Evaluation of nutrient retention in vegetated filter strips using the SWAT model

Control of nutrient fluxes to surface water bodies is required to achieve good chemical and ecological conditions of water quality. In basins with intensive agricultural activities, nutrients mainly originate from diffuse (non-point) sources. Therefore, best management practices (BMPs) are increasingly implemented to reduce nutrient input to streams. The positive impacts of BMPs on water quality have been proven in practice (e.g. Chaubey et al. 1995; Schmitt et al. 1999; Hunt et al. 2006; Sweeney & Newbold 2014) but these impacts need to be evaluated on the basin scale to guide river basin management plans, before the implementation of the BMP. Therefore, water quality assessment models that work on the basin scale are useful tools to evaluate diffuse source pollution and the impacts of BMPs. The SWAT (soil and water assessment tool) model (Arnold et al. 2002a) is a versatile tool that allows the simulation of a wide variety of BMPs, such as fertilizer and manure application rate and timing, cover crops, vegetated filter strips (VFSs), no-tillage farming, irrigation management, grassed waterways, wetland creation, and terracing.

The objective of this study is to evaluate the impact of VFS application as an agricultural BMP for improving surface water quality. For this purpose, SWAT is chosen, a semi-distributed water quality assessment model that works at the watershed scale, and applied on a small-sized stream basin. The effectiveness of the VFS is estimated by first determining nitrogen and phosphorus loads from diffuse pollution sources without applying any measures and then calculating reductions of these loads when the VFS is applied to agricultural areas. The dependence of the nutrient reduction rate on different widths of the VFS is also investigated.

The Nif stream basin is located in the Aegean region of Turkey and has a drainage area of 981 km². It is an environmentally stressed sub-basin of the Gediz river basin (GRB). The flowrate of the Nif stream fluctuates significantly during the year with an average minimum of 0.445 m³/s occurring in September and an average maximum of 6.904 m³/s occurring in March. The average annual flowrate is 2.542 m³/s. The mean annual precipitation in the basin is recorded as 782 mm and the mean minimum and mean maximum temperatures on long-term record are 9.8 °C and 22.1 °C, respectively. The stream has a length of 56.7 km with an average slope of 0.49% and drains water primarily from agricultural lands and forests, and also from industrial areas (Figure 1). Agricultural lands and forests cover 51.5% and 32.7% of the total stream watershed area, respectively. Industrial and residential areas together occupy 3.4% of the total area. Soil texture in the basin comprises chromic luvisols, eutric cambisols, and calcaric fluvisols. The land slope is an important factor in the generation of surface runoff, therefore important in the occurrence of diffuse source pollution. The basin features mostly land slopes greater than 10%, except for areas near the main reach of the stream, where usually slopes less than 1% are observed. These areas cover only about 15% of the total basin area. The study area is significant since the Nif stream is one of 663 water bodies in the country that are classified as nutrient-sensitive (Tubitak-MRC 2016). As for diffuse pollution loads, it is estimated that about 28,600 people live in residential areas that have no sewerage system and therefore are connected to septic tanks for wastewater disposal. Agriculture is important for the population living in the Nif stream basin. The total area reserved for agriculture is 41,145 ha. Chemical fertilizer use is estimated as 5,837 t per year. Manure originating from livestock is also used as fertilizer. Manure production rates are approximated using livestock animal numbers; the registered number of cows and cattle is about 18,850, and for sheep and goats is 39,620. Furthermore, there are seven dump sites within the boundaries of the basin that are estimated to have a minor contribution to the overall diffuse pollution load.

SWAT is a semi-distributed, process-based hydrologic and water quality model (Arnold et al. 2012a) developed by the US Department of Agriculture (USDA). Major model components in SWAT include hydrology, weather, sedimentation, soil temperature, crop growth, nutrients, pesticides, and agricultural management. This model is widely used for river basin-scale studies dealing with water management. The ArcGIS compatible version of SWAT, ArcSWAT (version 2012.10.8), is used for this study, which is basically an interface to the SWAT model (version 2015).

Primary input data for the SWAT model comprise CORINE land cover data, a digital elevation map, a soil texture map and meteorological records. Additionally, locations and number of septic tanks, locations of dump sites and information about fertilizer consumption and manure production are required for data input. The number of septic tanks is approximated using the count of the population that is not connected to the sewer system. Consequently, center points of residential areas not connected to the sewerage system can be assumed as representative locations for an equivalent number of septic tanks. Using the given information about fertilizer consumption, it is estimated that on an elemental basis 38.2 kg/ha of nitrogen and 3.5 kg/ha of phosphorus are applied annually. Manure production is calculated from livestock animal numbers. Production unit rates per livestock animal are assumed as 0.3 kg N/t animal per day and 0.1 kg P/t animal per day for cattle and cows; 0.42 kg N/t animal per day and 0.06 kg P/t animal per day for sheep and goats. These unit rates are adopted from a study by Tanik et al. (2013). Eventually, elemental nutrient input in the Nif stream basin that originates from manure application is determined as 31.7 kg N/ha and 9.3 kg P/ha per year. Total nitrogen (TN) and phosphorus inputs from fertilizing activities are determined as 69.9 and 12.8 kg/ha per year, respectively.

For diffuse pollution modeling purposes, the Nif stream basin is divided into 19 sub-basins and 595 hydrologic response units (HRUs). Sub-basins are derived from a 30-m resolution ASTER digital elevation map (Figure 2). Furthermore, an HRU is established based on land use, soil class, and land slope. Unique combinations of these properties constitute a single HRU; therefore properties affecting flow components of water and pollutant fluxes are assumed homogeneous within an HRU. Runoff and pollutants flowing out of each HRU are summed up for every sub-basin and routed to downstream sub-basins. The meteorological variables required by SWAT are precipitation, solar radiation, relative humidity and wind speed on a daily basis, also minimum and maximum daily temperatures. Missing records are filled in using the internal SWAT weather generator, which utilizes long-term statistical data pertaining to the relevant meteorological station. Runoff flowrates are simulated using the Soil Conservation Service's curve number method (Mockus 1969). Potential evapotranspiration for each HRU is estimated using the Penman–Monteith method and then adjusted into actual evapotranspiration using antecedent and current soil water conditions. As a first step of the modeling process, average monthly flowrates of surface runoff, lateral subsurface flow, seepage from groundwater, evapotranspiration and deep aquifer recharge are calculated for all sub-basins. SWAT evaluates these hydrologic components to eventually compute discharge rates for all reaches of the stream. These discharge rates are numerically assigned to outlet points that are located at the pour points of each sub-defined sub-basin. The most downstream located outlet point represents the outlet point for the entire stream basin. Here the Nif stream merges with the Gediz river.

Simulated discharge values at the basin outlet are compared to measured monthly discharge data to evaluate the performance of the model. The SUFI-2 algorithm in the SWAT-CUP software (Abbaspour 2011) is used for sensitivity analysis and model calibration. Model parameters adjusted during the calibration were curve number (CN2), groundwater revap coefficient (GW_REVAP), threshold depth of water in the shallow aquifer required for return flow to occur (GWQMN) and available water capacity of the soil (SOL_AWC). Several statistics, including the coefficient of determination (R²), root mean square error (RMSE), percent bias (PBIAS) and Nash–Sutcliffe (NS) efficiency, are used to facilitate the calibration process and evaluate the accuracy of model flow predictions. The calibration is performed in two stages: in the first stage, model parameters are adjusted manually by trial-and-error to capture main flow patterns in the basin; in the second stage of calibration, automatic calibration with SWAT-CUP is used for detailed adjustment of parameters. Following the calibration process, the flow-calibrated model is run to simulate diffuse source nitrogen and phosphorus loads for each sub-basin. Simulated nitrogen and phosphorus loads at the basin outlet are compared with yearly-average loads that are based on monthly nutrient concentration measurements (data provided in Tubitak-MRC (2016)). The simulation time is defined for an 8-year period (1.1.2009–31.12.2016) with a spin-up period of 6 years. The calibration period spans from May 2015 to April 2016, which is defined based on the availability of field data.

VFSs are densely vegetated areas that intercept runoff from upstream diffuse pollutant sources and remove pollutants by reducing overland flow velocity, thereby increasing infiltration. This process causes deposition of suspended solids and sorption of nutrients to the soil. They are located between surface water bodies and agricultural land, grazing land or forests. VFSs are also known as vegetative filter or buffer strips (Waidler et al. 2009). The performance of VFSs has been investigated extensively through field experiments or assessments of current VFS implementations (e.g. Abu-Zreig et al. 2004; Borin et al. 2005; Zhang et al. 2010). The VFS algorithm used in the SWAT model was originally derived from White & Arnold (2009). VFSs defined in a SWAT model reduce sediments, nutrients, bacteria, and pesticides but do not affect surface runoff (Arnold et al. 2012b).

After calibration of the model with an automatic procedure satisfactory model performance statistics are obtained; comparing modeled against measured discharge at the basin outlet yields R² = 0.70; RMSE = 1.85 m³/s; PBIAS = 15.8 and NS = 0.474. The average simulated and measured flowrate for the calibration period is 3.00 m³/s and 3.56 m³/s, respectively. The simulated yearly-averaged TN load at the basin outlet is 1,360.1 kg/d, which matches the yearly-averaged load of 1,386.5 kg/d based on measured concentrations. The simulated yearly-averaged total phosphorus load at the basin outlet is 44.3 kg/d, which is comparable to the yearly-averaged load of 60.8 kg/d based on measured concentrations. Simulated flow and measured flowrates for the calibration period at the Nif stream basin outlet are shown in Figure 4. It is important to note that the simulated flowrate does not include water input from point sources in the basin; therefore point source discharge rates are added to the simulated flowrate value, assuming they are constant for the entire year. The SWAT model slightly underestimated the stream flow, as is evident from the PBIAS value and the scatter plot. However, it is also evident that the measured flowrates for the period April–September 2016 are not consistent with precipitation data for the same period. Due to point source discharges in the basin that are not monitored, the stream flow has a significant component that cannot be accounted for. Furthermore, since flowrate measurements are instantaneous and thereby reflect flow during a very short time interval, they may not be representative of the average daily flow, in particular during times when the flow in the stream is not steady. The SWAT model calculates daily average flows and therefore some measured flowrates may not necessarily fit the simulated flowrate for certain months. Nevertheless, based on the calibration statistics it can be concluded that SWAT model performance was acceptable.

Basin average annual flow components are provided in Table 1. Based on the flow component distribution, it can be concluded that 532 mm evapotranspiration is expected to have occurred during the calibration period. This amount is approximately 67% of the annual precipitation. Groundwater recharge is the second largest flow component with 256 mm. It is interesting to note that only a small fraction of precipitation (2.1%) forms surface runoff. Septic inflow originating from settlements that are not connected to any sewer system is estimated as 93 mm, which can be considered as possibly effective on the water quality in the stream.

Besides hydrologic flow components in the Nif stream basin, nutrient fluxes from diffuse sources are predicted with the SWAT model for each sub-basin. Model results for nitrogen loads are broken down into nitrate (surface runoff, lateral subsurface flow, and groundwater components) and organic nitrogen. The model reports phosphorus loads as soluble mineral form, organic phosphorus and mineral phosphorus that is sorbed to sediment particles. In this study results are reported as sums, i.e. TN and total phosphorus and only for the calibration period of May 2015 to April 2016. Figure 5 illustrates simulation results for yearly average TN and phosphorus loads. It should be noted that here VFS is not applied yet. Results indicate that the total annual nitrogen load ranges from 3.380 to 216.9 t/yr, averaging 47.85 t/yr over the entire basin. Total phosphorus loads range from 0.946 to 53.64 t/yr, with a basin average of 13.36 t/yr. The spatial distribution of loads for both nutrients is heterogeneous and appear to be correlated with the relative area of land use class reserved for agriculture. The dominant components of the TN load are organic nitrogen. Mineral nitrogen in surface runoff is relatively insignificant. The most significant component of the total phosphorus load is mineral phosphorus sorbed to sediment particles. Therefore, erosion appears to be a controlling factor of diffuse source phosphorus inputs to the river.

Sub-basin number 17 is selected for the application of VFS and its evaluation of nutrient retention performance. This sub-basin is selected since nutrient loads to the Nif stream are significant (TN = 101.95 t/yr and TP = 32.33 t/yr) and it has large agricultural areas bordering the stream. The sub-basin has a land area of 13,260 ha, of which 24.1% is covered by agricultural areas cultivating different crops. VFSs are applied on banks of the stream and only to agricultural areas located next to the stream, which cover a total area of 1,645 ha and border a stream length of 7,314 m. Nutrient loads are obtained for the filter widths of 10, 20 and 30 m (on both stream banks) and compared to the case where no VFS is applied. Simulated nutrient loads for the 1-year calibration period are shown in Figures 6 and 7 for the case with no VFS applied. The occurrence of nutrient load peaks at times of intense precipitation is evident from these time series. During dry periods, the recession of nutrient loads is clearly visible. After precipitation events, the drop in loads is imminent and rapid; however, about 1 week after the event the decrease is relatively slow. It can be also concluded that TN loads last longer and fluctuate in a narrower range during the wet period of the year. Total phosphorus loads are not continuous and cease in particular during dry periods.

Nutrient retention efficiency of VFSs is evaluated by applying filter widths of 10, 20 and 30 m on both stream banks. Nutrient inputs to the main reach in sub-basin 17 are calculated using daily time steps. The cumulative inputs for sub-basin 17 over a 1-year period and the retention percentages are summarized in Table 2. The results show that TN input to the stream can be reduced from 6.181 kg/ha (8,195.6 kg/yr) to 5.831 kg/ha (7,731.5 kg/yr) by implementing a 30-m wide VFS. This reduction corresponds to only 5.7%, suggesting a minor impact on nitrogen retention. However, the VFS is more effective for phosphorus retention; input to the stream is reduced by 9.6% for a 30-m filter width. The increase in nutrient retention with increasing filter width seems to be linked non-linearly. It should be noted that a wider VFS implies higher implementation costs; therefore, a cost–benefit analysis must be done to select the most efficient solution. Since a thorough cost–benefit analysis is out of the scope of this study, the VFS areas provided in Table 2 can be viewed as proxies for implementation cost. Therefore, as a first approach it can be noticed that if nutrient retention per VFS area is considered as a cost–benefit indicator, a 20-m filter width is the best choice for nutrient retention. TN retention per VFS area for 10-m, 20-m and 30-m VFS application are determined as 0.089, 0.143 and 0.130 (% retention/ha), respectively. Total phosphorus retention per VFS area for 10-m, 20-m and 30-m VFS application are determined as 0.226, 0.266 and 0.219 (% retention/ha), respectively.

The different components of nutrients are also determined with the SWAT model. Nutrient component yields for sub-basin 17 are provided in Figure 8. It is evident that TN is mostly composed of organic nitrogen. The remainder is nitrate, which discharges mostly in the subsurface as opposed to runoff. Total phosphorus is mostly composed of mineral and organic phosphorus attached to sediment particles. Soluble phosphorus in runoff and groundwater comprises only a very small fraction and can be considered as insignificant with respect to the impact on water quality. The higher retention performance for phosphorus can be justified with its association to sediments. The efficiency of the VFS for TN and TP depends on the reduction of the sediment load (S_R) (see Equations (2) and (4)). However, TP retention has a stronger dependency on sediment reduction.

Results of a modeling study that investigates nutrient retention performance of VFSs as agricultural BMPs are presented. The impacts of a VFS on nutrient retention are simulated using the SWAT model. It can be concluded that the model adequately simulated hydrologic flow components and nutrient transport processes in the studied stream basin. Calibration of stream flow is complicated due to unmonitored point source discharges (e.g. wastewater discharges by industrial facilities) causing sometimes major fluctuations in the stream flowrate. VFS application has limited impact on nutrient retention because of constrained application to only certain reaches of the stream. The application of VFS must be restricted to some selected sub-basin since it is practically infeasible for the entire stream basin. Therefore, rather than an application of a single type of BMP, a diversification of BMPs is required. Lastly, a 30-m wide VFS can reduce TN and TP loads by up to 5.7% and 9.6%, respectively. However, the 20-m VFS appears to be a cost-effective solution for nutrient retention, considering VFS area as an approximation for implementation cost.

Data used in this study were partially provided by the Environment and Cleaner Production Institute of Tubitak-MRC. The author would like to thank the Institute's staff for their involvement in the study.