ASSESSMENTS OF SWAT FOR WATER QUALITY MODELING IN WATERSHED CONTAINING PADDY FIELD: CASE STUDY IN UPPER KASHIMA RIVER

22nd International Congress on Irrigation and Drainage 14-20 September 2014, Gwangju Metropolitan City, Republic of KoreaR58.1.10? ?

`

SYM (1) paper no. 171Assessments of SWAT for Water Quality Modeling in Watershed CONTAINING Paddy Field: Case Study in Upper Kashima River, japanHanhan A. Sofiyuddin[footnoteRef:1], Ryota Tsuchiya[footnoteRef:2], and Tasuku Kato[footnoteRef:3] [1: Master student in Department of International Environment and Agricultural Sciences, Faculty of Agriculture, Tokyo University of Agriculture and Technology, Japan; Research and development staff in Irrigation Experimental Station, Research Center for Water Resources, Research and Development Agency, Ministry of Public Works, Indonesia. E-mail: hanhan.ahmad@gmail.com.] [2: Master student in Department of International Environment and Agricultural Sciences, Faculty of Agriculture, Tokyo University of Agriculture and Technology, Japan.] [3: Associate Professor in Department of International Environment and Agricultural Sciences, Faculty of Agriculture, Tokyo University of Agriculture and Technology, Japan.]

ABSTRACT

Soil and Water Assessment Tool (SWAT) is very promising to simulate streamflow and water quality in an agricultural watershed. Many studies have reported that it can give acceptable result to assess non-point source pollution. In Kashima River, SWAT application is expected to aid the planning process in mitigating non-point source pollution, by means of environmental conservation scenario development. The present study was conducted to investigate the applicability of SWAT in Kashima River watershed. SWAT model was configured using ARC-SWAT 2009 and calibrated using SUFI-2 method. Model evaluation was done by examining model structure and interpreting calibration results using the information gathered from the field and literatures. This study showed that SWAT can model streamflow well regardless of its inappropriate model structure to represent the hydrological process in paddy fields. However, nitrogen simulation was not satisfactory. This is mainly due to complicated processes in ponded storage, water movement into the soil, ground water flow and denitrification.

Keywords: SWAT, paddy field, streamflow, total nitrogen, evaluation, Japan.

1.Introduction

Soil and Water Assessment Tool (SWAT) is a very promising model for hydrologic and water quality simulation in an agricultural watershed. SWAT is a basin scale time-continuous model that operates on a daily time step and is designed to predict water, sediment, nutrients, pesticides dynamics, and the impact of agricultural management practices on them. The model is physically based, computationally efficient, capable of continuous simulation over long time periods and proved by many researches to give reasonable performance in assessing non-point source pollution (Gassman et al., 2007). Furthermore, SWAT is an open source model thus it is continuously tested and developed by many researchers around the world. SWAT can give valuable insight regarding approach to solve water resources and non-point sources pollution issues. For examples, SWAT was used to develop Total Maximum Daily Load (Kang et al., 2006), evaluate Best Management Practices scenario (Dechmi and Skhiri, 2013), assess the impact of land use (Volk et al., 2009), assess the impact of climate change (Ficklin et al., 2009, Somura et al., 2009), and so on.

On its application for paddy fields land use, there are some differences in the validation results among researches. Some have successfully validated SWAT in watersheds containing paddy fields (Somura et al., 2012; Luo et al., 2011). In other researches, SWAT showed some limitation that may produce significant error. Thus, modified version of SWAT model is more preferable (Kim et al., 2003; Xie and Cui, 2011; Sakaguchi et al., 2014). The difference of validation results is related to the watershed characteristic and the issues being considered. Paddy fields have distinct features of water and pollutant dynamics compared to other agricultural land use. Thus, paddy fields affect differently to runoff (Hayase, 1999) and also pollutant balances (Feng et al. 2004; Takeda et al., 1997). Water quality dynamics in paddy fields is more complex because it is influenced not only by hydrological process but also biochemical interaction (Kato et al., 2011).

In our study area, Kashima River watershed, land use is dominated by agriculture and growing paddy is one of the agricultural activities. Paddy fields are located along the river and use the river both as a source of water for irrigation as well as as a sink for drainage effluent discharge. Under such condition, paddy fields have the opportunity to enhance the water quality (Ichino and Kasuya, 1998) and hence, proper irrigation management in paddy fields area can mitigate non-point source pollution from other agricultural areas, especially total nitrogen.

SWAT application is expected to give insight into the non-point source pollution and its mitigation possibilities. As the first step towards applying SWAT to watersheds containing paddy fields, this study is aimed to understand the performance and applicability of SWAT for paddy fields hydrological process that consists of both surface and ground water processes. The parameters considered in this study were streamflow and total nitrogen. These parameters are evaluated by uncertainty analysis.

2.Method

2.1Study Area

The study was conducted for the upper part of Kashima river watershed (Figure 1) in Chiba Prefecture, Japan. Major land use in study area is agriculture, comprising upland (38.1%) and paddy field (9.2%). River conveys drainage water from upstream agricultural area that is used for paddy irrigation. The area mainly consists of 2 soil types with relatively high permeability, i.e. Humic Andosols and Gley Soil.

Figure 1. Study location (Upper Kashima River watershed)

The river flows into Inbanuma Lake, which has serious water quality problem due to mixing with river water polluted by inflows from agricultural lands and habitated areas. Monthly average value of COD (9.6-14.0 mg/l), total nitrogen (1.9-3.1 mg/l) and total phosphorous (0.11-0.14 mg/l) is more than two times greater than the maximum allowable value (Chiba Prefecture Government, 2010). Compared to other lakes in Japan, Inbanuma Lake is considered to have the worst water quality (Inbanuma Lake Water Quality Council, 2011).

2.2.SWAT model

SWAT is a semi lumped hydrological model. Spatial heterogeneity is simplified by dividing watershed into sub-watersheds. Each sub-watershed is further discretised into Hydrological Response Unit (HRU). Watershed and sub-watershed are generated based on Digital Elevation Model (DEM). HRU are generated by overlaying land use, soil and slope data. Each HRU in sub-watersheds has specific combination of land use, soil and slope category. Calculation of hydrological component is conducted in each HRU. Afterwards, the outflow of each HRU is accumulated and routed as streamflow. SWAT model framework was documented in details by Neitsch et al. (2011).

By default, HRU simulation is based on the Soil Conservation Service Curve Number (SCS CN) procedure to divide rainfall into surface runoff and infiltration (Neitsch et al., 2011). This is an empirical procedure that calculates runoff based on rainfall-runoff relationships from small rural watershed. Runoff is calculated by equation:

(1)

(2)where Qsurf is the runoff (mm), Rday is rainfall (mm), Ia is initial abstractions (mm), S is retention parameter (mm) and CN is curve number of the day, representing the overall watershed response characteristics to rainfall.

Nitrogen processes is modelled in each HRU. In surface layer, nitrogen is estimated separately in several forms which dynamically changed i.e. organic Nitrogen, NO3-N, and NH4-N. Along with water movement, those N are discharged from surface layer. In soil layers and ground water, N discharges are assumed inorganic forms i.e. NO3-N. Schematically, modelled nitrogen process is represented in Figure 2.

Figure 2. Modelled nitrogen processes in SWAT (Neitsch et al., 2009)

To model the paddy field, Neitsch et al. (2011) recommend the use of pothole module. Originally, pothole module was developed in SWAT model to accommodate physical process in depression area. Runoff generated in HRU is flowing to the lowest portion of the potholes rather than directly contributing to the flow in the river. Storage dynamics in paddy field could be better represented by using this option. Furthermore, pothole module enables the simulation of nitrogen decaying process in ponded water.

2.3.Model Parameterisation

Model input file was generated using ARC-SWAT 2009 with following data:a. Digital Elevation Model (50 m mesh) by Geographical Survey Institute, Japanb. Land use (100 m mesh) by Ministry of Land, Infrastructure, Transport and Tourism, Japanc. Soil map by Japan Soil Association with soil vertical data by Eguchi et al. (2011)d. Weather data by Japan Metrological Agency (Automated Meteorological Data Acquisition System)e. Management data (irrigation, fertilizer rate, cropping season, etc) from local authorities

Watershed divided into 13 sub-watersheds and total 188 HRUs that have different properties of slope, land use, and soil types.

2.4.Model Calibration and Evaluation

The model was calibrated using SUFI-2 method with SWAT-CUP software. This method is capable in considering all uncertainty sources, such as uncertainty in driving variables (e.g., rainfall), conceptual model, parameters, and measured data (Abbaspour, 2014).

SUFI-2 started with some initial range of parameter value. Simulation was then conducted based on parameter generated by the Latin Hypercube Sampling within the initial range. At the end of each calibration round, SUFI-2 generates new parameter range that can give better model performance. The calibration is repeated until simulation results good performance, adjudged through p-factor, r-factor and goodness of fit criterion. Details of the algorithm are available in Abbaspour et al. (2004).

p-factor is the percentage of measured data bracketed by the 95% prediction uncertainty (95PPU). When all measured data are bracketed in 95PPU band, p-factor will be 100%. The r-factor is the average thickness of the 95PPU band divided by the standard deviation of the measured data. p-factor range between 0 and 100% and r-factor ranges between 0 and infinity. A p-factor of 1 and r-factor of 0 means that simulation exactly corresponds to measured data. Calibration is considered successful if r-factor is less than 1 while maintaining high enough p-factor (more than 80% for high quality data or more than 50% for low quality data) and best simulation has satisfactory goodness of fit (Abbaspour et al., 2007).

This study uses goodness of fit criterion proposed by Moriasi et al. (2007). RSR (root mean square error to observation standard deviation ratio) is the measure of magnitude of the error that defined as root mean square error divided by standard deviation of observation data. NSE (Nash Sutcliffe model efficiency) is the measure of how well the observed and simulated data fits 1:1 line. NSE ranges between - to 1. NSE less than 0 means that average value of observed data is the better predictor than simulation result while NSE equal to 1 means that the simulated and the observed data are exactly equal. PBIAS (per cent bias) is the measure of tendency of simulated data to overestimate or underestimate the observed value. Positive value indicates model underestimation while negative value indicates model overestimation. Generally, model simulation can be judged satisfactory if RSR 0.7, NSE > 0.5 and PBIAS 25% for streamflow and PBIAS 75% for nitrogen (Moriasi et al., 2007).

The model was evaluated by examining model structure and interpreting calibration results. Using the information from field and literatures, model structure and simulation process were examined to evaluate the applicability of the model. Additionally, model diagnostic analysis was conducted by analysing model performance in different time periods and flow regimes. Time period considered in this study are irrigation and non-irrigation period. Flow regime was separated into 3 types by adapting method in Wagener et al. (1999), i.e. FD, FQ and FS. The time steps with non-zero rainfalls, lagged by the time of concentration for the catchment, were classified as rainfall driven flow (FD). The remaining time steps with streamflow lower than a certain threshold value (mean of the logarithms) were classified as non-rainfall driven slow (FS) and the rest are classified as non-rainfall driven quick (FQ). The goodness of fit in each period or flow regime was plotted in a box-plot to determine in which period the model performs poorly (Guse et al., 2013).

3.Result and Discussion

3.1Paddy Field Characteristic in Kashima River

In the study area, the river conveys polluted drainage water from agricultural fields. Paddy field that lies along the river side uses river water for irrigation and the river for drainage. To some extent, this system allows paddy field to perform water purification function. The irrigation water is retained in the ponded storage for a quite long time so that the biochemical processes can reduce its nutrient content. Thus during drainage, paddy field contributes less polluted water to the river. Measurement conducted in 2012 and 2013 showed that nitrogen content in drainage water was generally lower than concentration in the river (Figure 3).

Figure 3. Total nitrogen measurement in 2012 and 2013

Purification characteristic depends on several factors. Generally in Japan, paddy fields can contribute to purification if pollutant concentration in irrigation water is above 2 mg/L (Misawa et al., 1999). Thus, the purification is very likely since the concentration in Kashima River is generally above 5 mg/L. Another factor is the retention time in paddy field. Retention time of more than 5-7 days is preferable for the purification to occur (Takeda et al., 1996, Feng et al., 2004). However, the retention time is a function of storage and water input (irrigation or precipitation). Thus, proper irrigation management is needed.

Currently, water pumped from river or deep aquifer is used for irrigation. Based on data from local irrigation office, gross irrigation for about one third of paddy field in study area is described in Figure 4. River is a major source of irrigation that comprises around 75% of total irrigation and maximum 45% of river outflow. Thus due to its significant amount, paddy field irrigation can affect greatly to streamflow. This emphasizes the importance of water management in paddy field to maintain streamflow as well as its water quality. StreamflowIrrigation from riverIrrigation from deep aquifer

Figure 4. Monthly river outflow and irrigation

3.2SWAT Model Calibration and Evaluation

Calibration was conducted using SUFI-2 method until p-value more than 50% and r-factor less than 1. The results are in Table 1, Figure 5, and Figure 6. Streamflow can be simulated well by both methods. Both models can bracket more than 70% of the data with r-factor less than 1. The resulting best simulation performance also give quite good performance with RSR less than 0.7, NSE more than 0.5 and PBIAS less than 10%. On the contrary, TN simulation shows not satisfactory results for both methods. Table 1. Performance of the model after calibrationModelOutputp-factor (%)r-factorBest Simulation

RSRNSEPBIAS (%)

SWAT with SCS-CNQ720.600.670.54-8.7

TN45 0.84 1.03 -0.07 -2.1

SWAT with PotholeQ870.960.690.52-4.8

TN520.791.06-0.126.6

Figure 5. Calibrated model (SWAT with SCS CN)

Figure 6. Calibrated model (SWAT with pothole module)

Theoretically, there is some differences in the assumption between the modelled process in SWAT and commonly modelling approaches in paddy field (Figure 7).

Figure 7. Schematic representation of simulated process in SCS CN (left), pothole module (middle) and paddy field (right)

SCS CN method was originally developed to model upland agricultural system. Using this method, runoff is generated directly as a fraction of rainfall that does not infiltrate to the soil. Thus, this method does not consider surface storage (ponded water), which is an important component to model runoff in paddy field. Many other modelling approaches configure paddy field differently from other land use and explicitly simulated the process in ponded water situation, as in Hayase (1999) and Khepar et al. (2000). Runoff from paddy field is commonly modelled as overflow which varied with the outlet height and initial ponding depth (Kim et al., 2003).

To model ponded water, pothole module is available in SWAT. However, there is also some different assumptions that can lead to model structural error. This module is originally developed to model closed depression area in young glacial till plains (Du et al., 2005). In pothole module, storage is assumed cone shaped so surface area is not constant. This can lead to underestimation of evapotranspiration since in paddy field the storage is almost cuboids with constant surface storage (Xie and Cui, 2011). Another different assumption is in seepage into the soil profile and hydrological process during non-ponding period as described by Sakaguchi et al. (2014).

However, these differences in model structures would not have significant effect on streamflow simulation. By arranging parameters during calibration process, quite satisfactory results were obtained. Resulting calibrated model also have quite good performance.

Different from streamflow simulation, total nitrogen simulation seems more sensitive to physical representation of the model. 95PPU can bracket only less than 60% of the observed data when r-factor less than 1 and best simulation was not having a satisfactory performance. One possible cause is due to not representative simulation in low-flow period. Figure 8 shows temporal RSR of both models. Generally both models can only represent streamflow in non-irrigation period and rainfall driven flow (FD). Poorest model performance was obtained at slow non-driven flow (FS). Since total nitrogen transport is mainly by ground water flow, it is obvious that poor performance of streamflow simulation during low flow period (FQ and FS) can lead to poor performance of overall total nitrogen simulation. Another possible cause is due to simulated denitrification and nutrient leaching process. Model structure differences are resulting in non-representative simulation of water movement in the soil. Thus, simulation of those processes is impaired. Furthermore, Kato et al. (2011) suggested that improved nutrient cycling algorithm is needed to model paddy field, especially the denitrification process in them.

Figure 8. Temporal performance (RSR) for streamflow simulation of SWAT with SCS CN (left) and SWAT with pothole (right). Dashed line is RSR at 0.7 indicating threshold for satisfactory model performance

4.Conclusion

In Kashima River Watershed, paddy fields play important role in determining streamflow and water quality. Thus, proper water management in paddy fields is very important. SWAT is a promising model to work on this idea by means of environmental conservation scenario development. However, there is some inappropriate model structure in SWAT to model paddy fields that hampers the application of SWAT in this watershed.

This study showed that SWAT can represent streamflow quite well regardless of its inappropriate model structure in representing the hydrological process in paddy fields. However, nitrogen simulation was quite sensitive to physical representation of the model so the result was not satisfactory. Thus, model improvement is needed, especially in the process of surface storage, water movement into the soil, ground water flow and denitrification. Further study is required to apply and evaluate the modification ideas.

REFERENCES

Abbaspour, K.C. 2014. SWAT-CUP 2012: SWAT Calibration and Uncertainty Programs - A User Manual. Accessed at http://www.neprashtechnology.ca/Downloads/SwatCup/ Manual/Usermanual_Swat_Cup.pdf (accessed May 2014).Abbaspour, K.C., A. Johnson, M.Th. van Genuchten. 2004. Estimating uncertain ow and transport parameters using a sequential uncertainty tting procedure. Vadose Zone Journal 3: 13401352.Abbaspour, K.C., J. Yang, I. Maximov, R. Siber, K. Bogner, J. Mieleitner, J. Zobrist, R. Srinivasan. 2007. Modelling hydrology and water quality in the pre-alpine/alpine Thur Watershed using SWAT. Journal of Hydrology 333 (24): 41330.Chiba Prefecture Government. 2010. Water Quality Historical Change - Inbanuma Lake ( - ). Available at https://www.pref.chiba.lg.jp/suiho/kasentou/inbanuma/5-shinchoku/inba-suishitsu.html (accessed April 2014; in Japanese). Dechmi, F., A. Skhiri. 2013. Evaluation of best management practices under intensive irrigation using SWAT model. Agricultural Water Management 123: 5564.Du, B., , J.G. Arnold, Saleh, A., Jaynes, D.B., 2005. Development and application of SWAT to landscapes with tiles and potholes. Transaction of ASAE 48 (3): 11211133.Eguchi, S., K. Aoki, K. Kohyama. 2011. Development of agricultural soil-profile physical properties database, Japan: SolphyJ. In Proceeding ASA-CSSA-SSSA International Annual Meetings. 16-19 Oct. 2011. San Antonio, Texas.Feng, Y.W., I Yoshinaga, E. Shiratani, T. Hitomi, H. Hasebe. 2004. Characteristics and behavior of nutrients in a paddy field area equipped with a recycling irrigation system. Agricultural Water Management 68 (1): 4760. Ficklin, D.L., Y. Luo, E. Luedeling, M. Zhang. 2009. Climate change sensitivity assessment of a highly agricultural watershed using SWAT. Journal of Hydrology 374 (12): 1629.Gassman, P. W., M. R. Reyes, C. H. Green, J. G. Arnold. 2007. The Soil and Water Assesment Tool: historical development, applications, and future research directions. Transactions of the ASABE 50 (4): 1211-1250.Guse, B., D.E. Reusser, and N. Fohrer. 2014. How to improve the representation of hydrological processes in SWAT for a lowland catchment temporal analysis of parameter sensitivity and model performance. Hydrological Processes 28 (4): 265170.Hayase, Y. 1999. Runoff analysis of paddy field based watersheds. In Advanced Paddy Field Engineering. Japanese Society of Irrigation, Drainage and Reclamation Engineering, Tokyo, Japan.Ichino, K., M. Kasuya. 1998. Nitrogen removal in paddy fields: an option for water quality improvement in rivers. Ecological Engineering 10 (2): 15964. Inbanuma Lake Water Quality Council. 2011. Introduction to Inbanuma Lake (). Avalilable at http://www.insuikyo.jp/environment/frame/ (accessed April 2014; in Japanese).Kang, M.S., S.W. Park, J.J. Lee, K.H. Yoo. 2006. Applying SWAT for TMDL programs to a small watershed containing rice paddy fields. Agricultural Water Management 79 (1): 7292.Kato, T., H. Somura, H. Kuroda, H. Nakasone. 2011. Simulation of nutrients from an agricultural watershed in Japan using the SWAT model. International Agricultural Engineering Journal 20 (3): 40-49.Khepar, S. D., A. K. Yadav, S. K. Sondhi, M. Siag. 2000. Water balance model for paddy fields under intermittent irrigation practices. Irrigation Science 19 (4): 199208.Kim, C.G., H.J. Kim, C.H. Jang, S.C. Shin, N.W. Kim. 2003. SWAT application to the Yongdam and Bocheong watersheds in Korea for daily stream flowstreamflow estimation. In Second International SWAT Conference Proceeding, TWRI Technical Report 266.Luo, P., K. Takara, B. He, W. Cao, Y. Yamashiki, D. Nover. 2011. Calibration and uncertainty analysis of SWAT model in Japanese river catchment. Journal of Japan Society of Civil Engineers, Ser B1 (Hydraulic Engineering) 67 (4): I_61- I_66.Misawa, Shiuichi. 1999. Water quality in irrigation and drainage. In Advanced Paddy Field Engineering. Japanese Society of Irrigation, Drainage and Reclamation Engineering, Tokyo, Japan.Moriasi, D.N., J.G. Arnold, M.W. Van Liew, R.L. Bingner, R.D. Harmel, T.L. Veith. 2007. Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Transactions of the ASABE 50 (3): 885-900.Neitsch, S.L., J.G. Arnold, J.R. Kiniry, J.R. Williams. 2011. Soil and Water Assessment Tool Theoretical Documentation Version 2009. Texas Water Resources Institute, Texas, USA. Available at http://twri.tamu.edu/reports/2011/tr406.pdf (accessed October 2013).Sakaguchi, A., S. Eguchi, T. Kato, M. Kasuya, K. Ono, A. Miyata, N. Tase. 2014. Development and evaluation of a paddy module for improving hydrological simulation in SWAT. Agricultural Water Management 137: 11622.Somura, H., I. Takeda, J.G. Arnold, Y. Mori, J. Jeong, N. Kannan, D. Hoffman. 2012. Impact of suspended sediment and nutrient loading from land uses against water quality in the Hii River Basin, Japan. Journal of Hydrology 450451: 2535. Somura, H., J. Arnold, D. Hoffman, I. Takeda, Y. Mori, M. Di Luzio. 2009. Impact of climate change on the Hii River Basin and salinity in Lake Shinji: a case study using the SWAT model and a regression curve. Hydrological Processes 23 (13): 18871900.Takeda, I., A. Fukushima, R. Tanaka. 1997. Non-point pollutant reduction in a paddy field watershed using circular irrigation system. Water Resources 31 (11): 2685-2692.Volk, M., S. Liersch, G. Schmidt. 2009. Towards the implementation of the European water framework directive? lessons learned from water quality simulations in an agricultural watershed. Land Use Policy 26(3): 58088. Wagener, T., D.P. Boyle, M.J. Lees, H.S. Wheater, H.V. Gupta, S. Sorooshian. 1999. A framework for development and application of hydrological models. Hydrology and Earth System Science 5 (1): 1326.Xie, X., Y. Cui. 2011. Development and test of SWAT for modeling hydrological processes in irrigation districts with paddy rice. Journal of Hydrology 396 (12): 6171.

8