**2.2 Database development for the model build**

110 Studies on Water Management Issues

Flysch bedrock of the case areas was formed in Eocene as a product of the sea sediments and undersea landslides. Flysch consists of repeated sedimentary layers of sandstones, marl, slate and limestone, which can quickly crumble under the influence of precipitation and temperature changes. Brown eutric soils are shallow and due to silt-loam-clay texture difficult for tillage, with appropriate agro-technical measures (deep ploughing, organic fertilisers) they obtain properties for vine or olive production. In case of inappropriate agricultural activities and land management, we can witness very strong erosion

Both areas are characterized by sub-Mediterranean climate (NE Mediterranean) with southwestern winds and warm and moist air. Average annual temperature at the station Bilje (the Reka catchment), for the period 1991−2009, was 13.3 °C, with the highest and lowest monthly average in August (22.8 °C) and January (4 ºC). Average annual rainfall in the period 1992 − 2008, was 1397 mm, with peaks between September and November (max. in September 184 mm). Average annual temperature at the Portorož station (the Dragonja catchment), for the period 1991 − 2009, was 14.1 °C, with the highest and lowest monthly average in August (23.4 °C) and January (5.2 °C). Average annual rainfall in the period 1993 − 2008, was 930 mm, with peaks between September and November (max. in September 130 mm). Both catchments are characterized by fractured aquifer, where water trapped between flysch layers forms surface springs. Alluvial aquifer in the valley bottom overlays impermeable flysch. River network, of the two areas is very extensive. Rivers character is torrential and mediterranean. The river Reka hydrograph recorded (1993-2008) the highest flow rates between October and January with the average flow of 0.98 m3s-1 in November and the maximum 24.50 m3s-1 in October 1998, and extremely low in the summers. Hydrograph of the river Dragonja recorded (1993-2008) the highest flows between November and April with the average flow of 1.41 m3s-1 in January and the maximum

64.70 m3s-1 in October 1993, however in the summer the river dries up every year.

characterized by overgrowing, which results in a disordered ownership structure.

The favourable climate and terrain influences at the higher average temperature, better lighting, soil temperatures, minimal risk of frost, wind prevents diseases development. Viticulture is economically most important agricultural sector in both areas, with important share of olive and vegetable productions in the Dragonja area, and fruit production in the Goriška Brda. Terracing is typical for both areas and depends on natural conditions, steepness of the slopes (erosion), geological structure (sliding) and climatic conditions (rainfall). In Goriška Brda (Reka) is 78% of vineyards terraced while in the Slovenian Istria (Dragonja) about 18%. Vine and olive growing are the sole agricultural sectors, which can withstand the cost of the terraces installation. Terraces in the Dragonja area are

The annual average concentration of sediment in the river Reka catchment for one year of research period (July 2008 – June 2009) was 32.6 mg l-1, nitrate (NO3-) 2.7 mg l-1 and TP concentration of 0.109 mg l-1. In the river Dragonja catchment average annual concentration of sediment in the research period (August 1989 – December 2008), was 29.1 mg l-1 (107

January 2007, the highest sediment concentration measured so far, was 1362 mg l-1. The water quality with exception of sediments does not cause any serious problems in these two study areas. Data shows that sediment concentrations are well in excess of Environment

2.7 mg l-1 (87 samples) and TP concentration of 0.043 mg l-1 (92 samples). In

processes.

samples), NO3-

Agency guide level (25 mg l-1).

Before the modelling a field tour to the research areas and review of available data was carried out (Table 1). Since the available data was insufficient for modelling, we perform additional monitoring of surface water quality at the Reka tributary Kožbanjšček hydrological station Neblo, excavation of soil profiles, laboratory measurements and using established model standards (texture, albedo, organic carbon etc) and water-physical soil


Table 1. Model input data sources for the Reka and Dragonja catchments

Modelling of Surface Water Quality by Catchment Model SWAT 113

( ) <sup>1</sup>

Root Mean Square Error – RMSE (3) is determined by calculating the standard deviation of the points from their true position, summing up the measurements, and then taking the square root of the sum. RMSE is used to measure the difference between flow (q) values simulated by a model and actual measured flow (q) values. Smaller values indicate a better

Percentage bias – PBIAS (%) (4) measures the average tendency of the simulated flows (q) to be larger or smaller than their observed counter parts (Moriasi et al., 2007). The optimal value is 0, and positive values indicate a model bias toward underestimation and vice versa.

1

=

*q*

Model calibration criteria can be further based on recommended percentages of error for annual water yields suggested from the Montana Department of Environment Quality (2005) who generalised information related to model calibration criteria (Table 2) based on a

Errors (Simulated-Measured) Recommended Criteria Error in total volume 10% Error in 50% of lowest flows 10% Error in 10% of highest flows 15% Seasonal volume error (summer) 30% Seasonal volume error (autumn) 30% Seasonal volume error (winter) 30% Seasonal volume error (spring) 30% Table 2. Model calibration hydrology criteria by Montana Department of Environment

For the detection of statistical differences between the two base scenarios and alternative scenarios Student t-test statistics should be used (α = 0.025, degrees of freedom (SP = n-1)), for comparing average annual value of two dependent samples at level of significance 0.05 (5). Variable, which has approximately symmetrical frequency distribution with one modus class, is in the interval *x* ±s expected 2/3 of the variables and in *x* ± 2s approximately 95% of the variables and in *x* ± 3s almost all variables. Confidence interval (*l1,2*) (6) for Student

> / *<sup>x</sup> <sup>t</sup> s n* − μ

distribution for all sample arithmetic means ( *x* ) can be calculated (6).

⎛ ⎞ <sup>−</sup> <sup>=</sup> ⎜ ⎟⋅ ⎝ ⎠

⎛ ⎞ <sup>−</sup> = − ⎜ ⎟

( )

*measured simulated*

*i i i*

*i average i*

1( ) *n simulated measured*

( ) 100%

( )

*n measured simulated t t i n measured t i*

*n*

<sup>−</sup> ⎝ ⎠

*measured measured*

1

*it t q q RMSE*

1

*q q PBIAS*

∑

=

1

=

∑

*NS n*

=

model performance. The range is between 0 (optimal) and infinity.

Ε

number of research papers.

Quality (2005)

*n*

2

∑ (2)

2

<sup>=</sup> <sup>−</sup> <sup>=</sup> ∑ (3)

∑ (4)

= (5)

2

properties (hydraulic conductivity, water-retention properties etc) (Saxton et al., 1986; Neisch et al., 2005; Pedosphere, 2009). For certain input data an expert assessment was performed, as required measured data was not available. For the purpose of this study we used the SWAT 2005 model and Geographic Information System (GIS) 9.1 software and ArcSWAT interface. Extensions necessary for SWAT functioning in GIS environment are Spatial Analyst, Project Manager and SWAT Watershed Delineator, which enables visualisation of the results.

SWAT is capable of simulating a single catchment or a system of hydrological linked subcatchments. The model of GIS based interface ArcSWAT defines the river network, the main point of outflow from the catchment and the distribution of subcatchments and Hydrological Response Units (HRU). HRUs are basically parts of each subcatchment with a unique combination of land use, soil, slope and land management. This allows the model modelling different ET, erosion, plant growth, surface flow, water balance, etc for each subcatchment or HRU, thus increases accuracy of the simulations (Di Luzio et al., 2005). The river Reka catchment was delineated on 9 subcatchment and 291 HRUs and the river Dragonja catchment on 16 subcatchments and 602 HRUs.

#### **2.3 Model performance objective functions**

The Pearson coefficient of correlation (R2) (unit less) for n time steps (1) describes the portion of total variance in the measured data that can be explained by the model. The range is from 0.0 (poor model) to 1.0 (perfect model). A value of 0 for R2 means that none of the variance in the measured data is replicated by the model, and value 1 means that all of the variance in the measured data is replicated by the model predictions. The fact that only the spread of data is quantified is a major drawback if R2 is considered alone. A model which systematically over or under predicts all the time will still result in good values close to 1.0 even if all predictions were wrong.

$$R^2 = \left(\frac{\sum\_{i=1}^n \{\text{simulated}\_i - \text{simulated}\_{\text{average}}\} (measured\_i - \text{measured}\_{\text{average}})}{\sqrt{\sum\_{i=1}^n (\text{simulated}\_i - \text{simulated}\_{\text{average}})^2} \sqrt{\sum\_{i=1}^n \{\text{measured}\_i - \text{measured}\_{\text{average}}\}^2}}\right)^2 \tag{1}$$

The Nash-Sutcliffe simulation efficiency index (ENS) (unit less) for n time steps (2) is widely used to evaluate the performance of hydrological model. It measures how well the simulated results predict the measured data. Values for ENS range from negative infinity (poor model) to 1.0 (perfect model). A value of 0.0 means, that the model predictions are just as accurate as using the measured data average. A value greater than 0.0 means, that the model is a better predictor of the measured data than the measured data average. The ENS index is an improvement over R2 for model evaluation purposes because it is sensitive to differences in the measured and model-estimated means and variance (Nash & Sutcliffe, 1970). A major disadvantage of Nash-Sutcliffe is the fact that the differences between the measured and simulated values are calculated as squared values and this places emphasis on peak flows. As a result the impact of larger values in a time series is strongly overestimated whereas lower values are neglected. Values should be above zero to indicate minimally acceptable performance.

properties (hydraulic conductivity, water-retention properties etc) (Saxton et al., 1986; Neisch et al., 2005; Pedosphere, 2009). For certain input data an expert assessment was performed, as required measured data was not available. For the purpose of this study we used the SWAT 2005 model and Geographic Information System (GIS) 9.1 software and ArcSWAT interface. Extensions necessary for SWAT functioning in GIS environment are Spatial Analyst, Project Manager and SWAT Watershed Delineator, which enables

SWAT is capable of simulating a single catchment or a system of hydrological linked subcatchments. The model of GIS based interface ArcSWAT defines the river network, the main point of outflow from the catchment and the distribution of subcatchments and Hydrological Response Units (HRU). HRUs are basically parts of each subcatchment with a unique combination of land use, soil, slope and land management. This allows the model modelling different ET, erosion, plant growth, surface flow, water balance, etc for each subcatchment or HRU, thus increases accuracy of the simulations (Di Luzio et al., 2005). The river Reka catchment was delineated on 9 subcatchment and 291 HRUs and the river

The Pearson coefficient of correlation (R2) (unit less) for n time steps (1) describes the portion of total variance in the measured data that can be explained by the model. The range is from 0.0 (poor model) to 1.0 (perfect model). A value of 0 for R2 means that none of the variance in the measured data is replicated by the model, and value 1 means that all of the variance in the measured data is replicated by the model predictions. The fact that only the spread of data is quantified is a major drawback if R2 is considered alone. A model which systematically over or under predicts all the time will still result in good values close to 1.0

( )( )

*i average i average i*

*simulated simulated measured measured*

⎛ ⎞ − − ⎜ ⎟ <sup>=</sup>

*i average i average i i*

*simulated simulated measured mesured*

− − ⎝ ⎠

The Nash-Sutcliffe simulation efficiency index (ENS) (unit less) for n time steps (2) is widely used to evaluate the performance of hydrological model. It measures how well the simulated results predict the measured data. Values for ENS range from negative infinity (poor model) to 1.0 (perfect model). A value of 0.0 means, that the model predictions are just as accurate as using the measured data average. A value greater than 0.0 means, that the model is a better predictor of the measured data than the measured data average. The ENS index is an improvement over R2 for model evaluation purposes because it is sensitive to differences in the measured and model-estimated means and variance (Nash & Sutcliffe, 1970). A major disadvantage of Nash-Sutcliffe is the fact that the differences between the measured and simulated values are calculated as squared values and this places emphasis on peak flows. As a result the impact of larger values in a time series is strongly overestimated whereas lower values are neglected. Values should be above zero to indicate

( ) ( )

∑ ∑ (1)

1 1

= =

*n n*

2

2 2

visualisation of the results.

Dragonja catchment on 16 subcatchments and 602 HRUs.

**2.3 Model performance objective functions** 

even if all predictions were wrong.

*n*

∑

=

2 1

minimally acceptable performance.

*R*

$$E\_{NS} = 1 - \left(\frac{\sum\_{i=1}^{n} (measured\_i - simulated\_i)^2}{\sum\_{i=1}^{n} (measured\_i - measured\_{average})^2}\right) \tag{2}$$

Root Mean Square Error – RMSE (3) is determined by calculating the standard deviation of the points from their true position, summing up the measurements, and then taking the square root of the sum. RMSE is used to measure the difference between flow (q) values simulated by a model and actual measured flow (q) values. Smaller values indicate a better model performance. The range is between 0 (optimal) and infinity.

$$RMSE = \sqrt{\frac{\sum\_{i=1}^{n} (q\_t^{\text{simulated}} - q\_t^{\text{measured}})^2}{n}} \tag{3}$$

Percentage bias – PBIAS (%) (4) measures the average tendency of the simulated flows (q) to be larger or smaller than their observed counter parts (Moriasi et al., 2007). The optimal value is 0, and positive values indicate a model bias toward underestimation and vice versa.

$$PBIAS = \left(\frac{\sum\_{i=1}^{n} (q\_t^{measured} - q\_t^{simulated})}{\sum\_{i=1}^{n} (q\_t^{measured})}\right) \cdot 100\% \tag{4}$$

Model calibration criteria can be further based on recommended percentages of error for annual water yields suggested from the Montana Department of Environment Quality (2005) who generalised information related to model calibration criteria (Table 2) based on a number of research papers.


Table 2. Model calibration hydrology criteria by Montana Department of Environment Quality (2005)

For the detection of statistical differences between the two base scenarios and alternative scenarios Student t-test statistics should be used (α = 0.025, degrees of freedom (SP = n-1)), for comparing average annual value of two dependent samples at level of significance 0.05 (5). Variable, which has approximately symmetrical frequency distribution with one modus class, is in the interval *x* ±s expected 2/3 of the variables and in *x* ± 2s approximately 95% of the variables and in *x* ± 3s almost all variables. Confidence interval (*l1,2*) (6) for Student distribution for all sample arithmetic means ( *x* ) can be calculated (6).

$$t = \frac{\overline{\chi} - \mu}{s \,/\sqrt{n}}\tag{5}$$

Modelling of Surface Water Quality by Catchment Model SWAT 115

surface runoff velocity of the river and Esco describes evaporation from the soil. For the sediment modelling the most important parameters are Spcon and Spexp that affect the movement and separation of the sediment fractions in the channel. Ch\_N − Manning coefficient for channel, determines the sediment transport based on the shape of the channel and type of the river bed material. Ch\_Cov − Channel cover factor and Ch\_Erod − Channel erodibillity factor proved to be important for the Dragonja catchment. Soil erosion is closely related to the surface runoff hydrological processes (Surlag, Cn2). The analysis showed importance of the hydrological parameters that are associated with surface and subsurface runoff (Cn2, Canmx, Sol\_Awc), evaporation (Revapmin, Esco, Blai), base flow (Alpha\_Bf) and groundwater (Rchrg\_Dp, Gwqmn), suggesting numerous routes by which sediment nitrate nitrogen (NO3-N) and TP are transported (Table 3). We noticed that the amount of N is also influenced by other parameters that are not included in the sensitivity analysis tool like Rate factor for humus mineralization of organic nutrients active N and P (CMN.bsn), half-life of nitrates and the shallow aquifer (HLIFE\_NGW.gw), fraction of algal biomass that is N (Al1.wwq). TP results are significantly affected by the parameters that control surface runoff (Cn2, Canmx, Usle\_P). Usle\_P factor adjusts the USLE value for a particular land management. This means that the soil loss from the terraced land is different, from non terraced slopes. Parameters which have a significant impact on P, but not included in the sensitivity analysis tool are: fraction of algal biomass that is P (Al2.wwq), P availability index (PSP.bsn), P enrichment ratio for loading with sediment (ERORGP.hru), BC4.swq, benthic sediment source rate for dissolved P in the reach (RS2.swq), organic P settling rate

**Base Sensitivity Analysis Objective function (SSQR)** 

**model Flow Sediment NO3-N TP Category** 

AlphaBf Ch\_N Revapmin Cn2 Cn2 Surlag Alpha\_Bf AlphaBf Ch\_K2 Spexp Esco Surlag Esco Cn2 RchrgDp Ch\_K2

Ch\_N Alpha\_Bf Sol\_Awc Slope

AlphaBf Ch\_Erod Sol\_Awc AlphaBf Ch\_K2 Ch\_Cov Cn2 Blai RchrgDp Ch\_N Revapmin Surlag Esco Spexp RchrgDp Cn2

Surlag Surlag Sol\_Z Sol\_Z

Table 3. SWAT parameters ranked by the sensitivity analysis for the Reka subcatchment 5

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

Surlag Spcon Cn2 Usle\_P Very important

Cn2 Spcon Blai Canmx Very important

Important (2-6)

Important (2-6)

(RS5.swq).

Brda

Dragonja

and Dragonja subcatchment 14 (1998 - 2005)

**4. Calibration and validation** 

$$t = \overline{\pi} \pm t\_{\frac{\alpha}{2}} (n - 1) \cdot \frac{s}{\sqrt{n}} \tag{6}$$


