The importance of the air temperature variable for the snowmelt runoff modelling using the SRM

Transcription

1 HYDROLOGICAL PROCESSES Hydrol. Process. 15, (2001) DOI: /hyp.1031 The importance of the air temperature variable for the snowmelt runoff modelling using the SRM C. Richard 1 * and D. J. Gratton 2 1 Department of Chemistry Biology, University du Québec à Trois-Rivières, 3351 boul. Des Forges, C.P. 500, Trois-Rivières, Québec, G9A 5H7, Canada 2 Department of Geography, University du Québec à Trois-Rivières, 3351 boul. Des Forges, C.P. 500, Trois-Rivières, Québec, G9A 5H7, Canada Abstract: Runoff regimes in most northern basins are controlled by the melting snow cover. A common method for evaluating runoff consists in correlating ambient air temperature and recorded hydrometric gauge values. The air temperature is the principal variable to estimate the importance of the melting of the snow cover when using a global conceptual model such as the snowmelt runoff model (SRM). The temperature, which is often only measured at one weather station, must be extrapolated to the whole basin according to some kind of lapse rate. This extrapolation often assumes that air temperature is representative for a wide region, which is often not the case. The estimation of temperature values is critical, especially for large basins where the surface processes are largely influenced by a forest cover. This project has two objectives: (1) applying a mostly high mountain SRM to the Batiscan River Basin, in the Province of Québec, an area occupied by a forest with a rolling hill topography; (2) investigate the impact of the extrapolation strategy for estimating temperature values and its importance in the runoff modelling. A statistical comparison between the different modelling attempts was performed. This allowed us to obtain a sensitivity analysis of the snow runoff modelling in relation to the extrapolation of the temperature values. Our results showed that the weather station, used to perform the runoff modelling, should be located in the most representative land cover of the study area. Otherwise, the values of a synthetic regional weather station were more reliable for the modelling. Finally, before pursuing any snowmelt modelling with the SRM, the temperature values must be evaluated based on the location of the weather station to see if they are representative of the total study area. Copyright 2001 John Wiley & Sons, Ltd. KEY WORDS SRM; air temperature; snowmelt; forest; modelling; low relief; northern basin; Batiscan river INTRODUCTION The fundamental purpose for any hydrological modelling in a drainage basin is a better understanding of water circulation as a process and as a resource (Kustas et al., 1994; Fergusen, 1999). In most northern hemisphere countries, snow melt is the principal source of water required for the economic, social development and growth of these regions (Horne and Kavvas, 1997). In Québec, like in any northern region, hydrological modelling is employed as a predictive tool to prepare for catastrophic events, like the 1996 Saguenay floods, or to analyse recurring spring events that bear important social and economic costs (Blöschl and Kirnbauer, 1991). The necessary information required to run a hydrological model is often so difficult to gather that a more simple empirical approach is often preferred to the deterministic schemes. One of the more popular drainage basin flow analysis models correlates ambient air temperature values with gauge station hydrometric measurements (Zuzel and Cox, 1975). Air temperature is presented as the principal variable to estimate snow melt in a conceptual model such as the snowmelt runoff model (SRM) (Blöschl, 1991; Brubaker et al., 1996). * Correspondence to: C. Richard, Department of Chemistry Biology, University du Québec à Trois-Rivières, 3351 boul. Des Forges, C.P. 500, Trois-Rivières, Québec, G9A 5H7, Canada. caroline richard@uqtr.uquebec.ca Received 1 May 2001 Copyright 2001 John Wiley & Sons, Ltd. Accepted 29 August 2001

2 3358 C. RICHARD AND D. J. GRATTON It is fundamental to obtain accurate and representative temperature values for the purpose of modelling. This is particularly important for basins covering large surface areas (Rango and Martinec, 1981). A strict and rigorous approach must be applied to determine the air temperature conditions over the total hydrological modelling area. Different studies have pointed out that accurate modelling results often come from the precise understanding of the atmospheric variables in play used in the modelling strategy (Leavesley, 1989). This project addressed two objectives. First, to apply the high mountain SRM to a mostly low relief and forested drainage basin and, second, to estimate the impact of the air temperature value acquisition strategy on the hydrological modelling. STUDY AREA The study area selected to analyse the performance of the SRM is the Batiscan River basin in the province of Québec, Canada, ( N, W). Draining a surface area of approximately 4325 km 2, the Batiscan River is one of many tributaries of the Saint Lawrence River. The basin relief is approximately 756 m, from 106 to 862 m above sea level (Figure 1). For the purpose of modelling, the study area was not subdivided because of the low relief environment, and the SRM was applied globally over the basin. Landsat-TM imagery 19 August 1995 Batiscan basin in Quebec La Tuque Quebec Montreal Kilometers Grande Anse Legend: Weather station under forest cover Weather station in open land Lac-aux-Sables Riviere Verte Ouest Weather station near a forest cover Hydrometric gauge in natural environment Herouxille Watershed: Area: km 2 Shawinigan Ste-Anna de la perade St-Narcisse Elevation range: 756 m. Heads: Lac Edouard (N-W), Lac Batiscan (N-E) Trois-Rivieres Aqueduc Kilometers Figure 1. The Batiscan River basin

3 SNOWMELT RUNOFF MODELLING USING THE SRM 3359 Forest cover occupies 73% the entire basin, approximately 3151 km 2, and is almost entirely public land. The southern part of the basin is covered with the mixed forest stands of the Canadian north-east and the northern part of the basin is principally boreal. The agricultural land use is limited to the extreme southern regions, covering 745 km 2 of the Batiscan basin. Finally, surface water, developed, and bare land cover approximately 429 km 2 of the study area. DATA AND METHODOLOGY The SRM used in this study utilized ambient air temperature values combined to a degree-day coefficient in order to estimate the ablation factor of the snow cover (Martinec et al., 1998). The SRM is a semi-distributed hydrological model that has been applied principally in high mountain environments, where it obtains excellent results (Fergusen, 1999). It has never been applied to a drainage basin in the Province of Québec. The daily snowmelt runoff values computed from the SRM model use the following linear equation: Q nc1 D [C Sn a n T n C 1T n S n C C Rn P n ] A k nc1 C Q n k nc1 1 where Q (m 3 s 1 ) is the average daily discharge, C S and C R are runoff coefficients expressing the losses as a ratio of runoff, a (cm C 1 day 1 ) is the degree-day factor, T ( C day 1 ) is the number of degree-days, 1T () ( C m 1 ) is the temperature lapse rate, S (%) is the snow-covered area, P is the precipitation in mm (snow or rain), A (km 2 ) is the area of the basin, k is the recession coefficient, n is the sequence of days during the discharge computation period and / is the conversion from cm km 2 day 1 to m 3 s. The principal runoff parameters for the SRM are, firstly, the runoff coefficients C (C S and C R ). Both these coefficients correspond to water taken out of the basin by means other than runoff estimated over a long time period (evaporation, evapotranspiration and infiltration). Secondly, the recession coefficient k identifies the amount of water that will immediately show as runoff. For this project, 14 different modelling attempts were performed over the Batiscan River basin for the 1996 snowmelt season (March, April, and May). Each modelling run was done using the same set of parameter values, which were previously calibrated and adjusted to the basin. Only the temperature and precipitation values were changed between each run. For this analysis, the precipitation values were estimated using the same method as employed to acquire the temperature values. Firstly, the parameters were determined with regard to the climatic and hydrological conditions of the study basin. These values are listed in Table I. T crit ( C) is the critical temperature, and L (h) is the time lag. Secondly, in order to run the SRM, we obtained information about the snow cover from satellite images. For almost two decades now, remote sensing has established itself more and more as an excellent tool to estimate the extent of the snow cover over large and inaccessible drainage basins (Singh, 1995). To estimate the change in snow cover over the Batiscan basin we relied on NOAA AVHRR imagery. These images are particularly efficient for the estimation of the snow cover over areas larger than 500 km 2 because of the 1Ð1 km 2 spatial resolution and the large 2400 km scanning width (Seidel et al., 1989). Furthermore, the Table I. The parameter values given in the modelling of the Batiscan River basin T crit L a C S C R k March 2Ð54 17Ð8 0Ð78 0Ð088 0Ð90 0Ð90 x D 1Ð0213 April 2Ð54 17Ð8 0Ð87 0Ð564 0Ð85 0Ð85 y D 0Ð0388 May 2Ð54 17Ð8 0Ð84 0Ð924 0Ð80 0Ð80

4 3360 C. RICHARD AND D. J. GRATTON acquisition of AVHRR satellite imagery offers the possibility of at least one image per day at regular time periods because of the sun synchronous orbit. Finally, the use of synthetic and individual weather stations is necessary to estimate the impact of changing temperature values in the hydrological modelling. Synthetic stations are defined as a series of temperature values calculated from two or more individual stations. Overall, nine individual and five synthetic stations are used for different runs of the SRM. These temperature stations are listed in Table II. Synthetic stations The SRM applied over a basin permits the use of only one series of meteorological variables. When more than one meteorological station was available, we created a synthetic station that used all values from available stations. Synthetic station data were estimated based on factors such as: topography, the vegetation cover of the station site, weighting each station according to a ratio of the area of influence over the total area of the drainage basin or a simple average temperature value taken from part or all of the individual stations. The synthetic station data should represent the general climate condition of the basin while reducing the potential local effect on the registered values. For our study we tested three types of synthetic station. Equal-weighted stations. The equal-weighted stations were first established by determining the most appropriate individual stations that registered the regional climatic condition prevailing over the drainage basin. The method used to identify the most representative stations follows a strategy of subdividing the total basin area along the lines of topographic and land-use characteristics or simply based on the linear distance between the individual stations and the border of the drainage basin. The chosen stations were not assigned any weights. The synthetic station was made up of the average temperature values of the chosen individual stations that contribute most significantly to the estimation of the weather conditions of the drainage basin (Figure 2a). For this analysis, two equal-weighted stations were developed to test the SRM. Firstly, a synthetic station (Table II, #2) was determined using the individual weather stations identified by SLURPAZ (Lacroix and Martz, 1998). This algorithm calculates the area of influence of each station based on the land-use classification of the area (forest cover, agricultural and bare land) and the distances between the individual stations and the drainage basin [Equation (2)]. ( ) ( ) ( ) C CÐÐÐ D x1 n D x2 n D xn n Weights N D 2 T Table II. The synthetic and individual weather stations used for the modelling of the Batiscan River basin Synthetic weather stations Individualweather stations # Station # Station Characteristics Alt. (m) 1 Weighted station from SLURPAZ 6 La Tuque Near St-Maurice Equal-weighted station from SLURPAZ 7 Grande-Anse Near St-Maurice Weighted station from Thiessen 8 Lac-aux-Sables Open land Equal-weighted station from Thiessen 9 Rivière-verte-Ouest Forest Average value station 10 Hérouxville Open land Shawinigan Near St-Maurice St-Narcisse Forest Ste-Anne-de-la-Pérade Open land Trois-Rivières-aqueduc Open land 55

5 SNOWMELT RUNOFF MODELLING USING THE SRM 3361 a) Equal-weighted station. 1. Batiscan basin with available weather stations and subdivision with Thiessen polygones. Equal-weighted station. Watershed subdivision with Thiessen polygons. A This equal-weighted station is an average using the most representative weather station for the study area determined by the Thiessen polygons. B F E I C G H D Weather stations used to estimated the equal-weighted station: Station A : La Tuque Station B : Grande-Anse Station C : Lac-aux-Sables Station D : Riviere-verte-Ouest Station E : 26% Station G : 1% Station H : 0.6% b) weighted station. 2. Weighting the weather stations compared to total study area. Weighted station. Watershed subdivision with Thiessen polygons. A 38% 26% Weighting the weather stations as a function of the physiographic characteristics, using SLURPAZ and the surface area compared to total study area, using Thiessen polygons. B 0.4% 25% D C 39% E 0.6% 1% H F G l Weather stations used to estimated the weighted station: Station A : 38% Station B : 0.4% Station C : 25% Station D : 26% Station E : 9% Station G : 1% Station H : 0.6% Figure 2. Method to define synthetic weather station where Weights N corresponds to the weight for the individual station N, x1, x2 toxn, according to the land-use classes, D n is the Euclidean distance between each digital elevation model (DEM) cell inside the basin and each individual station and this according to each land-use area; finally, T is the total number of DEM cells falling inside the basin.

6 3362 C. RICHARD AND D. J. GRATTON Overall, four individual stations identified by SLURPAZ (Table III) were used to calculate this equalweighted station [Equation (3)]: T 2 D T A C T C C T E C T G 3 4 where T 2 corresponds to the equal-weights synthetic station according to SLURPAZ, T A is the temperature value measured at the La Tuque station, T C that at Lac-aux-Sables, T E that at Hérouxville, and T G that at St-Narcisse. Secondly, the other synthetic station (Table II, #4) was obtained based on a Thiessen polygon delineation of the drainage basin using the individual stations as the centroids. The shapes of the polygons were not influenced by any physical characteristics such as topography and forest cover. However, only seven stations out of nine for which the individual area of influence overlaps the drainage basin were retained in the calculation of the synthetic station. This method allowed every station that touched the drainage basin to participate equally [Equation (4)]. T 4 D T A C T B C T C C T D C T E C T G C T H 4 7 where T 4 corresponds to the synthetic station with equal weights according to the Thiessen polygons strategy, T A is the temperature value measured at the La Tuque station, T B that at Grande-Anse, T C that at Lacaux-Sables, T D that at Rivière-verte-Ouest, T E that at Hérouxville, T G that at St-Narcisse and T H that at Ste-Anne. The weighted stations. These stations account for the importance (weights) of the individual stations over the total study area (Figure 2b). The weights are given to the selected stations identified previously for the estimation of the equal-weighted stations (Table III). The first of these synthetic weighted stations (Table II, #1) was obtained using SLURPAZ. The weights assigned to the individual stations selected by SLURPAZ were based on surface area overlapping the drainage basin and distance that separated the station with the study area [Equation (5)]: T 1 D 0Ð51T A C 0Ð39T C C 0Ð09T E C 0Ð01T G 5 4 where T 1 corresponds to the weighted synthetic station according to SLURPAZ, T A is the temperature value measured at the La Tuque station, T C that at Lac-aux-Sables, T E that at Hérouxville, and T G that at St-Narcisse. The second synthetic weighted station was determined based on Thiessen polygons. The weights for the seven individual stations selected were established using the proportion of the total drainage basin that each individual station overlaps [Equation (6)]: T 3 D 0Ð38T A C 0Ð004T B C 0Ð25T C C 0Ð26T D C 0Ð09T E C 0Ð01T G C 0Ð006T H 6 4 where T 3 corresponds to the weighted synthetic station according to the Thiessen polygons strategy, T A is the temperature value measured at the La Tuque station, T B that at Grande-Anse, T C that at Lac-aux-Sables, T D that at Rivière-verte-Ouest, T E that at Hérouxville, T G that at St-Narcisse and T H that at Ste-Anne. The regional station. The third and last synthetic station (Table II, #5) was created using the arithmetic average temperature values from all nine individual weather stations present in the vicinity of the drainage basin. This last synthetic station did not account for topography, surface cover or any particular area of influence [Equation (7)]: T 5 D T A C T B C T C C T D C T E C T F C T G C T H C T I 9 7

7 SNOWMELT RUNOFF MODELLING USING THE SRM 3363 Table III. Individual stations used to calculate the synthetic stations and their respective weights Stations identified Weights Stations identified by Weights by SLURPAZ (%) the Thiessen (%) polygons La Tuque 51 La Tuque 38 Lac-aux-Sables 39 Grande-Anse 0Ð4 Hérouxville 9 Lac-aux-Sables 25 St-Narcisse 1 Rivières-verte-Ouest 26 Hérouxville 9 St-Narcisse 1 Ste-Anne 0Ð6 where T 5 corresponds to the synthetic regional station, T A is the temperature value measured at the La Tuque station, T B that at Grande-Anse, T C that at Lac-aux-Sables, T D that at Rivière-verte-Ouest, T E that at Hérouxville, T F that at Shawinigan, T G that at St-Narcisse, T H that at Ste-Anne and T I that at Trois-Rivièresaqueduc. The individual stations Modelling runs were performed separately using the temperature record for each of the nine individual weather stations (Table II, #6 to #14) in or within the proximity of the drainage basin. RESULTS Modelling snowmelt in a forested environment using the SRM The accuracy of the modelling attempts was established using a determination coefficient (R 2 ), also known as the Nash Sutcliffe coefficient [Equation (8)] and the Dv [Equation (9)], which accounts for the seasonal estimation of the discharge: n ( Qi Qi) 0 2 R 2 D 1 i 1 n ( Qi Q ) 2 id1 8 Dv D V r V 0 r V r 9 where R 2 is a measure of the precision of the SRM results, Q i is the measured daily runoff, Qi 0 is the modelled daily runoff and Q is the average daily runoff that corresponds to the long time average of the measured daily runoff: Dv (%) is the difference between the total measured and simulated runoff, V r is the measured runoff volume and V 0 r is the simulated runoff volume. In this study the results varied between R2 D 0Ð07 and 0Ð73; five modelling attempts showed R 2 > 0Ð60 (Figure 3). The Dv varied between 7Ð21 and 69Ð29% (Figure 4). Modelling attempts over the full temporal period using each of the five synthetic stations generally show modelled runoff values that were less than the measured values (Figure 5). On the other hand, the modelling runs using each of the individual weather stations gave, in most cases, output runoff values higher than the measured values. This overestimation existed for a short period between the middle of April and the beginning

8 3364 C. RICHARD AND D. J. GRATTON R St-Narcisse Riv-verte- Ouest Equal- Weighted by Theissen Regional station Equal- Weighted by Slurpaz Ste-Anne Weighted by Theissen Shawinigan Grande- Anse La Tuque Herouxville Weighted by Slurpaz T-Raqueduc 0.07 Lac-Sables Weather stations Figure 3. Modelling precision (R 2 ) for the Batiscan River basin Dv St-Narcisse Riv-verte- Ouest Equal- Weighted by Theissen Regional station Equal- Weighted by Slurpaz Ste-Anne Weighted by Theissen Shawinigan Grande-Anse La Tuque Herouxville Weighted by Slurpaz T-R-aqueduc Lac-Sables Weather stations Figure 4. Modelling precision (Dv) for the Batiscan River basin of May, the peak snowmelt season. Only two of the nine individual stations did not overestimate the runoff values for this time period: the modelling run using the La Tuque and Lac-aux-Sables stations. Modelling snowmelt based on different series of temperature values The variations in R 2 were directly attributed to changes in the series of temperature values inserted in the 14 modelling runs. The synthetic weather stations. Firstly, the two modelling attempts using the equal-weighted synthetic stations showed R 2 > 0Ð60. The modelling run using the stations selected by SLURPAZ gave an R 2 just slightly above 0Ð60, whereas the run using the Thiessen polygon station selection showed R 2 D 0Ð66. Secondly, the two modelling efforts using weighted stations showed R 2 results that were inferior to 0Ð60. The R 2 results for SLURPAZ and the Thiessen approach were 0Ð34 and 0Ð54 respectively. Finally, the R 2 result obtained from the average regional station was 0Ð62.

9 SNOWMELT RUNOFF MODELLING USING THE SRM 3365 Q (m ³ /s) Mesured Weighted station by SLURPAZ Individual station - St-Narcisse March-96 April-96 May-96 Time Figure 5. Modelling the runoff for the Batiscan River using a synthetic and an individual weather station The individual weather stations. The nine modelling attempts using the individual weather stations gave results that were very different. Only two modelling runs gave R 2 results that were above 0Ð60. They were the modelling results provided from the Rivière-verte-Ouest (0Ð70) and the S-Narcisse (0Ð73) weather stations and they were the best modelling results obtained overall. DISCUSSION Although usually applied to high mountain environments, often in basins supporting a glacier cover, the SRM showed potential to model snowmelt from forested, low relief environments. Furthermore, examination of the modelling results clearly showed how the modelled and measured runoff was directly controlled by the temperature values. A closer look at the mean and variance of the temperature values for the time period of the study for the nine individual stations, in addition to the extrapolated values for the five synthetic stations, shows trends that do not necessarily come out in the modelling results (Table IV). Figure 6 presents temperature curves for the three most different stations: the synthetic regional station, and the La Tuque and St-Narcisse stations. The principal observation is that the temperature values, as one would expect, show a north to south gradient, as is suggested by the data from the La Tuque and Grande-Anse and Shawinigan stations. The temperature record is also influenced by the small variations in altitude. This is confirmed when comparing the Rivière-verte-Ouest station (206 m) with the St-Anne de la Pérade station (16 m). Both altitude and latitude conditions, however, are not directly reflected in the modelling results. More crucially, temperature values that were representative of the whole drainage basin were of primary importance for the SRM model. Prior to the analysis, all of the five synthetic and nine individual weather stations used could have been adequate for modelling the snow melt. In fact, the study shows that the physical characteristics of the drainage basin, such as topography and vegetation cover, and the position of the weather station had an important impact on the temperature values registered, and ultimately on the modelling accuracy. Modelling using the synthetic stations The modelling runs performed using the equal-weighted stations showed good results. These simple averages dampened the local climatic effects that existed around the individual weather stations. The temperature values registered from these synthetic stations were less spiky than those of the individual measurements.

10 3366 C. RICHARD AND D. J. GRATTON Table IV. Mean and variance of the temperature values for the nine individual and the five synthetic stations Mean Variance Synthetic weather stations Equal-weighted by SLURPAZ 2Ð58 61Ð05 Weighted by SLURPAZ 2Ð19 62Ð05 Equal-weighted by Thiessen 2Ð51 59Ð76 Weighted by Thiessen 2Ð06 61Ð06 Regional station 2Ð70 59Ð55 Weather stations in open land Lac-aux-Sables 2Ð95 72Ð97 Hérouxville 2Ð32 62Ð21 Ste-Anne 3Ð36 56Ð25 Trois-Rivières-aqueduc 3Ð72 58Ð55 Weather stations near St-Maurice River La Tuque 1Ð97 64Ð70 Grande-Anse 2Ð21 61Ð25 Shawinigan 3Ð01 60Ð64 Weather stations under forest cover Rivière-veste-Ouest 1Ð67 59Ð30 St-Narcisse 3Ð43 58Ð93 Temp Regional station La Tuque St-Narcisse March 96 April 96 May 96 Month Figure 6. Temperature value curves for the synthetic regional station, and the La Tuque and St-Narcisse stations The modelling run for the SLURPAZ weighted station, which accounted for topography, vegetation cover, and the area of influence of individual stations based on the distance to the drainage basin, gave poor results. Based on SLURPAZ weighting strategy, we expected the SLURPAZ weighted synthetic station to have performed better. A detailed analysis of the selected stations used to produce this synthetic station showed that they were not located in or close to a forested environment, but rather in open-land areas. The SLURPAZ station registered temperature values that accounted less for forested areas than other synthetic stations. The weighting strategy used by SLURPAZ, first, followed a classification of the weather stations according to surface cover and topography and, second, determined the weights according to the distance of each station

11 SNOWMELT RUNOFF MODELLING USING THE SRM 3367 from the drainage basin. In our study, SLURPAZ selected two stations, Lac-aux-Sables and Hérouxville, which were both located inside the drainage basin but in open-land areas. This selection had an important impact on the modelling results. For the same reason, the modelling results from the weighted station from the Thiessen polygons strategy also showed poor results. Again, dividing the area using the Thiessen polygons did not account for land cover. In both cases, the weighting selection strategy was based first on distance rather than on land cover. It was more evident that land cover played a principal role in the SRM results of low relief environments and that any weighting scheme that did not account for land cover first would fail. The modelling results using a simple average from all nine individual station showed good results. In the same way as the equal-weighted synthetic stations, average temperature values reduced the local temperature effects of particular stations. Most of the nine individual weather stations used in this study were, in part, situated closer to the neighbouring drainage basin, the Saint-Maurice River, or if situated in the Batiscan River drainage basin they were placed in open-land areas. Averaging these values tends to reduce the local effects. Modelling using the individual weather stations We obtained very different results using data from each of the nine individual stations. The best results came from weather stations placed inside or close to forested environments, namely the Rivière-verte-Ouest and S-Narcisse weather stations. The energy exchange between the forest canopy and the melting snow cover was crucial during the melt season, especially for conifer forests. The temperature values registered in these areas tended to correspond to the temperature conditions that occurred in the overall basin, which is covered mostly by conifer forests. The worst results came from the La Tuque, Shawinigan and Grande-Anse stations. Although in distance they were not far, these stations were all closer to the Saint-Maurice River than to the Batiscan River. Though representing similar forested environments, the weather conditions over these stations are influenced by the more incised Saint-Maurice River. In this case topography plays a more important role than distance. The modelling runs performed using the Lac-aux-Sables, the Sainte-Anne-de-la-Pérade, the Trois-Rivièresaqueduc and Hérouxville stations all show poor results. All of these stations are located in open-land areas. Furthermore, except for the Trois-Rivières-aqueduc station, the other three stations are all inside or less than 7 km from the Batiscan River drainage basin. Again, these results show that distance was a secondary variable in the selection of the temperature values to model snowmelt with the SRM. Land cover was by far more important. If we link the results obtained for R 2 and the Dv for each modelling run, some stations show potential to model the entire modelling period in addition to specific runoff events. The Dv values obtained for the Ste-Anne, St-Narcisse and Rivière-verte-Ouest stations show that, in addition to modelling the entire melt period with relative accuracy, these stations produce runoff simulations with the typical snowmelt runoff peaks. On the other hand, the Trois-Rivières aqueduc and Grande-Anse stations show the typical snowmelt peak but give poor seasonal modelling results. For the case of the synthetic stations, it is the synthetic regional stations that give the best overall seasonal modelling results and give better results for modelling particular events. The impact of the temperature values in the snowmelt modelling To reduce the effects of any one station, the use of a regional station was an ideal solution when many stations are available but none are located on the principal land covers of the study basin. The regional station damps the local effects. On the other hand, many extreme climatic events were attenuated, making the regional station inappropriate for modelling over short time periods (a few days to a couple of weeks). To model over short time periods, an individual station ideally located under the principal land cover was better suited. The

12 3368 C. RICHARD AND D. J. GRATTON Month 1 ier March 31 1 ier May April 30 1 ier 31 Lac-Sables Hérouxville Grande- Anse La Tuque St- Narcisse Shawinigan T-Raqueduc Riv-v- Ouest Ste-Anne Weather station - Precipitation Figure 7. Precipitation events in space and time over the Batiscan River basin stations showed the high-frequency changes in the weather conditions that translated into peak flow outputs from the model (Figure 5). As stated earlier, the precipitation values were estimated in the same way as the temperature values. Although they were an important component of the modelling performance, they had a significantly smaller impact, compared with temperature, on the modelling result. In fact, in the SRM, the snow cover distribution was far more important at the beginning of the snowmelt season than the amount of snow on the ground. At this time of year, any new precipitation, if solid, contributed to the thickening of the snow-cover, or, if liquid, increased the water content of the snow on the ground. The importance of the precipitation variable was found in the spatial distribution of the precipitation events across the basin. However, when the snow cover became saturated, precipitation became more important. But again, the spatial distribution

13 SNOWMELT RUNOFF MODELLING USING THE SRM 3369 of the precipitation events was crucial. A liquid precipitation event that occurs over the whole basin, if falling on snow, will increase the melting snow cover, and the runoff from this event will be added to the runoff from the melting snow; if falling on the ground, it will simply produce runoff. Consequently, the precipitation variable was much less important for modelling with the SRM than the temperature variable. Although precipitation will increase snowmelt, it was temperature that controlled the ablation factor. A cold rain had a smaller impact than a warm rain. In the case of the Batiscan River basin, the precipitation events were, in general, uniformly distributed over the study area (Figure 7). Some stations registered an isolated precipitation event, but this was generally followed or preceded by an important event that covered the whole basin. From these results, identifying the ideal position for a weather station to model snowmelt using the SRM should be based on general topography (in higher relief environments) and on land cover (in lower relief environments). In study areas similar to the Batiscan River basin, this station should be placed on the principal land-cover type of the area at an intermediate elevation (median). This is almost the case for the Rivièreverte-Ouest station in this study area. CONCLUSION This study focused on testing the SRM, principally developed for high-relief mountain environments, in a relatively large low-relief forested environment. The results showed that the SRM was capable of obtaining good results in this type of environment. This study also focused on the importance of the temperature variable. Again, our results showed that temperature had a considerable impact on the modelling results. Identifying a representative set of temperature values was crucial for this modelling attempt. Although the SRM does account for land-cover type in its modelling strategy, the location of the weather station in relation to the principal land-cover type of the study area played an important role in obtaining good modelling results. In this study, the topography and land cover present at each weather station location had a considerable influence on the temperature values, and consequently on the modelling results. ACKNOWLEDGEMENTS The NOAA AVHRR satellite images were gratefully provided to us by Laboratoire de Télédétection of the Université du Québec à Chicoutimi. We also wish to extend our gratitude to the anonymous reviewers for their constructive comments on the manuscript. REFERENCES Blöschl G The influence of uncertainty in the air temperature and albedo on snowmelt. Nordic Hydrology 22: Blöschl G, Kirnbauer R Point snowmelt models with different degrees of complexity internal processes. Journal of hydrology 129: Brubaker K, Rango A, Kustas W Incorporating radiation inputs into the snowmelt runoff model. Hydrological Process 10: Fergusen RI Snowmelt runoff models. Progress in Physical Geography 23(2): Horne FE, Kavvas ML Physics of the spatially averaged snowmelt process. Journal of Hydrology 19: Kustas WP, Rango A, Uijlenhoet R A simple energy budget algorithm for snowmelt runoff model. Water Resources Research 30(5): Lacroix M, Martz LW Using the TOPAZ digital landscape analysis model and the SLURPAZ interface to generate SLURP input files. Manual for the SLURP Hydrological Model, V International Water Management Institute: Izmir, Turkey; Appendix 2: Leavesley GH Problems of snowmelt runoff modelling for a variety of physiographic and climatic conditions. Hydrological Sciences Journal 34(6): Martinec J, Rango A, Roberts R Snowmelt Runoff Model (SRM) User s Manual. Geographica Bernensia, Department of Geography, University of Bern; 83. Rango A, Martinec J Accuracy of snowmelt runoff simulation. Nordic Hydrology 12:

14 3370 C. RICHARD AND D. J. GRATTON Seidel K, Burkart R, Baumann R, Martinec. J Snow cover monitoring by satellites and real time runoff forecasts. Proceeding IGARSS, Vancouver; Singh VP Watershed modelling. In Computer Models of Watershed Hydrology. Water Resources Publications: Highland Ranch, CO; Zuzel JF, Cox LM Relative importance of meteorological variables in snowmelt. Water Resources Research 11(1):