sophocleous, M., Yin, Z., Hongrui, R., Ouyang, Z., 2008. Assessment ofcrop growth and soil water modules in SWAT2000 using extensive field exper-
nd Environment 139 (2010) 675–688
iment data in an irrigation district of the Yellow River Basin. J. Hydrol. 352,139–156.
Lord, E.I., Shepherd, M.A., 1993. Developments in the use of porous ceramic cups formeasuring nitrate leaching. J. Soil Sci. 44, 435–449.
Maidment, D.R., 1992. Handbook of Hydrology. McGraw-Hill, Co., New York.Nadafian, H., 2007. Simulation of groundwater pollution around drinking water
wells in Hamadan city. M.Sc. Dissertation, Shahid Beheshti University, Tehran,Iran (in Persian).
Nour-Mohamadi, G., Siadat, A., Kashani, A., 1997. Agronomy Cereal Crops, vol. 1,Shahid Chamran University, Ahwaz, Iran (in Persian).
Peralta, J.M., Stockle, C.O., 2001. Dynamics of nitrate leaching under irrigated potatorotation in Washington State: a long-term simulation study. Agric. Ecosyst. Env-iron. 88, 23–34.
Pohlert, T., Huisman, J.A., Breuer, L., Frede, H.G., 2007. Integration of a detailedbiogeochemical model into SWAT for improved nitrogen predictions-Modeldevelopment, sensitivity, and GLUE analysis. Ecol. Model. 203, 215–228.
Pratt, P.E., 1984. Nitrogen use and nitrate leaching in irrigated agriculture. In: Hauck,R.D., et al. (Eds.), Nitrogen in Crop Production. ASA, Madison, WI, pp. 319–333.
Randall, G.W., Iragavarapu, T.K., 1995. Impact of longterm tillage systems for con-tinuous corn and nitrate leaching to tile drainage. J. Environ. Qual. 24, 360–366.
Rahmani, A., 2003. Study and Investigation of Pollution in Groundwater ofHamadan–Bahar Plain. Environmental Organization of Hamadan, Iran (in Per-sian).
Reichert, P., Borsuk, M.E., 2005. Does high forecast uncertainty preclude effectivedecision support? Environ. Modell. Softw. 20 (8), 991–1001.
Rostamian, R., Jaleh, A., Afyuni, M., Mousavi, S.F., Heidarpour, M., Jalalian, A.,Abbaspour, K.C., 2008. Application of a SWAT model for estimating runoff andsediment in two mountainous basins in central Iran. Hydrol. Sci. J. 53 (5),977–988.
Saleh, A., Du, B., 2004. Evaluation of SWAT and HSPF within BASINS program forthe upper North Bosque River watershed in central Texas. Trans. ASAE 47 (4),1039–1049.
Santhi, C., Arnold, J., Williams, J., Hauck, L., Dugas, W., 2001. Application of a water-shed model to evaluate management effects on point and non-point sourcepollution. Trans. ASAE 44, 1559–1570.
Schuol, J., Abbaspour, K.C., Sarinivasan, R., Yang, H., 2008a. Estimation of freshwateravailability in the West African sub-continent using the SWAT hydrologic model.J. Hydrol. 352, 30–42.
Schuol, J., Abbaspour, K.C., Yang, H., Srinivasan, R., Zehnder, A.J.B., 2008b. Modellingblue and green water availability in Africa. Water Resour. Res. 44, 18, W07406.
Van Griensven, A., Meixner, T., 2006. Methods to quantify and identify the sources ofuncertainty for river basin water quality models. Water Sci. Technol. 53, 51–59.
Volk, M., Liersch, S., Schmidt, G., 2009. Towards the implementation of the EuropeanWater Framework Directive? Lessons learned from water quality simulations inan agricultural watershed. Land Use Policy 26, 580–588.
Vrugt, J.A., Gupta, H.V., Bouten, W., Sorooshian, S., 2003. A shuffled complex evo-lution metropolis algorithm for optimization and uncertainty assessment ofhydrologic model parameters. Water Resour. Res. 39 (1201), p18.
Williams, J.R., Jones, C.A., Dyke, P.T., 1984. A modeling approach to determining therelationship between erosion and soil productivity. Trans. ASAE 27, 129–144.
Winchell, M., Srinivasan, R., Di Luzio, M., Arnold, J.G., 2007. ArcSWAT Interface for
SWAT2005–User’s Guide. Blackland Research Center, Texas Agricultural Stationand Grassland, Soil and Water Research Laboratory, USDA Agricultural ResearchService, Temple, TX.
Yang, J., Reichert, P., Abbaspour, K.C., Xia, J., Yang, H., 2008. Comparing uncertaintyanalysis techniques for a SWAT application to Chaohe Basin in China. J. Hydrol.358, 1–23.
