Evaluation of streamflow simulati...
Hydrological Sciences���Journal���des Sciences Hydrologiques, 53(5) October 2008 Special issue: Advances in Ecohydrological Modelling with SWAT Open for discussion until 1 April 2009 Copyright �� 2008 IAHS Press 961 Evaluation of streamflow simulation by SWAT model for two small watersheds under snowmelt and rainfall ��TIENNE L��VESQUE1,2, FRAN��OIS ANCTIL1, ANN VAN GRIENSVEN3 & NICOLAS BEAUCHAMP1 1 D��partement de G��nie Civil, Universit�� Laval, Qu��bec, Qu��bec G1K 7P4, Canada 2 SNC-Lavalin, Division Transport et Infrastructures, Qu��bec, Qu��bec G2K 2E2, Canada email@example.com 3 UNESCO-IHE Water Education Institute, Department of Hydroinformatics and Knowledge Management, PO Box 3015, 2601 DA Delft, The Netherlands Abstract The degradation of the river water quality in Canadian rural catchments is of concern. In these catchments, the Soil Water Assessment Tool (SWAT) model can help better understand the problems related to diffuse pollution. The numerous documented applications of SWAT have been dominated by areas uniquely driven by rainfall. Given that Canadian hydroclimatic conditions differ due to the presence of a seasonal snowpack of long duration, evaluation of the hydrological performance needs to be performed prior to attempting any water quality simulations. The objective of the present work is to evaluate the hydrological behaviour of the SWAT model under snowmelt and rainfall for two small watersheds located in southeastern Canada. Different calibration schemes are evaluated including seasonal effects. One-year calibration gave satisfactory daily performances measured with Nash-Sutcliffe efficiency (NS) ranging between 61 and 83% and deviations of volume (Dv) between ���10 and 1%, while in validation, NS was 40���73% and Dv between ���20 and ���3%. The SWAT model has difficulties in reconciling both seasons. When winter and summer data are used separately to calibrate the model, the model performance is still much better for the winter season than for the summer one. However, the latter is considerably improved when only summer observations are provided for calibration. Conversely, calibration based strictly on the winter observations provides no real advantage over that based on all available data. A two-step composite calibration, which optimizes the SWAT snow accumulation and melt-related parameters on the winter data, after all other model parameters have been optimized on the summer data, provides a compromise. Key words calibration seasonal effects performance snowmelt modelling southeastern Canada SWAT water quality Evaluation de la performance hydrologique du mod��le SWAT pour deux petits bassins versants influenc��s par la fonte nivale et la pluie R��sum�� La d��gradation de la qualit�� des eaux dans les bassins versants canadiens �� vocation agricole est pr��occupante. Dans ces bassins versants, le mod��le Soil Water Assessment Tool (SWAT) serait utile pour mieux comprendre la probl��matique de pollution diffuse. Les nombreuses applications document��es de SWAT ont ��t�� limit��es �� des r��gions uniquement contr��l��es par la pluie. ��tant donn�� que les conditions hydro- climatiques canadiennes diff��rent par la pr��sence d���une couverture de neige saisonni��re de longue dur��e, une ��valuation de la performance hydrologique doit ��tre effectu��e avant de tenter des simulations de la qualit�� de l���eau. L���objectif des pr��sents travaux est d�����valuer le comportement hydrologique du mod��le SWAT sujet �� des ��v��nements de fonte nivale et de pluie pour deux petits bassins versants localis��s dans le sud-est du Canada. Diff��rentes approches de calage sont ��valu��es en incluant les effets saisonniers. Un calage sur un an a permis d���obtenir des performances journali��res satisfaisantes avec des coefficients de Nash-Sutcliffe (NS) variant entre 61 et 83% et des ��carts de volume (Dv) compris entre ���10 et 1%, alors qu���en validation NS a vari�� entre 40 et 73% et Dv entre ���20 et ���3%. Le mod��le SWAT a des difficult��s �� concilier les deux saisons. Lorsque les donn��es hivernales et estivales sont utilis��es s��par��ment pour caler le mod��le, la performance du mod��le est toujours bien meilleure en hiver qu���en ��t��. Cette derni��re est cependant consid��rablement am��lior��e lorsque seules les observations estivales sont utilis��es pour le calage. En contrepartie, un calage bas�� sur les seules observations hivernales n���apporte aucun avantage concret par rapport �� un calage avec toutes les donn��es disponibles. Un calage composite en deux temps, qui optimise les param��tres de SWAT relatifs �� l���accumulation et �� la fonte de la neige avec les donn��es hivernales, apr��s optimisation de tous les autres param��tres du mod��le avec les donn��es estivales, m��ne �� un compromis. Mots clefs calage effets saisonniers performance mod��lisation de la fonte nivale sud-est du Canada SWAT qualit�� de l���eau INTRODUCTION The water quality of Canadian streams is, at times, unsatisfactory. In an assessment, Simard (2004) notes important exceedences of summer median total phosphorus and nitrogen, mainly attributable
Etienne L��vesque et al. Copyright �� 2008 IAHS Press 962 to loads arising from agriculture. Turbidity exceedences and bacterial contamination are also noted. Schindler et al. (2006) point out that agriculture, along with urbanization and industry, are responsible for the higher concentration of nitrogen in the surface water and groundwater of southern Canada. This is likely to apply to other water quality indicators, such as total suspended solids and phosphorus. In agriculturally-dominated catchments, the Soil Water Assessment Tool (SWAT) model might be useful to simulate such water quality issues. Control strategies could then be proposed in an effort to lower concentrations of total suspended solids, nitrogen and phosphorus, in agreement with government targets. In a comprehensive literature review of the SWAT model applications, Gassman et al. (2007) show that SWAT might assist in solving water resource management and diffuse pollution issues, for a large range of scales and environmental conditions. However, the vast majority of these applications concern watersheds driven by rainfall. Few assessments have been carried out in Nordic conditions. Winter is a low-flow period characterized by the accumulation of solid precipitation, and minimum effluent dilution capacity and oxygen replenishment (Healy & Hicks, 2004). Sub- freezing air temperature persists up to snowmelt, which typically produces, over a period of about two weeks, the most abundant flood of the year. In eastern Canada, spring runoff is particularly affected by snowmelt, which delays the availability of the water and leads to a significant spring flood, especially when rainfall is superimposed on snowmelt. However, this phenomenon cannot be observed directly, and hydrological models must rely on a complex snowmelt routine to account for such events (Ferguson, 1999). Knowledge of the snow water equivalent and energy budget are thus crucial to hydrological modelling in Nordic countries. However, the availability of ruler-based measurements is usually not dense enough to embody all snow cover conditions within a watershed. Furthermore, the restricted number of available observations makes it difficult to estimate the snow energy budget. Snowmelt routines are thus prone to some uncertainties. In contrast, spring floods have been reported transporting a large part of the sediment and nutrient annual loads (Jamieson et al., 2003 Gollamudi, 2006 Quilb�� et al., 2006 Michaud et al., 2007), stressing the need for a functional snow hydrology component (Zhang et al., 2008). The objective of the present work is to evaluate the daily hydrological behaviour of the SWAT model under snowmelt and rainfall for two small watersheds located in southeastern Canada. Different calibration schemes are evaluated, including seasonal effects, specific for winter or summer. MATERIALS AND METHODS The SWAT model The SWAT model (release 2005) is a continuous physically-based distributed river basin model simulating water, sediment and pollutant yields (Gassman et al. 2007). It was developed in the early 1990s to assist water resource managers to assess impacts of land-use management on water, and diffuse pollution for large ungauged catchments with different soil types, land uses and management practices (Arnold & Fohrer, 2005). The watershed needs first to be divided into sub-basins, each containing a main channel and a specific combination of land use, soil type and management practices, which will allow the speci- fication of hydrological response units (HRU). Water balance computations are performed at this level of spatial discretization, and contributions of each HRU are then averaged out to represent water yield to the main channel. Water is then routed to the outlet of the watershed. Review of SWAT applications Pioneering work by Peterson & Hamlet (1998), evaluating the hydrological routines of the SWAT model at the daily time step, underline difficulties in baseflow and snowmelt predictions. At that time, the snowmelt routine was based on temperature index and a constant snowmelt rate factor.
Evaluation of streamflow simulation by SWAT model Copyright �� 2008 IAHS Press 963 Since then, the snowmelt routine has been refined following Fontaine et al. (2002). Major additions comprise temperature and spatial coverage evaluation of the snowpack, and inclusion of seasonal variation of the snowmelt rate. Possibilities of subdividing each sub-basin into 10 elevation bands have also been included. These modifications improved streamflow simulation performance in comparison to results obtained with the previous snowmelt routine, for a typical Rocky Mountain basin (4999 km2) (Fontaine et al. 2002). Wang & Melesse (2005) evaluated the actual SWAT snowmelt algorithm on a Minnesota (USA) watershed (4334 km2) subjected to an average annual snowfall of 146 mm. They report satisfactory monthly performances and acceptable daily performances. A detailed season-by- season evaluation shows that better spring (March���May) performance generally leads to mediocre autumn (September���November) performance and/or mediocre winter (December���February) performance. Lemonds & McCray (2007) reported successful SWAT application for the Blue River watershed (867 km2) in Summit County, Colorado, USA, when compared to average monthly observations. Snowmelt and snow formation parameters were identified as the most important calibration parameters, along with the several groundwater parameters. Zhang et al. (2008) also compared the performance of the SWAT model with different snowmelt algorithms, for runoff simulation in a macroscale mountainous river basin in the headwaters of the Yellow River, China extending over 114 345 km2. Their results showed that, in general, the SWAT model simulates monthly runoff well, and that, after calibration, the temperature index plus elevation band model provides equally good performance as the energy budget-based SNOW17 model. Under southeastern Canadian conditions, Michaud et al. (2007) obtained satisfactory calibration performance at the daily time step, at small scales (6���8 and 385���561 km2), but the model was only validated with an independent data set on a 385 km2 watershed. SWAT snowmelt algorithm The following mass balance performed at the HRU scale allows SWAT to keep track of the snowpack: mlt SNO SNOi SNOi ��� ��� + = +1 Esub Ps (1) where SNOi and SNOi+1 are the water content at day i and i+1 (mm H2O), Ps is the solid precipitation on day i (mm H2O), Esub is the sublimation on day i (mm H2O) and SNOmlt is the snowmelt on day i (mm H2O). Daily mean air temperature dictates snowfall accumulation and snowmelt. Total precipitation is classified as solid or liquid based on a threshold mean air temperature. Part of the estimated daily potential evapotranspiration (ETP) is allowed to be lost by sublimation. A temperature index is used to obtain snowmelt estimates based on the following relationship: ��� ��� ��� ��� ��� ���Tsnow ��� + = mlt mlt 2 SNOcov SNO Tmlt Tmx b (2) where bmlt is the melt factor (mm H2O day-1 ��C-1), SNOcov is the fraction of HRU area covered by snow, Tsnow is the snowpack temperature (��C), Tmx is the daily maximum air temperature (��C) and Tmlt is the base temperature above which snowmelt is allowed (��C). A lagging factor (l) sets the influence of the previous-day snowpack temperature (Tsnow,i-1) on the mean air temperature of the current day (Tair,i ) in the evaluation of the current-day snowpack temperature (Tsnow,i) through: ( l Tair,i l) Tsnow,i���1 Tsnow,i �� + ��� = 1 (3) The snowmelt factor is allowed to increase to reflect the length of the day as the season progresses. A minimum snowmelt factor (bmlt,mn) and maximum snowmelt factor (bmlt,mx) occurring at the winter and summer solstices, respectively, control these seasonal variations through: