You are on page 1of 13

Desalination 250 (2010) 274–286

Contents lists available at ScienceDirect

Desalination
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / d e s a l

Hydrological and water quality modeling in a medium-sized basin using the Soil and
Water Assessment Tool (SWAT)☆
Vassilios Pisinaras, Christos Petalas, Georgios D. Gikas, Alexandra Gemitzi, Vassilios A. Tsihrintzis ⁎
Department of Environmental Engineering, School of Engineering, Democritus University of Thrace, 67100 Xanthi, Greece

a r t i c l e i n f o a b s t r a c t

Article history: The newest version of Soil and Water Assessment Tool (SWAT2005), coupled with a GIS interface
Received 24 November 2007 (AVSWATX), was applied to Kosynthos River watershed located in Northeastern Greece. The 440 km2
Accepted 1 August 2008 drainage basin was discretized into 32 sub-basins using an automated delineation routine. The multiple
Available online 21 October 2009
hydrologic response unit (HRU) approach was used and the basin was discretized into 135 HRUs. The model
was calibrated and verified using continuous meteorological data from three stations, and runoff and
Keywords:
nutrient concentrations measured at four monitoring sites located within the main tributaries of the
SWAT model
Calibration
watershed, for the time period from November 2003 to November 2006. Calibration and verification results
Verification showed good agreement between simulated and measured data. Model performance was evaluated using
Hydrology several statistical parameters, such as the Nash–Sutcliffe coefficient and the normalized objective function.
Nutrients The validated model was also used to test the effect of several land use change and crop management
Management scenarios scenarios in runoff and nutrient loadings. The study showed that SWAT model, if properly validated, can be
used effectively in testing management scenarios in Mediterranean watersheds. The SWAT model
application, supported by GIS technology, proved to be a very flexible and reliable tool for water decision-
making, especially under the need for harmonization with the Water Framework Directive.
© 2009 Elsevier B.V. All rights reserved.

1. Introduction different stages of the legislative process [2,3], starting with relatively
simple ones during the characterization phase of the WFD, and more
In recent years many efforts have been made worldwide on the complex ones during the river basin management planning stage. To
abatement of point source pollution; as a result, the major cause of end up with a successful river basin management plan, in addition to
water quality deterioration of the water bodies is mostly associated describing current conditions, a variety of environmental conditions
today with non-point source pollution, due to the intensification of should be evaluated with the use of mathematical models, in an effort
agricultural activities and the development of large urban centres [1]. to forecast short and long-term impacts on the aquatic system. In the
The EU Water Framework Directive (WFD) is relatively new case of diffuse agricultural pollution, various land management op-
legislation that establishes an integrated approach on management tions have to be tested with the model.
and protection of Europe’s aquatic environment. The principal Because of the complexity of the hydrologic processes, hydrologic–
objective of the WFD is to achieve good chemical and ecological based, distributed parameter models and GIS constitute a powerful
status for receiving waters by 2015, and mandates Member States to combination for water quantity and quality assessment [4,5]. There are
develop river basin management schemes. This planning mechanism several reasons that enforce the combination of the aforementioned
is intended to ensure integrated management of the river environ- models with GIS for water resources management, the most important
ment, providing a decision-making framework for setting environ- of which are [6]: the automation of data input and output in the pre-
mental objectives. However, the management of water quality from and post-processing stage of model development, as well as the ability
non-point sources would require very expensive monitoring efforts. to develop interactive post-processing tools that provide the opportu-
Mathematical modeling is a necessary step in the implementation nity for easier understanding of hydrologic system function; and, the
of the WFD. The application of different types of models is required at continuous increase in data availability and quantity, which gives the
opportunity to investigate important hydrological variables.
This paper presents the combined application of hydrological
☆ Presented at the 1st Conference on Environmental Management, Engineering, model SWAT with GIS technology as a management tool for a
Planning and Economics (CEMEPE), Skiathos, Greece, 24–28 June, 2007. medium-sized Mediterranean basin located in Northeastern Greece.
⁎ Corresponding author. Laboratory of Ecological Engineering and Technology,
Department of Environmental Engineering, School of Engineering, Democritus University
The study aimed to assess the SWAT model performance in the area,
of Thrace, 67100 Xanthi, Greece. Tel.: +30 25410 79393; fax: +30 25410 78113. and evaluate the current management practices and several manage-
E-mail addresses: tsihrin@otenet.gr, tsihrin@env.duth.gr (V.A. Tsihrintzis). ment scenarios. The suitability of SWAT in the development of a River

0011-9164/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.desal.2009.09.044
V. Pisinaras et al. / Desalination 250 (2010) 274–286 275

Basin Management Plan for Mediterranean basins has been evaluated 20 years, the lagoon has suffered severe impacts due to point and non-
in terms of: 1) model performance; and 2) ability of the model to point sources of pollutants [8,9]. Although several measures have been
simulate relevant management scenarios for the region. taken over the last decade to reduce point source pollution, such as
sewage treatment and diversion from the watercourses, proper solid
2. Materials and methods waste management and industrial waste elimination, there are still
pollutants entering the lagoon, mainly associated with agricultural
2.1. Study area description land use practices [8,9]. For this, the different crop categories and the
area each crop covers in the watershed was determined based on
Kosynthos river basin is located in Thrace District, in north-eastern information collected from the local authorities. Then, the application
Greece. Kosynthos river total length is approximately 52 km, it rates of nitrogen and phosphorus fertilizers for the most significant
originates from Rhodope Mountains and after traversing a basin of crops were estimated based on a manual compiled by the local
about 440 km2 that includes mountain terrain, agricultural plains and authorities for the farmers of the study area, indicating the application
urban areas, it discharges to Vistonis lagoon (Fig. 1). The major rates and the application periods of the fertilizers. As shown in Table 1,
sources that can affect its water quality originate from agricultural, major crops for the study area are wheat, corn, cotton and tobacco.
urban and industrial activities taking place in the lower reaches of the There are significant quantities of N and P entering the watershed
basin. Water needs of the area are satisfied mainly by groundwater from fertilizers (about 1190 tons N and 162 tons P), indicating that
abstracted from numerous wells. The natural environment of the agricultural activities significantly affect the nutrient budget.
study area is still relatively unaffected in the greatest part of the basin. The watershed was ungaged. For this, river flow and water quality
Geologically, Kosynthos catchment belongs entirely to the Rho- monitoring was undertaken at four monitoring sites (MSs) along
dope massif, consisting of old metamorphic rocks (gneisses, marbles, Kosynthos river and its tributaries (Fig. 1). Monitoring site 1 (MS1)
schists), observed mainly in the northern part of the study area. was located in tributary Gerakas, at the northern of the city of Xanthi.
Moreover, igneous rocks (granites, granodiorites) have intruded Monitoring site 2 (MS2) was situated in the city of Xanthi. Monitoring
Rhodope massif, through magmatic events in tertiary times, and site 3 (MS3) was located in a significant tributary (Kimmeria Creek) of
outcrop in the southern part of the area, together with quaternary Kosynthos river near Kimmeria village, which confluences with
sediments. Precipitation averages 791 mm annually in the plain area, Kosynthos upstream of monitoring site 4 (MS4), located downstream
ranging from 368 to 1307 mm, while in the mountainous area it of the city of Xanthi. This study focuses on data collected between
averages 1044 mm annually, ranging from 539 to 1828 mm. October 2003 and November 2006 for river flow, and between
Kosynthos river water quality is an important aspect, as water is November 2004 and November 2006 for nitrate and soluble
used for irrigation purposes and also recharges the Xanthi's plain phosphorus concentrations. River flow measurements were con-
aquifer, which constitutes the potable and irrigation water supply of ducted using a Valeport model 801 flowmeter, while nitrates and
about 50,000 inhabitants. Moreover, Kosynthos river discharges into soluble phosphorus concentration were determined by spectropho-
Vistonis lagoon which is one of the most significant ecosystems of tometry according to standard methods [10]. A detailed study about
Greece, protected by the Ramsar Convention, and is considered as a water quantity and quality characteristics of Kosynthos river has been
first priority site under EU Natura 2000 network [7]. Over the past presented by Pisinaras et al. [11].

Fig. 1. Map of the study area, including the monitoring sites (MSs).
276 V. Pisinaras et al. / Desalination 250 (2010) 274–286

Table 1 2.3. Model parameterization


Crop distribution in the watershed, fertilizer application rates and total nutrient
quantities entering the watershed due to agricultural activities.
For the SWAT simulations the available topography, land use, soil
Crop type Cultivated % of the N fertilizer P fertilizer types and meteorological data had to be aggregated. AVSWATX gives
area (ha) cultivated application rate application rate the opportunity for pre-processing the data by applying some of the
area (kg/ha) (kg/ha)
ArcView GIS functions. This involves the creation of the river network,
Wheat 2146.0 32.1 130 15.3 the basin area, and the sub-basins. The latter step is crucial, since it
Corn 2490.3 37.3 300 39.2
creates the boundaries for the following simulation [24]. It is well
Sunflowers 4.6 0.1 140 37.1
Rice 0.5 0.0 95 19.6 known that the quality of the DEM will have a strong influence on the
Alfalfa 122.2 1.8 35 48.0 final output of the hydrological model [25]. A 50 × 50 m resolution
Cotton 385.2 5.8 125 21.8 DEM was used in this study.
Tobacco 947.9 14.2 40 19.6 The CORINE Land Cover 1:100,000 vector map was used in this
Tomatoes 72.8 1.1 167.5 65.4
study. The CORINE land cover consists of a geographical database
Other 376.9 7.6 169.1 2.5
Total 6546.5 100.0 Total N applied: Total P applied: describing vegetation and land use in 44 classes, grouped in three
1193.05 tons 162.22 tons nomenclature levels. The CORINE land cover classification codes were
converted to the SWAT land cover/plant codes, so a reclassified and
aggregated land use data set was made indicating that 0.16% of the
whole basin belongs to the “Residential-High Density” class (URHD),
2.2. Model description 1.25% belongs to the “Range-Brush” class (RNGB), 0.31% belongs to the
“Pasture” class (PAST), 7.45% belongs to the “Range-Grasses” class
The AVSWATX [12] version of the SWAT model was used. This (RNGE), 0.11% belongs to the “Wetland-Mixed” class (WETL), 0.34%
version integrates the latest version of Soil and Water Assessment Tool belongs to the “Commercial” class (UCOM), 76.07% belongs to the
(SWAT2005) [13,14] with ArcView. In AVSWATX, the pre-processing “Forest-Mixed” class (FRST), 0.32% belongs to the “Transportation”
of the data is done by applying some of the ESRI ArcView GIS class (UTRN), 12.90% belongs to the “Generic Agricultural Land” class
functionalities. (AGRR) and 1.09% belongs to the “Residential-Med/Low Density”
SWAT constitutes a river basin or watershed-scale, distributed (URML) class. Unfortunately, a detailed crop type map was not
model, which simulates the rainfall-runoff process, sediment trans- available. For this an average N and P fertilizer application was applied
port and nutrient loads in large watersheds, where complex soil, land for all agricultural (AGRR) HRUs.
use and management patterns are observed [13]. The SWAT model Data on soil attributes were obtained from soil maps provided by
has been developed by the USDA Agricultural Research Service (ARS), the Greek Department of Agriculture and a significant previous work
it incorporates features of several previous ARS models and is the for the study area [8,26]. For each sub-basin, the soil percentage in
evolution of the Simulator for Water Resources in Rural Basins clay, silt, sand, as well as percent of organic matter, were estimated for
(SWRRB) model [15]. There have also been several other models that up to six soil layers from soil section data. Then, the dominant soil
contributed to the development of SWAT, such as the Chemicals, type was determined by using the USDA-SCS soil texture classification
Runoff and Erosion from Agricultural Management Systems model with the largest coverage in the HRU. A hydrologic category (A to D)
(CREAMS) [16], the Groundwater Loading Effects of Agricultural was assigned to each HRU according to USDA-SCS [20]. All this data
Management Systems model (GLEAMS) [17], and EPIC [18]. The water was entered to the Soil Database of AVSWATX manually or in dbf
budget equation is the basis for the simulation of the hydrologic cycle format.
in SWAT [13,19]. Total runoff hydrographs are computed based on Weather data from three meteorological stations was collected for
runoff calculated separately at each sub-basin, and then routing the simulation period. One of the meteorological stations is located in
through several channels. A modified version of SCS curve number the mountainous part of the basin, while the other two are located in
method [20] is used for surface runoff computation, while the Modi- the lowlands. The weather data collected includes the daily
fied Universal Soil Loss Equation (MUSLE) [21] is used for erosion and precipitation rate, daily maximum/minimum temperature, daily
sediment yield calculation. Nutrient load and concentration predic- values of wind speed, daily values of solar radiation and daily relative
tion is based on a modification of the code in EPIC model [18,22]. humidity values.
Finally, soil surface and plant data are used to calculate evapotrans- Evapotranspiration is one mechanism by which water is removed
piration in the watershed, while precipitation and temperature data from a catchment. Three methods are provided within the SWAT
can be either provided as time series data, or simulated using a first- model for potential evapotranspiration (PET) estimation; the Pen-
order Markov chain model in the case when meteorological data time man–Monteith method [27], the Priestley–Taylor method [28] and the
series are not available. Hargreaves method [29]. The Hargreaves method was used in the
The distributed SWAT model with the use of AVSWATX interface calculation of PET in Kosynthos River catchment.
is parameterized three-dimensionally by spatial and relational data- In SWAT model application, the watershed is discretized into
bases [5]. Horizontal variability of input parameters, such as land subwatersheds, whose size depends on the threshold value (CSTV).
use, is provided by grid data, stored and operated by ArcView. CSTV is defined by the user and constitutes an important parameter
Relational databases of soil properties serve for parameterization for the definition of the HRUs, which allows watershed discretization
of a vertical model structure, because they are linked to spatial in more detail [30,24]. The model developer proposes a series of
modeling units. predefined relative CSTVs for the watershed size, and the model user
The watershed discretization in the SWAT model is approached selects the appropriate CSTV from several possible relative threshold
through subwatersheds defined by the watershed digital elevation values, as described by Romanowicz et al. [24]. According to FitzHugh
model, and Hydrologic Response Units (HRUs), which comprise and Mackay [30], significant attention has to be paid when defining
similar land use and soil type combinations within the subwa- the CSTV. CSTV assignment is followed by the watershed disaggre-
tershed. The Watershed Delineation module of AVSWATX is based gation into homogeneous subwatersheds and HRUs, where the
on some elementary raster functions provided by ArcView and the various hydrological attributes are assigned [24]. The classification
Spatial Analyst extension [5], in combination with the standard of the Kosynthos River catchment resulted in 32 sub-basins. With a
methodology based on the eight-pour point algorithm with steepest threshold value of 10% for land use and for soil types the number of
descent [23]. HRUs is 135.
V. Pisinaras et al. / Desalination 250 (2010) 274–286 277

2.4. Model calibration and verification 3. Another way to assess the calibration is through the use of
scattergrams [37,38] where predicted quantities are plotted
SWAT input parameters are physically based and are allowed to against observed ones. In a scattergram, a regression straight line
vary within a realistic uncertainty range for calibration. The SWAT of the following form is also fitted through the data:
model contains many difficult to measure or not measurable para-
meters [31]. Calibration techniques are generally referred to as either Pi = γOi ð4Þ
manual or automated. Santhi et al. [32] suggested a generalized
manual calibration procedure, indicating the most sensitive input and its slope γ is compared to the 1:1 slope (perfect match). The value of
parameters, acceptable model evaluation results and sensible ranges the slope γ is a measure of the over- (γ > 1.0) or under-prediction
of parameters uncertainty. A manual calibration procedure has also (γ < 1.0) of the model compared to the observed data. In addition, the
been presented by Gikas et al. [8]. square of the correlation coefficient R2 of the regression line is
Coffey et al. [33] recommended using the R2 and modeling computed. The lower the value of R2 falls below 1.0, the worse the
efficiency objective functions while Gikas et al. [8] recommended data correlation is, i.e., the greatest is the scatter of the data around the
using the R2, the root mean square error (MSE), the normalized line. Therefore, best calibration requires that values for both slope γ and
objective function (NOF) and scattergrams for calibration of daily R2 be as close to 1.0 as possible.
streamflow, sediment and nutrient data. The following modeling The calibration of the model was conducted for the period from
evaluation indices were used in this study: November 1st, 2003 to October 31st, 2005, using the field measure-
1. The root mean square error (RMSE) and the normalized objective ments collected at the four monitoring stations. The calibration
function (NOF) [34,35] were computed based on the following procedure was conducted first for water volume and flow rate, and
equations: then for nutrient quantities, separately for each monitoring station.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi One important rule for proper calibration is to begin calibration from
u2 N 3ffi monitoring stations located upstream and then proceed to monitoring
u
u6 ∑ ðPi  Oi Þ 7 2
u6 7
stations located downstream. Thus, the calibration sequence for
RMSE = u u6
i=1
7 ð1Þ Kosynthos River basin was the following: MS1, MS2, MS3 and MS4.
t4 N 5
Total water volume and discharge in each monitoring station was
calibrated in two steps: first, a curve number value was selected using
standard SCS tables. The default curve number value assigned by the
RMSE AVSWATX database, according to the land use and soil hydrologic
NOF = ― ð2Þ
O group of each HRU was varied within the range from ±5 of this value
until predicted and observed values at each monitoring station
where Pi are the model predicted values, Oi are the observed values approached [39]. Then the soil available water capacity (SOL-AWC)
— was calibrated; this is estimated as the difference between the in situ
for the N observations, and O is the mean of observed values.
According to Kornecki et al. [34], the ideal value of NOF is 0.0. water field capacity and the permanent wilting point, and represents
However, a model is acceptable for NOF values in the range from 0.0 the water volume that should be available to plants, if the soil,
to 1.0 when site specific data are available for calibration. In that case, inclusive of rock fragments, was at field capacity. These steps were
the model can be used to test scenarios associated with management repeated until an acceptable fit to observed water volume at the outlet
practices. was obtained. Further agreement of observed and predicted values
was achieved by adjustment of the groundwater parameters GW_RE-
2. The Nash–Sutcliffe coefficient [36], which is calculated by the VAP, REVAPMN, GWQMN and RCHRG_DP. Finally, better adjustment of
following equation: the shape of the hydrograph was achieved by varying the base flow
n alpha factor ALPHA_BF. Through this process the hydrologic budget
2
∑ ðOi  Pi Þ was continuously checked in order to avoid serious errors. Because of
i=1
NSC = 1  ð3Þ this ESCO and EPCO values were properly adjusted.
n
2
∑ ðOi  OÞ Nutrient loadings were calibrated at each monitoring station
i=1 according to the following procedure: first, the groundwater contri-
bution to stream nitrate concentrations was adjusted using GWNO3
The optimal statistical value occurs when the NSC value is closer to 1. and NPERCO parameters in order to calibrate the nitrate loadings.

Table 2
Several parameter values used for calibration of the SWAT model.

Variable name Model processes Description Normal range Actual value used

CN2 Flow Curve number −5 to + 5 (from SCS table values) − 3 to + 3 (from 65 to 78)
ESCO Flow Soil evaporation compensation factor 0–1 0.95
EPCO Flow Plant uptake compensation factor 0–1 1
SOL_AWC Flow Soil available water capacity 0–1 0.11–0.14
GW_REVAP Flow Groundwater revap coefficient 0.02–0.20 0.02
GWQMN Flow Threshold depth of water in shallow aquifer for 0.0–300.0 200
percolation to occur
RCHRG_DP Flow Deep aquifer percolation fraction 0.0–1.0 0–0.5
N Flow Manning's n for channel 0–1 0.05–0.1
ALPHA_BF Flow Base flow alpha factor 0.0–1.0 0.024–0.048
NPERCO Nitrate nitrogen Nitrogen percolation coefficient 0.0–1.0 0.2
GWNO3 Nitrate nitrogen Concentration of nitrate in groundwater contribution – 0.10
to streamflow from sub-basin
PPERCO Soluble phosphorus Phosphorus percolation coefficient 10.0–17.5 10
GWSOLP Soluble phosphorus Concentration of soluble phosphorus in groundwater – 0.08
contribution to streamflow from sub-basin
278 V. Pisinaras et al. / Desalination 250 (2010) 274–286

Fig. 2. Observed and simulated flow, and corresponding scattergrams at each monitoring station for the calibration period.
V. Pisinaras et al. / Desalination 250 (2010) 274–286 279

Fig. 3. Observed and simulated nitrate loadings, and corresponding scattergrams at each monitoring station for the calibration period.
280 V. Pisinaras et al. / Desalination 250 (2010) 274–286

Fig. 4. Observed and simulated soluble phosphorus loadings, and corresponding scattergrams at each monitoring station for the calibration period.
V. Pisinaras et al. / Desalination 250 (2010) 274–286 281

Then, groundwater contribution to steam phosphorus concentration percolation fraction (RCHRG_DP) were chosen to simulate the signif-
(GWMINP) was adjusted using GWMINP and PPERCO parameters in icant groundwater recharge [11] across the alluvial cone of Kosynthos
order to match the observed soluble phosphorus loadings with the river at the south part of the study area. The lowest values of both
predicted ones at all monitoring stations. nitrogen and phosphorus percolation coefficients were determined as
After model calibration was finished, model verification followed. most appropriate for Kosynthos river watershed. Finally, the concentra-
Model verification is the process of performing the simulation, using a tions of nitrate and soluble phosphorus in groundwater contribution to
different time-series of input data, without changing any parameter streamflow from its sub-basin were 0.1 and 0.08 mg/L, respectively.
values that may have been adjusted during calibration. The purpose of Typical comparisons of observed and predicted values are
model verification is to establish whether the model can estimate presented in Figs. 2–4 for each monitoring station and for the three
output for locations, time periods or conditions other than those that parameters, i.e., flow rate, nitrate nitrogen and soluble phosphorus
the parameter values were adjusted to fit. Model verification was quantities. One can see that all predicted values at all stations, for the
performed using meteorological and field data collected from three parameters for the entire simulated period show quiet good
November 2005 until November 2006. agreement with measured values.
Calibration statistics results are presented in Table 3 for all sites
2.5. Development of land use change and crop management scenarios and the three parameters. One can see that the NOF values are less
than 1.0 in all cases, thus the model can safely be used for estimating
The ability of a model to perform different management scenarios and determining results associated with management practices [34].
is a powerful tool for the decision-making process. For this, several Scattergrams for each parameter at three sampling sites, one in each
management scenarios were tested after calibration and verification sub-basin, are presented in Figs. 2–4. Values for the slope γ of Eq. (4)
of SWAT model in Kosynthos watershed in order to assess the impact are close to 1.0 for all parameters, particularly for flow rate (between
of land use change or crop management change, both in water flow 0.981 and 1.071). Except for MS2 (γ = 0.987) where nitrate quantities
and nutrient loadings. were slightly underpredicted, for all the other monitoring stations
Three land use change scenarios were applied in order to evaluate nitrate quantities were slightly overpredicted (γ between 1.139 and
the land use change impact on water flow. The first scenario, which is 1.292). Soluble phosphorus was generally slightly underpredicted (γ
the “100% Deforestation” scenario, is a pessimistic and extreme between 0.765 and 0.960). Correlation coefficient values are close to
scenario, and assumes the deforestation of the whole watershed 1.0 for all the simulated parameters (R2 between 0.617 and 0.958).
possibly by a forest fire. The “20% Expansion of Urban area” scenario Similarly to R2, NSC values were close to 1.0 for the three simulated
assumes an expansion of the urban area by 20%. This change is small in parameters ranging 0.617 and 0.915 (Table 3).
relation to the area of the whole watershed as urban area covers only
2% of the total watershed area. Finally, an expansion of the agricultural 3.2. Model verification
land by 20% was assumed in order to assess the impact of an increase
of agricultural activities. The last two land use change scenarios were Model verification was performed using meteorological and field data
developed through GIS by buffering the urban area polygons for the collected from November 2005 until November 2006. Calibrated
“20% Expansion of Urban area” scenario and the agricultural polygons parameter values were retained the same for the verification period as
for the “20% Expansion of Agricultural land” scenario. summarized in Table 3. Typical visual comparison of observed and
In order to evaluate the impact of alternation of different crops in predicted values for the verification period and for all monitoring stations
Kosynthos watershed, four different scenarios of crop management were is shown in Figs. 5–7. Similarly to calibration, these figures show that all
applied. For each alternative scenario, only one crop was considered predicted values at all stations, for the three parameters, for the entire
covering the entire arable area: wheat, corn, cotton or tomato. The verification period show good agreement with measured values.
fertilizer application rates shown in Table 1 were applied according Accuracy of the predictions from verification runs was determined with
to each crop type for the whole agricultural part of the watershed. the three methods also used in calibration, i.e., NOF computation Eqs. (1),
(2), use of scattergrams and Eq. (4), and the Nash–Suttclife coefficient
3. Results and discussion determination Eq. (3). NOF, NSC, γ and R2 values for the verification
period are presented in Table 3. One can see that the NOF values are less
3.1. Model calibration results than 1.0 in all cases (and in most times less than 0.5) (Table 3), thus the

Table 2 presents an overview of the SWAT2005 parameter changes


Table 3
applied during the model calibration. These changes reflect, to a large
Goodness-of-fit criteria used for calibration and verification of SWAT model.
degree, the special characteristics of this Mediterranean watershed. The
default curve numbers set by the AVSWATX user interface, i.e., the Station Parameter Calibration Verification
values recommended by the SCS Handbook [20] were sometimes Flow Nitrate Sol. Flow Nitrate Sol.
reduced by 3 units in the subwatersheds where runoff volumes needed phosphorus phosphorus
to be reduced, reflecting better soil drainage than the conditions MS1 NOF 0.402 0.157 0.218 0.274 0.331 0.254
assumed in the default SWAT database. In an opposite situation CN was NSC 0.815 0.904 0.781 0.679 0.727 0.751
increased. The runoff lag coefficient ALPHA_BF started from 0.024 and γ 1.046 1.193 0.931 1.103 0.768 0.960
R2 0.839 0.943 0.779 0.772 0.682 0.724
ended to 0.048 when storm recessions needed to be less steep. The soil
MS2 NOF 0.266 0.150 0.207 0.279 0.263 0.303
evaporation compensation factor and plant uptake compensation factor NSC 0.915 0.897 0.784 0.859 0.765 0.608
were kept at their default value of 0.95 and 1, respectively. ESCO and γ 1.000 0.987 0.786 0.944 0.841 0.765
EPCO decrease resulted in very high evapotranspiration values thus R2 0.898 0.884 0.931 0.825 0.653 0.750
affecting water balance. The relatively homogeneous soils which are MS3 NOF 0.479 0.196 0.327 0.208 0.286 0.290
NSC 0.741 0.873 0.616 0.906 0.767 0.727
mainly sandy clays and sandy loams resulted in little variance for soil
γ 0.981 1.292 0.931 0.990 1.175 0.824
available water capacity values. The groundwater coefficient that R2 0.790 0.958 0.617 0.923 0.793 0.800
controls the amount of water that moves from the shallow aquifer to MS4 NOF 0.351 0.178 0.217 0.259 0.226 0.227
the root zone, i.e., the SWAT revap parameter, was kept to the default NSC 0.858 0.861 0.764 0.889 0.815 0.785
γ 1.071 1.239 0.785 1.010 1.186 0.8900
value of 0.02 to allow more movement of water from the shallow aquifer
R2 0.865 0.922 0.909 0.867 0.767 0.791
to the unsaturated root zone. High values (up to 0.5) for the deep aquifer
282 V. Pisinaras et al. / Desalination 250 (2010) 274–286

Fig. 5. Observed and simulated flow, and corresponding scattergrams at each monitoring station for the verification period.
V. Pisinaras et al. / Desalination 250 (2010) 274–286 283

Fig. 6. Observed and simulated nitrate loadings, and corresponding scattergrams at each monitoring station for the verification period.
284 V. Pisinaras et al. / Desalination 250 (2010) 274–286

Fig. 7. Observed and simulated soluble phosphorus loadings, and corresponding scattergrams at each monitoring station for the verification period.
V. Pisinaras et al. / Desalination 250 (2010) 274–286 285

Table 4
Estimated water flow at the outlet of Kosynthos watershed for year 2006, for different land use change scenarios.

Land use scenario Flow Nitrate Soluble phosphorus

Volume (× 106 m3) % change Loading (tn) % change Loading (tn) % change

Current land use 153.86 0.00 320.90 0.00 11.30 0.00


Conversion of 20% of the forested 157.08 + 2.09 617.87 + 92.54 27.09 + 139.76
area to agricultural land
20% expansion of urban area 153.90 + 0.02 320.81 − 0.03 11.35 + 0.42
100% deforestation 186.38 + 21.13 333.23 + 3.84 13.82 + 22.33

Table 5
Estimated nutrient loads to Vistonis lagoon from different crop management scenarios in Kosynthos watershed for 2006.

Scenario Nitrate Soluble phosphorus

Fertilizer applied Nitrate loading (tn) % change Fertilizer applied Soluble phosphorus % change
as N (kg/ha) as total P (kg/ha) loading (tn)

Present crop management 167.7 320.9 0.00 25.2 11.3 0.00


Only corn 300.0 721.3 + 124.77 39.24 13.1 + 15.38
Only cotton 125.0 234.0 −27.08 21.8 9.8 − 13.37
Only wheat 130.0 242.3 −24.49 15.2 8.0 −29.28
Only tomato 167.5 322.0 + 0.34 65.4 16.0 + 41.12

model can safely be used for estimating and determining results 4. Conclusions
associated with management practices [34]. Scattergrams for each
parameter and site are presented in Figs. 5–7. Values for the slope γ of The newest version of SWAT (SWAT2005) assisted by the
Eq. (4) are close to 1.0 for all parameters, particularly for flow rate (γ AVSWATX integrated GIS environment was applied in Kosynthos
between 0.944 and 1.103) and nitrate quantities (γ between 0.768 and river watershed located at Northeastern Greece. Both calibration and
1.186). Soluble phosphorus is again slightly underpredicted (γ between verification results of SWAT model applied in Kosynthos watershed
0.765 and 0.960). Correlation coefficient values are closer to 1.0 for flow showed quite good agreement between observed and simulated data.
rate (R2 between 0.772 and 0.923) and lower for the other parameters (R2 The methodology described proved to be useful in performing
between 0.653 and 0.793 for nitrates, and between 0.724 and 0.800 for watershed-scale simulations under typical Mediterranean conditions.
soluble phosphorus), indicating some scatter around the straight line of The SWAT model application, supported by GIS technology, proved to
Eq. (4). Similarly to R2, NSC values were close to 1.0 for the three be a very useful tool in evaluating management alternatives of both
simulated parameters ranging between 0.679 and 0.906 (Table 3). land use change and crop management in rural basins. This fact makes
Generally, the agreement between observed and predicted values SWAT a flexible and reliable tool for water decision-making, especially
can be considered good for all parameters used, especially for flow under the need for harmonization with the WFD.
and nitrate, while soluble phosphorus was slightly underestimated in
all torrents. Therefore, the calibrated model, according to the input
parameters shown in Table 2, reproduces well the measured References
quantities during the verification time period, and thus, it can be [1] V. Novotny, H. Olem, Water Quality: Prevention, Identification and Management of
safely used in testing management scenarios. Diffuse Pollution, Van Nostrand Reinhold, USA, 1994.
[2] S. Rekolainen, J. Kamari, M. Hiltunen, A conceptual framework for identifying the
need and role of models in the implementation of the Water Framework Directive,
International Journal of River Basin Management 4 (2004) 347–352.
3.3. Land use change and crop management scenarios [3] C.F. Dilks, S.M. Dunn, R.C. Ferrier, in: R. Srinivasan, J. Jacobs, D. Day, K. Abbaspour
(Eds.), Evaluation of SWAT for use in Development of a River Basin Management Plan
Table 4 shows the water flow change at the outlet of Kosynthos river for the Ythan Catchment, UK3rd International SWAT Conference, Zurich, 11–15 July,
2005.
watershed. The “100% Deforestation” Scenario affects the water flow [4] V.A. Tsihrintzis, H.R. Fuentes, R. Gadipudi, Modeling prevention alternatives for
mostly due to the fact that the largest part of the watershed (almost nonpoint source pollution at a wellfield in Florida, Water Resources Bulletin,
75%) was altered. Also a significant decrease in soluble phosphorus Journal of the American Water Resources Association (AWRA) 32 (2) (1996)
317–331.
loadings was observed. The “20% Expansion of Urban area” scenario [5] M. Di Luzio, R. Srinivasan, J.G. Arnold, A GIS-coupled hydrological model system
indicates that the small expansion of urban areas in relation to the total for the watershed assessment of agricultural nonpoint and point sources of
watershed area does not affect significantly the water flow. The pollution, Transactions in GIS 8 (1) (2004) 113–136.
[6] J.P. Wilson, H. Mitasova, D.J. Wright, Water resource applications of Geographic
agricultural land expansion is also increasing the water flow, but it Information Systems, Journal of Urban and Regional Information Systems
mainly affects the nutrient loads as they become nearly double and Association 12 (2000) 61–79.
more than double for nitrate and soluble phosphorus, respectively. [7] S. Dafis, E. Papastergiadou, K. Georgiou, D. Babalonas, T. Georgiadis, M.
Papageorgiou, T. Lazaridou, V. Tsiaoussi, Directive 92/43/EEC: The Project “Habitat”
In Table 5, the nutrient transport from Kosynthos watershed is in Greece: Netwotk Natura 2000, Life Contract B4-3200/84/756, DG XI Commission
summarized as a result of the four alternative crop scenarios. The “only of the European Communities, The Goulandris Mouseum of Natural History, Greek
cotton” and “only wheat” scenarios predict a considerable reduction of Biotope/Wetland Center, 1997.
[8] G.D. Gikas, T. Yiannakopoulou, V.A. Tsihrintzis, Modeling of non-point source
both nutrients. On the other hand, the “only corn” scenario predicts a
pollution in a Mediterranean drainage basin, Environmental Modeling and
significant increase of the transport of both nutrients, which for nitrate Assessment 11 (2006) 219–234.
reaches 124.8%. The “only tomato” scenario predictions showed an [9] G.D. Gikas, T. Yiannakopoulou, V.A. Tsihrintzis, Water quality trends in a coastal
almost equal nitrate loading, but also showed a significant increase in lagoon impacted by non-point source pollution after protective measures,
Hydrobiologia 563 (1) (2006) 385–406.
soluble phosphorus loading reaching 41.1%. All scenarios indicate [10] APHA, Standard Methods for the Examination of Water and Wastewater, 20th
significant loadings of nutrients entering Vistonis Lagoon. edition, 1998.
286 V. Pisinaras et al. / Desalination 250 (2010) 274–286

[11] V. Pisinaras, C. Petalas, A. Gemitzi, V.A. Tsihrintzis, S. Kasnis, Water quantity and lished in Spatial Accuracy Assessment: Land information Uncertainty in Natural
quality monitoring of Kosynthos River, North-Eastern Greece, Global NEST. Resources, Ann Arbor Press, Chelsea, Michigan, 1999, 323 pp.
International Journal 9 (3) (2007) 259–268. [26] G.D. Gikas, A Study of the Aquatic Ecosystem of Vistonis, Doctoral Dissertation,
[12] M. Di Luzio, G. Mitchell, N. Sammons, AVSWAT-X Short TutorialThird Conference Department of Civil Engineering, Democritus University of Thrace, Xanthi, Greece,
on Watershed Management to Meet Water Quality Standards and Emerging 2002 (in Greek).
TMDL. USA, March 5–9, 2005. [27] J.L. Monteith, Evaporation and the environment, The State and the Movement of
[13] S.L. Neitsch, J.G. Arnold, J.R. Kiniry, J.R. Williams, Soil and Water Assessment Tool, Water in Living Organisms, XIXth Symposium, Cambridge University Press,
Theoretical Documentation. Version 2005, Blackland Research Center, Agricultural Swansea, 1965, pp. 205–234.
Research Service, Texas, USA, 2005. [28] C.H.B. Priestley, R.J. Taylor, On the assessment of surface heat flux and evaporation
[14] S.L. Neitsch, J.G. Arnold, J.R. Kiniry, J.R. Williams, Soil and Water Assessment Tool using large-scale parameters, Monthly Weather Review 100 (1972) 81–92.
Input/Output File Documentation, Version 2005, Blackland Research Center, [29] H.G. Hargreaves, Z.A. Samani, Reference crop evapotranspiration from tempera-
Agricultural Research Service, Texas, USA, 2005. ture, Applied Engineering in Agriculture 1 (1985) 96–99.
[15] J.G. Arnold, J.R. Williams, A.D. Nicks, N.B. Sammons, SWRRB: A Basin Scale [30] T.W. FitzHugh, D.S. Mackay, Impacts of input parameter spatial aggregation on an
Simulation Model for Soil and Water Resources Management, Texas A&M agricultural nonpoint source pollution model, Journal of Hydrology 236 (1–2)
University Press, College Station, 1990. (2000) 35–53.
[16] W.G. Knisel, CREAMS: a field-scale model for chemicals, runoff and erosion from [31] G. Heuvelmans, B. Muys, J. Feyen, Evaluation of hydrological model parameter
agricultural management systems, Conservation Research Report No. 26, U.S. transferability for simulating the impact of land use on catchment hydrology,
Department of Agriculture, Science and Education Administration, 1980. Physics and Chemistry of the Earth 29 (2004) 739–747.
[17] R.A. Leonard, W.G. Knisel, D.A. Still, GLEAMS: groundwater loading effects [32] C. Santhi, J.G. Arnold, J.R. Williams, W.A. Dugas, R. Srinivasan, L.M. Hauck, Validation
on agricultural management systems, Transactions of ASAE 30 (5) (1987) of the SWAT model on a large river basin with point and nonpoint sources, Journal of
1403–1428. the American Water Resources Association 37 (5) (2001) 1169–1188.
[18] J.R. Williams, The EPIC model, in: V.P. Singh (Ed.), Computer Models of Watershed [33] M.E. Coffey, S.R. Workman, J.L. Taraba, A.W. Fogle, Statistical procedures for
Hydrology, Water Resources Publications, 1995, pp. 909–1000. evaluating daily and monthly hydrologic model predictions, Transactions of ASAE
[19] J.G. Arnold, P.M. Allen, G. Bernhardt, A comprehensive surface-ground water flow 47 (1) (2004) 59–68.
model, Journal of Hydrology 142 (1) (1993) 47–69. [34] T.S. Kornecki, G.J. Sabbagh, D.E. Storm, Evaluation of runoff, erosion and
[20] USDA-SCS, National Engineering Handbook, Hydrology Section 4, chap. 4–10, US phosphorus modeling system—SIMPLE, Journal of the American Water Resources
Dept. of Agriculture, Soil Conservation Service, Washington, DC, USA, 1972. Association 4 (1999) 807–820.
[21] J.R. Williams, H.D. Berndt, Sediment yield prediction based on watershed [35] W.C. Hession, V.O. Shanholtz, S. Mostaghimi, T.A. Dillaha, Uncalibrated performance
hydrology, Transactions of ASAE 20 (6) (1977) 1100–1104. of the finite element storm hydrograph model, Transactions of ASAE 1994 (3) (1994)
[22] R. Williams, A.D. Nicks, J.G. Arnold, Simulator for water resources in rural basins, 777–783.
Journal of Hydraulic Engineering, ASCE 111 (6) (1985) 970–986. [36] J.E. Nash, J.E. Sutcliffe, River flow forecasting through conceptual models. Part I. A
[23] S. Jenson, J. Domingue, Extracting topographic structure from digital elevation discussion of principles, Journal of Hydrology 10 (3) (1970) 282–290.
data for Geographic Information System analysis, Photogrammetric Engineering [37] V.A. Tsihrintzis, R. Hamid, Runoff quality prediction from small urban catchments
and Remote Sensing 54 (1988) 1593–1600. using SWMM, Hydrological Processes 12 (1998) 311–329.
[24] A.A. Romanowicz, M. Vanclooster, M. Rounsevell, I. La Junesse, Sensitivity of the [38] V.A. Tsihrintzis, C.B. Sidan, Modeling urban stormwater runoff using the Santa
SWAT model to the soil and land use data parameterisation: a case study in the Barbara Method, Water Resources Management 12 (1998) 139–166.
Thyle catchment, Belgium, Ecological Modelling 187 (2005) 27–39. [39] J.G. Arnold, R.S. Muttia, R. Srinivasan, P.P. Allen, Regional estimation of base flow and
[25] P. Defourny, G. Hecquet, T. Philippart, in: K. Lowell, A. Jaton (Eds.), Digital Terrain groundwater recharge in the Upper Mississippi river basin, Journal of Hydrology 227
Modeling: Accuracy Assessment and Hydrological Simulation Sensitivity, Pub- (2000) 21–40.

You might also like