**4. Calibration and validation**

During the model calibration parameters are varied within an acceptable range, until a satisfactory correlation is achieved between measured and simulated data. Usually, the parameters values are changed uniformly on the catchment level. However, certain

Modelling of Surface Water Quality by Catchment Model SWAT 117

Official measurements of a flow showed that on certain days the flow was not present or it was negligible. Model does not neglect extremely low flows, as is evident from the cumulative distribution of the flow (Fig. 2). Errors in flow measurements, in the worst case

The ENS values for total flow fall into the category of satisfactory results (Moriasi et al., 2007, Henriksen et al., 2003), R2 values fall into the category of good results, RMSE into the category of very good results (Henriksen et al., 2003) and PBIAS into the category of very good and good results (Moriasi et al., 2007). The reasons for lower results of the objective functions in the validation lie in the representation of the soil, rainfall and in the river flow

may be upto 42 % and in best case upto 3 % of the total flow (Harmel et al., 2006).

**Reka Dragonja** 

1993 - 1997

(Total Flow)

**ENS** 0.61 0.61 0.39 0.69 0.55 0.57 0.45 0.42 **R2** 0.72 0.64 0.57 0.70 0.66 0.59 0.49 0.49 **RMSE** 0.13 0.82 1.21 0.74 0.35 1.06 1.98 1.50 **PBIAS** -12.79 7.04 -14.19 19.40 1.49 4.69 23.15 -3.31 Table 5. Daily time step river flow performance statistics for the rivers Dragonja and Reka

2006 - 2008

Base Flow

Calibration Validation

Total Flow (Total Flow)

2006 - 2008

1994 - 1996

Calibration Validation

Total Flow

for the calibration (2001-2005) and validation periods

data uncertainty.

**Objective function** 

**River Flow - Reka (m3 s-1)**

**01/01/01**

**0.00 0.01 0.10 1.00 10.00 100.00**

**River Flow - Reka (m3 s-1)**

**04/01/01**

**07/01/01**

**10/01/01**

**01/01/02**

**04/01/02**

**07/01/02**

**10/01/02**

**01/01/03**

**04/01/03**

**07/01/03**

**Month/Day/Year**

**1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Percentile (River Flow)**

**10/01/03**

**01/01/04**

**04/01/04**

**07/01/04**

**10/01/04**

**01/01/05**

**04/01/05**

**07/01/05**

**Measured Simulated (c)**

**Measured Simulated (a)**

**10/01/05**

**01/01/06**

**River Flow - Dragonja (m3 s-1)**

**01/01/01**

**0.00 0.01 0.10 1.00 10.00 100.00**

Fig. 2. Comparison between simulated (SWAT) and measured daily flows (m3 s-1) (a, b) and cumulative distribution (c, d) of daily river flows for the calibration period (2001-2005)

**River Flow - Dragonja (m3 s-1)**

**04/01/01**

**07/01/01**

**10/01/01**

**01/01/02**

**04/01/02**

**07/01/02**

**10/01/02**

**01/01/03**

**04/01/03**

**07/01/03**

**Month/Day/Year**

**1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Percentile (River Flow)**

**10/01/03**

**01/01/04**

**04/01/04**

**07/01/04**

**10/01/04**

**01/01/05**

**04/01/05**

**07/01/05**

**Measured Simulated (d)**

**Measured Simulated (b)**

**10/01/05**

**01/01/06**

Base Flow

parameters (Sol\_Awc, Cn2, Canmx) are exceptions, because of the spatial heterogeneity. Firstly manual calibration, parameter by parameter, should be carried out with gradual adjustments of the parameter values until a satisfactory output results (ENS and R2> 0.5) (Moriasi et al., 2007, Henriksen et al., 2003). This procedure may be time consuming for inexperienced modellers. In the process of autocalibration only the most sensitive parameters are listed that showed the greatest effect on the model outputs. For each of the parameter a limit range (max, min) has to be assigned.

Validation is performed with parameter values from the calibrated model (Table 4) and with the measured data from another time period. Due to the data scarcity, the model was validated only for the hydrological part (flow). The river Reka water quality data covers only one year of daily observations, which was only enough for the calibration. For the river Dragonja a 14 years long data series of water quality was available, but the data was scarce in the number of observations (for sediment, NO3- and TP only 92, 73, 75, 78 measurements). It should be pointed out that samples taken during monitoring represents only the current condition of the river in a certain part of the day (concentration in mg l-1), while the simulated value is a total daily transported load (kg day-1) in a river.

Calibration of the daily flow for the rivers Reka and Dragonja catchments was performed for the period from 1998 to 2005. According to the availability of data we selected different periods for the daily flow validation of the Reka (1993–1997, 2006–2008) and Dragonja (1994–1996, 2006–2008). Due to the lack of data, on sediment, NO3- and TP, we performed only the calibration for Kožbanjšček (1. 7. 2008 – 30. 6. 2009) and Dragonja (1994–2008).


Legend: 1 - forest, permanent crops, grassland, arable; 2 - subcatchment 1-2-5, subcatchment 3-4-6-7-8-9; D - default value - depends on soil type, land use and modeller set up

Table 4. Hydrological parameters, ranges and final values selected for the calibration of models (SWAT) for the rivers Reka and Dragonja catchments

#### **4.1 Hydrology calibration and validation**

Objective functions show that the simulated total flows are within the acceptable range (Table 5, Fig. 2). Correlation coefficient (R2) for a daily flow is influenced by low flows.

parameters (Sol\_Awc, Cn2, Canmx) are exceptions, because of the spatial heterogeneity. Firstly manual calibration, parameter by parameter, should be carried out with gradual adjustments of the parameter values until a satisfactory output results (ENS and R2> 0.5) (Moriasi et al., 2007, Henriksen et al., 2003). This procedure may be time consuming for inexperienced modellers. In the process of autocalibration only the most sensitive parameters are listed that showed the greatest effect on the model outputs. For each of the

Validation is performed with parameter values from the calibrated model (Table 4) and with the measured data from another time period. Due to the data scarcity, the model was validated only for the hydrological part (flow). The river Reka water quality data covers only one year of daily observations, which was only enough for the calibration. For the river Dragonja a 14 years long data series of water quality was available, but the data was scarce in the number of observations (for sediment, NO3- and TP only 92, 73, 75, 78 measurements). It should be pointed out that samples taken during monitoring represents only the current condition of the river in a certain part of the day (concentration in mg l-1), while the

Calibration of the daily flow for the rivers Reka and Dragonja catchments was performed for the period from 1998 to 2005. According to the availability of data we selected different periods for the daily flow validation of the Reka (1993–1997, 2006–2008) and Dragonja (1994–1996, 2006–2008). Due to the lack of data, on sediment, NO3- and TP, we performed only the calibration for Kožbanjšček (1. 7. 2008 – 30. 6. 2009) and Dragonja (1994–2008).

**Calibrated values Parameter Default Range Reka Dragonja** 

Legend: 1 - forest, permanent crops, grassland, arable; 2 - subcatchment 1-2-5, subcatchment 3-4-6-7-8-9;

Objective functions show that the simulated total flows are within the acceptable range (Table 5, Fig. 2). Correlation coefficient (R2) for a daily flow is influenced by low flows.

Table 4. Hydrological parameters, ranges and final values selected for the calibration of

D - default value - depends on soil type, land use and modeller set up

models (SWAT) for the rivers Reka and Dragonja catchments

**4.1 Hydrology calibration and validation** 

1 Alpha\_Bf 0.048 0–1 0.30058 0.45923 2 Canmx1 0 0–20 8, 4, 2 8, 4, 2 3 Ch\_K2 D 0–150 7.0653 3.7212 4 Ch\_N D 0–1 0.038981 0.04363 5 Cn2 D –25/+25% –8, –15 2 +14 6 Esco 0.95 0–1 0.8 0.75 7 Gw\_Delay 31 0–160 131.1 60.684 8 Gw\_Revap 0.02 0–0.2 0.19876 0.069222 9 Gwqmn 0 0–100 100 0.79193 10 Sol\_Awc D +50% no change no change 11 Surlag 4 0.01–4 0.28814 0.13984 ENS 0.61 0.57

parameter a limit range (max, min) has to be assigned.

simulated value is a total daily transported load (kg day-1) in a river.

Official measurements of a flow showed that on certain days the flow was not present or it was negligible. Model does not neglect extremely low flows, as is evident from the cumulative distribution of the flow (Fig. 2). Errors in flow measurements, in the worst case may be upto 42 % and in best case upto 3 % of the total flow (Harmel et al., 2006).

The ENS values for total flow fall into the category of satisfactory results (Moriasi et al., 2007, Henriksen et al., 2003), R2 values fall into the category of good results, RMSE into the category of very good results (Henriksen et al., 2003) and PBIAS into the category of very good and good results (Moriasi et al., 2007). The reasons for lower results of the objective functions in the validation lie in the representation of the soil, rainfall and in the river flow data uncertainty.


Table 5. Daily time step river flow performance statistics for the rivers Dragonja and Reka for the calibration (2001-2005) and validation periods

Fig. 2. Comparison between simulated (SWAT) and measured daily flows (m3 s-1) (a, b) and cumulative distribution (c, d) of daily river flows for the calibration period (2001-2005)

Modelling of Surface Water Quality by Catchment Model SWAT 119

**Parameter Default Range Calibrated values** 

Nitrate nitrogen (NO3-N)

Total phosphorus (TP)

Sediment *Reka – Kožbanjšček Dragonja* 1 SpCon 0.0001 0.0001–0.01 0.002 0,002 2 SpExp 1 1–1.5 1.3 1 3 Ch\_Erod 0 0–1 0.092 0,06 4 Ch\_Cov 0 0.05–0.6 0.1 0,1

**ENS 0.23 0.70 ENS percentile 0.83 0.73 R2 0.24 0.80 RMSE 10.35 19.81 PBIAS –0.15 –6.33** 

1 Nperco 0.2 0.01–1 1 0,2 2 Al1 0.08 0.07–0.09 0.071 0,08 3 CMN 0.0003 0.0001–0.001 - 0,0001 4 HLIFE\_NGW 0 0–200 - 0,02

**ENS 0.40 0.10 ENS percentile 0.72 0.78 R2 0.46 0.17 RMSE 79.89 5.11 PBIAS 21.24 –3.43** 

1 Pperco 10 10–17.5 15 10 2 Phoskd 175 100–200 175 200 3 Al2 0.015 0.01–0.02 0.003 0,001 4 PSP 0.4 0.01–0.7 0.22 0,04 5 ERORGP 0 0.001–5 0 0,003 6 BC4 0.35 0.01–0.7 0.1 0,1 7 RS2 0.05 0.001–0.1 0.1 0,1 8 RS5 0.05 0.001–0.9 0.08 0,001

**ENS –0.05 0.36 ENS percentile 0.95 0.85 R2 0.11 0.46 RMSE 48.17 0.18 PBIAS 3.43 49.21** 

models calibration periods (Reka 2008 - 2009; Dragonja 1994 - 2008)

5 FRT\_surface 0.2 0–1 management

9 FRT\_surface 0.2 0–1 management

5 USLE\_P 1 0–1 slope dependent slope dependent

dependent

dependent

Table 6. SWAT water quality parameters, their ranges and the final values chosen for the

management dependent

management dependent

#### **4.2 Sediment, nitrogen and phosphorus calibration**

Sediment calibration is essential for the proper P calibration, as P is preferentially transported adsorbed on the sediment particles. Parameters used for the calibration were USLE\_P, SPCON, SPEXP, CH\_EROD, CH\_COV. Simulation results for the river Reka show lower ENS = 0.23 and a good result in predicting the variability of ENSpercentile = 0.83 (Table 6). In the case of Dragonja, model achieved good results for ENS = 0.70 and ENSpercentile = 0.73. PBIAS values fall within the category of very good results as deviation is less than 15% (Moriasi et al., 2007).

Parameters with impact on the N calibration results were FRT\_SURFACE, NPERCO, AL1, CMN, HLIFE\_NGW. The river Dragonja statistic is lower (ENS = 0.10, ENspercentile = 0.78) and for the river Reka is in satisfactory range with ENS = 0.40 and ENspercentile = 0.72 (Table 6). The PBIAS results fall into the very good (Dragonja) and satisfactory (Reka) category (Moriasi et al., 2007). The lower performance of the objective functions is connected to data scarcity in the Dragonja catchment with only 73 measurements in 14 years and in river Reka with only one year of daily data. Therefore, it is difficult to say whether the model is a good predictor of nitrate nitrogen (NO3-N ) loads and dynamics. Monthly sampling rate leads to inaccurate estimates of the transported loads of nutrients in rivers (Johnes, 2007); especially NO3- (Harmel et al, 2006).

#### **4.3 Model performance indicators**

An important step before calibrating sediment and water quality parameters is to look at other model performance indicators. Three main parameters are crop growth, evapotranspiration (ET) and Soil Water Content (SWC), as all of them have a great effect on the water balance. Evapotranspiration is a primary mechanism by which water is removed from the catchment. It depends on air temperature and soil water content. The higher the temperature, the higher is potential evapotranspiration (PET) and consequently ET, if there is enough of water in the soil. A simple monthly water balance between monthly precipitation and PET showed that average monthly water balance in the Reka catchment (station Bilje) is negative between May and August (Fig. 3). In the Dragonja catchment (station Portorož) water balance is negative from April to August (growing season) (Fig. 3).

Fig. 3. Comparison of simulated and measured (Environment Agency of Republic of Slovenia - EARS) water balance (mm) for the Reka subcatchments 8 and Dragonja subcatchment 14

Sediment calibration is essential for the proper P calibration, as P is preferentially transported adsorbed on the sediment particles. Parameters used for the calibration were USLE\_P, SPCON, SPEXP, CH\_EROD, CH\_COV. Simulation results for the river Reka show lower ENS = 0.23 and a good result in predicting the variability of ENSpercentile = 0.83 (Table 6). In the case of Dragonja, model achieved good results for ENS = 0.70 and ENSpercentile = 0.73. PBIAS values fall within the category of very good results as deviation is less than 15%

Parameters with impact on the N calibration results were FRT\_SURFACE, NPERCO, AL1, CMN, HLIFE\_NGW. The river Dragonja statistic is lower (ENS = 0.10, ENspercentile = 0.78) and for the river Reka is in satisfactory range with ENS = 0.40 and ENspercentile = 0.72 (Table 6). The PBIAS results fall into the very good (Dragonja) and satisfactory (Reka) category (Moriasi et al., 2007). The lower performance of the objective functions is connected to data scarcity in the Dragonja catchment with only 73 measurements in 14 years and in river Reka with only one year of daily data. Therefore, it is difficult to say whether the model is a good predictor of nitrate nitrogen (NO3-N ) loads and dynamics. Monthly sampling rate leads to inaccurate estimates of the transported loads of nutrients in rivers (Johnes, 2007); especially

An important step before calibrating sediment and water quality parameters is to look at other model performance indicators. Three main parameters are crop growth, evapotranspiration (ET) and Soil Water Content (SWC), as all of them have a great effect on the water balance. Evapotranspiration is a primary mechanism by which water is removed from the catchment. It depends on air temperature and soil water content. The higher the temperature, the higher is potential evapotranspiration (PET) and consequently ET, if there is enough of water in the soil. A simple monthly water balance between monthly precipitation and PET showed that average monthly water balance in the Reka catchment (station Bilje) is negative between May and August (Fig. 3). In the Dragonja catchment (station Portorož) water balance is negative from April to August (growing season) (Fig. 3).

Fig. 3. Comparison of simulated and measured (Environment Agency of Republic of Slovenia - EARS) water balance (mm) for the Reka subcatchments 8 and Dragonja

**-250 -200 -150 -100 -50 0 50 100 150 200 250 300**

**Water Balance (mm)**

**1 2 3 4 5 6 7 8 9 10 11 12 Month**

**Measured EARS Portorož (1971 -2000) Simulated Dragonja (2001-2005) Sim. Max. W.B. Sim. Min. W.B.**

**4.2 Sediment, nitrogen and phosphorus calibration** 

(Moriasi et al., 2007).

NO3- (Harmel et al, 2006).

**-250 -200 -150 -100 -50 0 50 100 150 200 250 300**

subcatchment 14

**Water balance (mm)**

**4.3 Model performance indicators** 

**1 2 3 4 5 6 7 8 9 10 11 12 Month**

**Measured EARS Bilje (1971-2000) Simulated Reka (2001-2005) Sim. Max. W.B. Sim Min. W.B.**


Table 6. SWAT water quality parameters, their ranges and the final values chosen for the models calibration periods (Reka 2008 - 2009; Dragonja 1994 - 2008)

Modelling of Surface Water Quality by Catchment Model SWAT 121

The aim of this scenario was to investigate possible effects of the agri-environmental measures on the river water quality. To achieve the aim seven different scenarios were

The field erosion buffer strips scenario (EVP) is a function of how to minimize influences of diffuse pollution resulting from agricultural activities without drastic management changes. They are planted or indigenous bands of vegetation that are situated between source areas and receiving waters to reduce surface runoff velocities and to remove pollutants from surface and subsurface runoff. The effectiveness of strips is closely correlated with their slope and width (Dillaha et al., 1989). An option of 3 m wide strips was modelled on all arable (AGRC, AGRR), vineyard (VINE), orchard (ORCI, ORCE) in olive grove (OLEA)

Organic farming scenarios on 20 % of the area (EKO20) and on the 100 % area (EKO100) aim to reduce the use of mineral fertilizers and to reduce the intensity of production. Special organic rotations with green manure and composted farmyard manure were created. The lack of P was compensated with the use of triple-superphosphate that is allowed in organic production. Both organic scenarios were designed to ensure normal production for the

Steep meadows, being an agricultural landscape, should be cut regularly, but due to the steep slopes and the associated costs and risks, are abandoned and overgrown. Scenarios having steep meadows with slope inclination above 35 % (S35) and 50 % (S50) should prevent overgrowth. To verify the effects of scenarios on water quantity and nutrients transport, meadows (TRAV) of both case studies located on slopes greater than 35 % and 50 % were changed into the forest (FRSD) (Fig. 6). In the S35 scenario 18 % (Reka) and 3.6 % (Dragonja) of grassland was changed into forest, which is equivalent to 1.43 % (Reka) and 0.67 % (Dragonja) of the total catchments. In the S50 scenario only 2 % (Reka) and 0.3 % (Dragonja) of grassland was changed into forest, which is equivalent to 0.16% (Reka) and

than 35 % and 50 % for the Reka and Dragonja catchment

Fig. 6. Hydrological response units with the grassland land use (TRAV) and slopes greater

Conservation of vineyards on steep slopes has proved to be difficult because of unprofitable production. Economic reasons were followed by a trend of wine production abandonment.

applied to the study area EVP, EKO20, EKO100, S35, S50, STV35, ETA.

**5. Agri-environmental scenarios** 

0.06% (Dragonja) of the total catchments.

HRUs.

market.

Water that enters the soil may move along one of the several different pathways. It may be removed by plant uptake or evaporation; it may percolate past the bottom of the soil profile or may move laterally in the profile. However, plant uptake removes the majority of water that enters the soil profile (Neitsch et al., 2005). The soil water content will be represented correctly if crops are growing at the expected rate and soils have been correctly parameterized. Figure 4 shows the average of HRU for both catchments, with a silt clay soils, with the prevailing surface runoff and slow lateral subsurface flow. Soils exit the field capacity in the spring and return to that state in the autumn (Fig. 4). Soils in the summer are often completely dry with occasional increasing induced by storms.

Fig. 4. Comparison of simulated soil water content (mm) for the HRU No. 38 (Reka) and HRU No. 182 (Dragonja) and observed precipitation (mm) in the calibration period (2001−2005)

The plant growth component of SWAT is a simplified version of the plant growth model. Phenological plant development is based on daily accumulated heat units, leaf area development, potential biomass is based on a method developed by Monteith, a harvest index is used to calculate yield, and plant growth can be inhibited by temperature, water, N or P stress. (Neitsch et al., 2005). In the crop database a range of parameters can be changed to meet the requirements for optimal plant growth. We used default SWAT database parameters that were additionally modified (Frame, 1992). An example crop growth profile for development of leaf area index (LAI) and plant biomass (BIOM) for vineyard is presented on figure 5.

Fig. 5. Simulated vineyard biomass growth (kg ha-1) and leaf area index (m2 m-2) for the HRU No. 38 in the river Reka catchment
