Journal of Hydrology (2007) 334, 64– 72 available at www.sciencedirect.com journal homepage: www.elsevier.com/locate/jhydrol Hydrological modelling of a small catchment using SWAT-2000 – Ensuring correct flow partitioning for contaminant modelling N Kannan a,* , S.M White b, F Worrall c, M.J Whelan d a Blackland Research and Extension Centre, Texas A&M University, 720, East Blackland Road, Temple, TX 76502, USA Institute of Water and Environment, Cranfield University, Silsoe, MK45 4DT Bedfordshire, UK c Department of Earth Sciences, University of Durham, DH1 3LE County Durham, UK d Safety and Environmental Assurance Centre, Unilever Colworth Laboratory, Sharnbrook, MK44 1LQ Bedfordshire, UK b Received 14 November 2005; received in revised form 23 September 2006; accepted 28 September 2006 KEYWORDS SWAT; Hydrological modelling; Colworth; Small catchment; Flow partitioning; Curve number; Crop growth The performance of the SWAT-2000 model was evaluated using stream flow at the outlet of the 142 Colworth catchment (Bedfordshire, UK) This catchment has been monitored since October 1999 The soil type consists of clay loam soil over stony calcareous clay and a rotation of wheat, oil seed rape, grass, beans and peas is grown Much of the catchment is tile drained Acceptable performance in hydrological modelling, along with correct simulation of the processes driving the water balance were essential first requirements for predicting contaminant transport Initial results from SWAT-2000 identified some necessary modifications in the model source code for correct simulation of processes driving water balance After modification of the code, hydrological simulation, crop growth and evapotranspiration (ET) patterns were realistic when compared with empirical data Acceptable model performance (based on a number of error measures) was obtained in final model runs, with reasonable runoff partitioning into overland flow, tile drainage and base flow ª 2006 Elsevier B.V All rights reserved Summary Introduction Diffuse-source pollution of the aquatic environment has received increased attention in recent years The impacts of * Corresponding author Tel.: +1 254 774 6122; fax: +1 254 774 6001 E-mail address: kannan@brc.tamus.edu (N Kannan) diffuse-source pollutants, such as pesticides, on stream ecology are of considerable interest in the context of new legislation in Europe, particularly, the Water Framework Directive (WFD: Chave, 2001) The control of such pollutants at source (e.g via efficient land management practices) is often seen as the optimal solution to potential problems However, conducting field experiments to better-understand diffuse-source pollution and design appropriate management 0022-1694/$ - see front matter ª 2006 Elsevier B.V All rights reserved doi:10.1016/j.jhydrol.2006.09.030 Hydrological modelling of a small catchment using SWAT-2000 solutions can be prohibitively expensive There is, therefore, a need for modelling tools to predict pesticide loss under varying land use, management and climate Historically, management decisions to control diffusesource pollution have often not fully considered the interactions between climate, soil and hydrology (Thorsen et al., 1996) Contaminant transfer via runoff is a complex function of rainfall timing, antecedent hydrology, slope and soil characteristics and of the properties of the contaminant under consideration (Wauchope and Leonard, 1980) Therefore, models designed to address this kind of problem require a robust description of the hydrological processes responsible for transport and of any partitioning and transformation processes operating As part of the CEFIC (European Chemical Industries Council) LRI-funded (Long-range Research Initiative) TERRACE project (TErrestrial Runoff modelling for Risk Assessment of Chemical Exposure: White et al., 2001), a number of models were reviewed in terms of their potential for predicting diffuse-source transfers The criteria considered were: Capability for application to large-scale catchments (>100 km2) Capability for interface with a Geographic Information System (GIS) A physically reasonable representation of hydrological and contaminant transport processes Input data requirements that allow the model to be applied in a wide variety of European situations A model that could be made available as part of a freely accessible package A model validated for pesticides, preferably in a European setting Three suitable models were identified for further exploration: the physically based event model ANSWERS-2000 (Beasley, 1991; Bouraoui and Dillaha, 1996, 2000); the empirically based SWATCATCH model (Brown and Hollis, 1996; Hollis and Brown, 1996; Holman et al., 2001), and SWAT (Arnold et al., 1993; Neitsch et al., 2001a) Of these, SWAT was considered to best-achieve the above criteria It represents a trade off between physical complexity and input data requirements that is believed to be achievable across Europe As land use and management are known to be key controls over diffuse source pollution, the flexibility offered by the SWAT modelling approach gives maximum potential for defining sustainable and low environmental impact farming practices As a first step, SWAT was applied to predict pesticide transfers from land to surface water for a small catchment in Bedfordshire, UK (Kannan et al., 2006a) The hydrological modelling component of the work is discussed in this paper Outlook The ultimate objective of the work described here is the simulation of pesticide transport from land to surface water This requires an accurate estimation of chemical transfer via both surface and subsurface flow Leaching of pesticide through the soil profile depends on infiltration 65 and percolation rates, which, thus, need to be well described In addition to matching predicted and observed stream flow it is, therefore, essential to partition runoff correctly into different hydrological pathways This, in turn requires a robust simulation of the processes driving water balance such as crop growth and evapotranspiration (ET) Study area and data availability The study catchment (Fig 1) is located near Sharnbrook, Bedfordshire, UK (in an area bounded by National Grid References SP 495000, SP 263000 and SP 499000, SP 263000) The total catchment area is 141.5 The predominant soil series is Hanslope, consisting of clay loam soil over stony, calcareous clay (1:25 000 outline soil map R112 TL14; http://www.silsoe.cranfield.ac.uk/nsri/services/cf/gateway/pdf/bibliography.pdf, last accessed on September 25, 2005) Most of the catchment is covered by arable fields in which a rotation of wheat, oil seed rape, grass, beans and peas is grown Many of the fields in the catchment have extensive drainage systems, mostly installed during the 1960s using clay tile drains with gravel backfill at an approximate spacing of 40 m Secondary drainage treatments include mole drainage and sub-soiling All field drains eventually discharge into the main stream, which runs through the centre of the study area The remainder of the catchment consists of woodland, grass and some concrete areas Soil horizon data with key properties such as land use group, depth of horizon, percentage of sand, silt, clay, organic carbon, bulk density, saturated hydraulic conductivity and water content at different tension values for each horizon were obtained from the National Soil Resources Institute (http://www.silsoe.cranfield.ac.uk/nsri/services/cf/ gateway/pdf/bibliography.pdf, last accessed on September 25, 2005) The Hanslope soil association provides some of the most extensive cereal growing land in Eastern England The soils are developed in chalky till on low plateaux Although, the soils of this particular soil association have slowly permeable subsurface horizons, they are seldom waterlogged The soil type in the study area has prolonged opportunities for spring cultivation, even in wet years 30-Minute interval rainfall data for the catchment were collected from September 1999 to December 2002 Daily maximum and minimum temperature values are also recorded for the catchment During the simulation period, the highest and lowest temperatures recorded were 30.6 °C and À8.9 °C, respectively Solar radiation and wind speed data were downloaded from the British Atmospheric Data Centre (BADC) web site for the nearest weather station to the study area (Bedford) The average annual wind speed during the simulation period was 4.54 m sÀ1 and the maximum value recorded was 13.04 m sÀ1 More details on wind speed estimation and the measurement device used can be found in http://badc.nerc.ac.uk/data/surface/ukmo_guide.html#5.5 (last accessed on August 30, 2006) Relative humidity values were computed from dew point temperature (from BADC) and daily maximum and minimum temperature (after Oke, 1987) The average relative humidity during the simulation period was 83% A detailed list of management operations (e.g tillage, sowing and harvesting, 66 N Kannan et al Figure Location of study area fertilizer and pesticide application rates) carried out in the catchment (with dates) was available An automatic flow recording system was installed by Agricultural Development and Advisory Service (ADAS) at the catchment outlet to measure stream flow The system continuously records flow using a Wessex flume equipped with an ultrasonic probe to record water depth and velocity in the flume The ultrasonic probe was linked to an electronic data capture system based on a Campbell Scientific CR10 data logger The data were transferred for processing on a daily basis by means of a mobile phone link Model description – SWAT SWAT (Soil and Water Assessment Tool) is a conceptual model developed to quantify the impact of land management practices in large, complex catchments (Arnold et al., 1993; Neitsch et al., 2001a) It operates with a daily time step although sub-daily rainfall can also be used (with the Green and Ampt infiltration method) SWAT incorporates simulation of weather, crop growth, evapotranspira- tion, surface runoff, percolation, return flow, erosion, nutrient transport, pesticide fate and transport, irrigation, groundwater flow, channel transmission losses, pond and reservoir storage, channel routing, field drainage, plant water use and other supporting processes Tile drainage is simulated when the soil water content exceeds field capacity in a soil layer Estimation of tile drainage is a function of the depth of drains, time required for the tile drains to bring the soil layer to field capacity and a drainage lag parameter SWAT divides sub-catchments into hydrological response units (HRUs), which are unique combinations of soil and land cover Flow is not routed between HRUs but routing is used for flow in the channel network A large number (hundreds or thousands) of HRUs can be continuously simulated using SWAT Model setup The Digital Elevation Model (DEM) of the catchment was prepared using contour data from the 1:25,000 scale topographic map of the study area Detailed land use information, Hydrological modelling of a small catchment using SWAT-2000 obtained from ADAS, was used to prepare the land use map of the catchment The soil map was prepared based on the information obtained from National Soil Resources Institute (NSRI) The Arc View-SWAT interface (AVSWAT-2000 version 1.0) was used to delineate the catchment boundary and the burning-in option was used to derive the drainage network A visual inspection of the derived drainage network and network delineated on the paper map showed good agreement The multiple HRU option available in the AVSWAT interface was used with the objective of representing each field as a separate HRU As a result, the study area was discretised into three sub-basins and 18 hydrological response units (HRUs) Methods Model performance evaluation criteria Model performance was evaluated using a range of different error measures: Percent BIAS (PBIAS), Persistence Model Efficiency (PME), Nash and Sutcliffe Efficiency (NSE), and Daily Root Mean Square (DRMS) error criteria (Table 1) The power of these model performance measures decreases from PBIAS to DRMS in the above-mentioned order (Gupta et al., 1999) 67 et al., 2002; Moriasi et al., in press) ESCO (soil evaporation compensation factor), AWC (available water capacity), GWQMN (a threshold minimum depth of water in the shallow aquifer for base flow to occur), GWREVAP (groundwater reevaporation coefficient), REVAPMN (minimum depth of water in shallow aquifer for re-evaporation to occur), Ksat (saturated hydraulic conductivity of the first soil layer) and curve number (CN) parameters were manually adjusted (one at a time) for calibration The performance evaluation of daily hydrological modelling for the combined calibration and validation periods is discussed here From the perspective of PME (65.85%) and NSE (67.87%), the model performance is acceptable with regard to the target values of the model performance evaluation criteria considered (Table 1) In addition, the DRMS estimation criterion (0.78 mm) is low which also indicates good model performance In the case of PBIAS, the value obtained (11.95) is above the optimum value of zero indicating under estimation of stream flow in general In summary, according to the performance evaluation criteria, the overall model performance is good, indicating the suitability of SWAT for hydrological modelling of this catchment Problems identified – implementation of necessary remedial measures Initial hydrological modelling Data from the period September 1, 1999 to June 29, 2001 were used as the simulation period for calibration and validation Because of their simplicity and limited data requirements, the NRCS-curve number method for rainfall-runoff modelling and the Hargreaves method for estimation of evapotranspiration were used for initial model runs In accordance with the hydrological behaviour of the study area, tile drainage and surface runoff (together) are considered as the quick response component and base flow as the slow response component of runoff Base flow is separated from the observed flow using an automated digital filter technique (Nathan and McMahon, 1990) proposed by Arnold et al (1995) The filter has three passes and pass gave acceptable base flow values for the hydrograph (Kannan, 2003) Calibration of stream flow was carried out in accordance with SWAT user manual and other published literature from SWAT users (e.g Santhi et al., 2001; Lenhart Table The overall predicted water balance generated by the initial calibrated SWAT run (expressed as percentage of rainfall) is as follows: surface runoff (9.38%); throughflow (0%); baseflow (23.53%); tile drainage (21.05%); evapotranspiration (47.98%) Separate measurements of surface runoff, base flow and tile drainage are not available for our study area Results of field-scale investigations to ascertain the relative proportion of surface runoff and base flow contributing to stream flow conducted at another site (Boxworth, UK) with similar characteristics to Colworth (Pers Commun., John Hollis, 2002) were used as a qualitative check for the SWAT predictions reported here From the breakdown of water balance components, it can be seen that the SWAT-predicted surface runoff is relatively high In addition, the percentage of rainfall lost through evapotranspiration is well below the normally expected values (Smith, 1976) This was identified as being principally due to problems associated with modelling crop growth In fact, the predicted Error measures used for analysing the performance of hydrological modelling Error measure Evaluation criterion Target values Percent Bias (PBIAS) Average tendency of simulated flows to be larger/smaller than the observed Persistence Model Efficiency (PME) Relative magnitude of residual variance (noise) to the variance of errors obtained by the use of a simple persistence model 100 >0: acceptable performance Nash and Sutcliffe Efficiency (NSE) Relative magnitude of residual variance (noise) to the variance of flows (information) 100 >0: acceptable performance Daily Root Mean Square (DRMS) estimation criterion Standard deviation of model prediction errors ffi0: acceptable performance Values obtained >0: under-estimation;