Flood producing mechanisms identification in southern British Columbia, Canada

Flood producing mechanisms identification in southern British Columbia, Canada

Journal of Hydrology 227 (2000) 218–235 www.elsevier.com/locate/jhydrol Flood producing mechanisms identification in southern British Columbia, Canad...

390KB Sizes 0 Downloads 17 Views

Journal of Hydrology 227 (2000) 218–235 www.elsevier.com/locate/jhydrol

Flood producing mechanisms identification in southern British Columbia, Canada A. Loukas a,*, L. Vasiliades b, N.R. Dalezios c b

a Department of Civil Engineering, University of Thessaly, Pedion Areos, 383 34 Volos, Greece Department of Management of Environment and Natural Resources, University of Thessaly, Pedion Areos, 383 34 Volos, Greece c Department of Agriculture, University of Thessaly, Pedion Areos, 383 34 Volos, Greece

Received 17 May 1999; received in revised form 25 September 1999; accepted 22 October 1999

Abstract The causes of peak flows in two climatically different mountainous-forested basins of British Columbia have been identified. The U.B.C. watershed model was used to identify the causes of peak flows, since this model separately calculates the runoff components, i.e. rainfall, snowmelt and glacier runoff. The results showed that the flood flows in the maritime basin of Upper Campbell are mainly generated by rainfall during the fall months and winter rain-on-snow events. Rainfall runoff constitutes the largest percentage of peak flow for all types of events. On the other hand, the flood flows in the inland basin of Illecillewaet are mainly produced by spring rain and snowmelt events, snowmelt events alone and summer events when runoff from the glacier melt contributes to peak discharge. However, snowmelt runoff is the dominant component of peak flows. Based on these findings, flood frequency analysis showed that considering the flow component frequency distributions marginally improves the probability distribution flows in the two examined watersheds. q 2000 Elsevier Science B.V. All rights reserved. Keywords: Peak flows; Hydrologic simulation; Flood causes; Flood frequency

1. Introduction The annual series of peak flows at a station is normally treated as a sample from an implied single population. In fact, most annual flood series in Canada contain floods of two or more physical types; in particular, floods because of snowmelt and rainfall may occur at different times of year and have quite different characteristics. In many cases, it is necessary to categorize the peak flows according to the causes that produce them. Such a case is when, in flood frequency studies, the * Corresponding author. Fax: 130-42174239. E-mail addresses: [email protected] (A. Loukas), [email protected] (L. Vasiliades), [email protected] (N.R. Dalezios).

presence of floods from two or more populations produces a peculiar shape of frequency curve (Watt et al., 1989). The categorization of flood flows into various physical types should provide a better and reliable estimate of the magnitude of design floods which, in turn, is necessary for the design of hydrotechnical projects. Another reason for the classification of flood flows is the case of forest management practices. Forest management can increase the magnitude of peak flows by altering a variety of hydrologic processes. For example, removal of vegetative canopy decreases interception losses, increases net precipitation, reduces evaporation and thereby increases soil moisture and, therefore, increases the magnitude of subsequent runoff events (Harr et al., 1975; Troendle

0022-1694/00/$ - see front matter q 2000 Elsevier Science B.V. All rights reserved. PII: S0022-169 4(99)00182-1

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

219

Fig. 1. Location of the Upper Campbell and Illecillewaet watersheds and the monitoring stations within the watersheds.

and King, 1985). Roads and skid trails generate overland flow, whereas a road cut can convert subsurface flow to surface runoff (Megahan, 1983). More evident effects of logging practices are anticipated in snowdominated areas, where forest logging affects the amount and rate of snowmelt and the resulting runoff (Harr, 1981; Berris and Harr, 1987). In raindominated areas, forest harvest with minimal soil compaction should not affect the magnitude of flood flows, since the major hydrologic changes are the reduction in interception and evapotranspiration (Wright et al., 1990). The scale-dependence of the effects of forest management practices on the magnitude of peak flows has raised a lot of debate. For example, Jones and Grant (1996) analyzed long records of flows from two pairs of 60–101 ha experimental basins and three pairs of large adjacent basins ranging from 60 to 600 km 2 in western Cascades, Oregon. They concluded that forest harvesting and forest road construction have increased peak discharges by as much as 50% in the small basins and over 100% in the large basins. However, re-analysis of the same records by Thomas and Megahan (1998) showed that forest logging practices significantly increased the peak flows (up to 90%) in small basins and for the smallest peaks, whereas no detectable effects have been found for peak flows with return periods greater

than 2 years. On the other hand, their results for the large basins were inconclusive. Practically, the classification of flood flows according to flood producing mechanisms considerably complicates the analysis because a dense network of hydrometeorological stations is needed, which is not available in most areas, especially in the mountainous regions. Such a study requires the concurrent analysis of meteorological and flow data and is feasible for small basins (MacDonald and Hoffman, 1995). However, in large basins, it is difficult to make the analysis, unless a hydrologic model is used that accurately simulates the flood flows and the runoff produced by different processes based on limited meteorological data. Such a model, the U.B.C. watershed model, has been used in this study (Quick, 1995a,b). The main objectives of this study were: 1. to determine the types of the flood events in two basins located in two climatically different areas of British Columbia; 2. to evaluate the importance of the flood producing mechanisms; 3. to assess the reliability of flood frequency analyses, considering whether the flood flows are caused by different runoff mechanisms or, are derived from a single population.

220

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

2. Study areas and data base The U.B.C. watershed model has been applied to two British Columbia watersheds, the Upper Campbell and the Illecillewaet watersheds. The Upper Campbell watershed is located on the east side of the Vancouver Island mountains and drains to the north and east into the Straight of Georgia (Fig. 1). The 1194 km 2 basin is very rugged with peaks rising to 2200 m and with mean basin elevation of 950 m. At the lower elevations of the watershed a reservoir has been formed by a dam, which controls the Upper Campbell and Buttle Lakes and is about 50 km long and up to 5 km wide. The climate of the area is characterized as a maritime climate with wet and mild winters and dry and warm summers. Most of the precipitation is generated by cyclonic frontal systems that develop over the North Pacific Ocean and travel eastwards. The annual precipitation is about 2000 mm, of which about 60% is in the form of rain. The wettest period is the period between November to March, whereas the precipitation during the 6 month period from April to September accounts for only 25% of the annual precipitation. Significant snowpacks are accumulated, especially at the upper elevations, but the snowpack partly melts during the accumulation period from October to April and strongly melts from May until the disappearance of snow at the higher elevations, usually by mid-August. Although data from two meteorological stations, one at 370 m and the other at 1470 m, were available in the watershed, only daily precipitation and daily maximum and minimum temperature from the lower station have been used because the upper station was considered unreliable, especially during intense snowstorms mainly due to ice-capping of the gauge. Streamflow was measured at the mouth of the watershed. Additional data from three snowcourse sites, Lower Wolf Creek, Middle Wolf Creek and Upper Wolf Creek at elevations 640, 1070 and 1490 m, respectively, were used to compare the observed snowpack accumulation with the simulated snowpack. Fig. 1 shows the locations of the streamflow gauge, the snow courses and the hydrometeorologic stations. The second watershed examined is the Illecillewaet River basin, which is located on the west slopes of the

Selkirk Mountains in southeastern British Columbia (Fig. 1). The size of the watershed is 1150 km 2 and its elevation ranges from 1000 to 2250 m. The Illecillewaet River is a tributary of the Columbia River and contributes to the Arrow Lakes reservoir (Fig. 1). The climate of the area is continental with cold winters and warm summers with frequent hot days. The basin is located about 500 km inland from the Coast Mountains and its climate is influenced primarily by the maritime Pacific Ocean air masses and by weather systems moving eastwards. The long term average precipitation ranges from 950 mm at Revelstoke at an elevation of 443 m, which is located close to the mouth of the watershed, to 2160 mm at Glacier Mount Fidelity station at 1875 m (above mean sea-level (a.m.s.l.)). The water equivalent of the average annual snowfall for the same stations is 445 and 1518 mm, respectively. As a result, substantial snowpacks develop during winter at all elevations in the basin. The snowpack in the valley bottom at Revelstoke is usually depleted by the end of April, but permanent snowfields and a 76 km 2 glacier exist at the highest elevations. In this basin, data from three meteorological stations have been used. Precipitation and temperature data are used from the Revelstoke station at 443 m, Glacier Rogers Pass station at 1323 m and Glacier Mount Fidelity station at 1875 m. Also, the simulated snowpack accumulation has been compared to observed data from three snowcourses. The snowcourses used are Glacier at 1250 m, Mount Fidelity at 1875 m and Mount Abbott at 1980 m (a.m.s.l.). Streamflow data from the station located at the mouth of the watershed are used to assess the simulated runoff from the watershed. Fig. 1 shows the location of these stations.

3. Hydrologic model The U.B.C. watershed model has been used for the simulation of flow and identification of the peak flow cause in the two watersheds. The U.B.C. watershed model was first developed 30 years ago, and has been continuously updated to its present form (Loukas and Quick, 1996). This model has been applied to a variety of climatic regions, ranging from coastal to

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

221

Fig. 2. Conceptualization of U.B.C. watershed model.

inland mountain regions of British Columbia including the Rocky Mountains, and the subarctic region of Canada. The model has also been applied to the Himalayas and Karakoram ranges in India and Pakistan, the Southern Alps in New Zealand and the

Snowy Mountains in Australia. This ensures that the model is capable of simulating runoff under a large variety of conditions. The model conceptualizes the watersheds in a number of elevation bands. Each elevation band can

222

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

Table 1 Calibration statistics of the U.B.C. model Hydrologic year

Upper Campbell 1983–1984 1984–1985 1985–1986 1986–1987 1987–1988 1988–1989 1989–1990 Illecillewaet 1970–1971 1971–1972 1972–1973 1973–1974 1974–1975 1975–1976 1976–1977 1977–1978 1978–1979 1979–1980 1980–1981 1981–1982 1982–1983 1983–1984 1984–1985 1985–1986 1986–1987 1987–1988 1988–1989 1989–1990

Nash–Sutcliffe efficiency (Eff.)

Coefficient of determination (r 2)

Runoff volume difference (dV, %)

Mean observed flow (m 3/s)

Mean simulated flow (m 3/s)

0.69 0.80 0.83 0.73 0.63 0.68 0.82

0.70 0.81 0.83 0.74 0.63 0.70 0.83

0.13 29.66 0.09 0.37 26.10 20.72 29.89

75.5 59.7 75.5 89.6 70.2 63.6 65.5

75.6 53.9 75.5 89.9 66.0 63.1 59.0

0.94 0.95 0.94 0.95 0.93 0.93 0.94 0.95 0.95 0.95 0.90 0.92 0.93 0.93 0.94 0.93 0.92 0.91 0.93 0.93

0.95 0.95 0.94 0.95 0.94 0.93 0.94 0.96 0.96 0.95 0.90 0.93 0.93 0.94 0.94 0.93 0.92 0.91 0.94 0.94

23.88 0.15 20.01 25.83 3.94 0.40 5.82 20.32 6.30 20.08 20.97 20.17 20.30 21.58 0.24 24.83 2.22 0.32 20.52 22.90

52.0 63.1 43.0 59.6 47.5 65.5 44.7 50.3 49.9 50.2 55.5 57.0 52.3 52.9 49.4 54.3 51.9 49.4 47.3 55.3

49.9 63.2 43.0 56.2 49.3 65.8 47.3 50.1 53.1 50.1 54.9 56.9 52.1 52.1 49.6 51.6 53.0 49.5 47.0 53.7

have variable characteristics and land uses such as forested, open, impermeable and glacierised areas. Fig. 2 shows the flow chart of the model structure. Precipitation and temperature data are used as input to the model. The model is used to distribute these meteorological data with elevation. Based on temperature, the model identifies the type of precipitation, i.e. rain or snow and calculates the snowpack accumulation as a function of elevation. Snowmelt is calculated using a simplified energy balance method. Equations using only maximum and minimum temperature data have been developed and used for the estimation of cloud cover, albedo and wind speed and from these parameters the net short-wave radiation, long-wave radiation, convective and advective heat transfer and rain melt are estimated (Quick, 1995a).

Furthermore, the U.B.C. model computes the runoff separately from rainfall, snowmelt and glacier melt for each elevation band, respectively, and distributes the runoff into four runoff components with a soil moisture control mechanism (Fig. 2). The four runoff components, namely direct or surface runoff, medium or interflow runoff, slow or upper zone groundwater runoff and very slow or deep zone groundwater runoff, are simulated by using the linear storage technique. Using this technique, the runoff from each land use portion of each elevation band is calculated separately and is added to the runoff from the other land use areas of the elevation band to produce the runoff from the band. This procedure is repeated for each elevation band and the summation of the runoff from all bands provides the watershed runoff for the time step, which

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

can vary from 1 h to 1 day. Details of the model structure are presented in Quick (1995b).

4. Application-method of analysis A watershed model should be able to reproduce reasonably well the historical streamflow record in order to use it for the classification of peak flows. In this study, the U.B.C. watershed model was calibrated using daily-recorded streamflow for the Upper Campbell basin for the years—1983–1990, and for the Illecillewaet watershed for the years—1970– 1990. The statistics used to validate the performance of the model are: the mean observed and the mean simulated flow, the runoff volume difference, the coefficient of determination (r 2), and the NashSutcliffe efficiency (Eff) defined as: n X

…Qobsi 2 Qsimi †2

Eff ˆ 1 2 iˆ1 n X

…1† …Qobsi 2 Qobs †

2

iˆ1

where, Qobsi is the observed flow on day i, Qsimi is the simulated flow on day i, and Qobs is the average observed flow for the simulation period. Table 1 shows the statistics for the calibration periods of the model for the two study watersheds. The data include dry and wet years. The calibration results for the Upper Campbell watershed (Table 1) show that the Nash–Sutcliffe efficiency ranges from 0.63 to 0.83, the coefficient of determination ranges from 0.63 to 0.83 and the runoff volume difference ranges from 29.89 to 0.37%. For the Illecillewaet watershed, the Nash–Sutcliffe efficiency ranges from 0.90 to 0.95, the coefficient of determination ranges from 0.90 to 0.96 and the runoff volume difference ranges from 25.83 to 6.30% (Table 1). Runoff statistics and the comparison of simulated and observed hydrographs for wet and dry years indicate that the model simulates the recorded streamflow well. The models’ ability to simulate the snow accumulation has been tested by comparing the simulated snow accumulation with data from snowcourses. The comparison is made by using the simulated snow accumulation from one or two band elevations. It should be mentioned that the simulated values are

223

average values over the area of the elevation band or bands, whereas the observed values represent the accumulation at the location of the snowcourse. Owing to paper length limitations only the comparisons with the snowcourses at the higher elevations are shown. At high elevations deep snowpacks are observed and their melting affects the runoff distribution and flood generation in the two examined watersheds. For the Upper Campbell watershed the simulated snow accumulation of band 6, with mean elevation of 1485 m, has been compared with data from the Upper Wolf Creek snowcourse, which is at an elevation of 1490 m (a.m.s.l.). Fig. 3 shows the comparison for the years 1984–1985 (Fig. 3a) and 1986–1987 (Fig. 3b), which are the driest and wettest recorded years, respectively. This comparison shows that the U.B.C. model simulates reasonably well, the accumulation of the snowpack and the subsequent melting of the snowpack. The same comparison has been made for the Illecillewaet watershed using the average simulated snow accumulation at elevation bands 6 and 7, with mean elevations of 1915 and 2085 m, respectively, and snow data from the snowcourse of Mount Abbot, which is at 1980 m a.m.s.l. Fig. 4 presents the comparison for years 1972–1973 (Fig. 4a) and 1975–1976 (Fig. 4b), which are the driest and wettest years of the 20 years of record, respectively. The comparison indicates that the U.B.C. model estimates reasonably well, the snow accumulation at the two studied watersheds. It should be mentioned that there are no data for the melting period and thus no comparison between simulation and observations can be made. However, the comparison of the simulated and observed snowpacks along with the results of the flow calibration indicate that the U.B.C. model can be used for the simulation of flow at the two examined watersheds. After the calibration of the U.B.C. watershed model, the annual observed flows are plotted along with the simulated flows (Fig. 5a) and the component flows (Fig. 5b), i.e. snowmelt, rainfall and glacier runoff for each year of the two examined watersheds, respectively. Visual inspection of the U.B.C. model’s output files and their graphical presentation have been used to identify the causative runoff of certain peak

224

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

Fig. 3. Comparison between the simulated snow accumulation and the observed data from Upper Wolf Creek snowcourse–Upper Campbell watershed.

flows. The peak flows have been classified into five categories:

5. summer runoff events with contribution of glacier melt.

1. 2. 3. 4.

The last category of events applies only to Illecillewaet watershed, where a 76 km 2 glacier exists at high elevations. In this study, two analyses of annual flood series

rainfall events; snowmelt events; winter rain-on-snow events; spring rain and snowmelt events;

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

225

Fig. 4. Comparison between the simulated snow accumulation and the observed data from Mount Abbott snowcourse–Illecillewaet watershed.

and partial series have been implemented. For the partial series analysis, the flow threshold is taken equal to the triple long-term mean daily streamflow (Gellens, 1991) according to the positive “runs theory” (Dracup et al., 1980a, b). Based on this screening procedure any flow above the abovedefined threshold value is considered. The threshold values for the two examined watersheds are 214.8 m 3/s for the Upper Campbell watershed and 157.8 m 3/s for

the Illecillewaet basin. If two peaks have occurred within four days time interval, only the larger peak flow is considered. For each peak flow the contribution of each of the runoff components, i.e. rainfall, snowmelt and glacier melt runoff, are calculated as a percentage of the simulated peak flow. Finally, a frequency analysis of the flood flows is performed to test the reliability of the hypothesis that the flood flow consists of a sample from a single population.

226

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

Fig. 5. (a) Comparison of observed and simulated flow for Illecillewaet watershed (1975–1976); and (b) various components of simulated flow.

5. Results 5.1. Causes of peak flows Table 2 presents the analysis of maximum annual peak flows for the two examined watersheds. In the Upper Campbell watershed, four out of seven annual maximum peak flows were generated, mainly by

rainfall, during the fall months (Table 2). For these events the percentage of runoff produced by rain ranged from 78 to 94%. In contrast, the two largest events were produced by winter rain-on-snow and one maximum annual flow was generated by spring rain and snowmelt. Table 2 indicates that the annual maximum flow events in the Upper Campbell watershed are mainly produced by rainfall, which contributes

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

227

Table 2 Causes of annual maximum flows and their contribution (causes: (1) rainfall events; (2) snowmelt events; (3) winter rain-on-snow events; (4) spring rain and snowmelt events; (5) summer runoff events with contribution of glacier melt) Hydrologic year

Upper Campbell 1983–1984 1984–1985 1985–1986 1986–1987 1987–1988 1988–1989 1989–1990 Illecillewaet 1970–1971 1971–1972 1972–1973 1973–1974 1974–1975 1975–1976 1976–1977 1977–1978 1978–1979 1979–1980 1980–1981 1981–1982 1982–1983 1983–1984 1984–1985 1985–1986 1986–1987 1987–1988 1988–1989 1989–1990

Observed flow (m 3/s)

Cause

Snowmelt contribution (%)

806.6 628.7 572.4 865.9 347.4 569.0 873.0

3 1 4 3 1 1 1

9 0 25 3 1 0 2

0 0 0 0 0 0 0

82 94 60 89 78 83 92

254 360 241 328 246 275 246 205 225 225 234 272 436 309 272 341 284 214 271 263

4 4 4 2 5 4 4 2 4 4 4 2 5 4 4 2 4 4 5 4

39 44 53 77 37 50 67 77 50 23 57 70 6 37 63 82 56 69 27 56

1 0 0 0 22 0 0 0 0 1 0 0 8 1 0 0 0 0 11 0

9 36 14 0 0 6 2 0 20 25 5 0 60 24 7 0 23 7 18 4

significantly into high percentages of runoff, even during snowmelt events. For example, in the maximum peak flow of the year 1985–1986, which was produced by spring rain and snowmelt, rainfall runoff contributed 60% of the total peak flow, whereas snowmelt runoff contributed only 25% of the total peak. Contemporary field research (Kattelmann, 1985; Singh et al., 1997) has shown that winter rain-onsnow events coupled with spring snowmelt ripen the snowpack. The snowpack becomes saturated and preferential pathways are developed, which speed up the delivery of rain and snowmelt water through the snowpack. Furthermore, field measurements (Dingman, 1994) indicated that the highest rates of snowmelt are observed during rain-on-snow events

Glacier contribution (%)

Rainfall contribution (%)

when warm moist air moves over the ripe snowpack. These processes can produce large floods in Britain (Archer et al., 1994) and western United States (Kattelmann, 1989; Brunengo, 1990). The previous analysis considers only the annual maximum peak flows. In many years, however, the second and third largest annual flow could be higher than an annual maximum flow of another year. For this reason, a second analysis has been performed considering the peaks above a certain threshold value as discussed in the previous section. Table 3 shows the results of this analysis for the Upper Campbell watershed. Using the triple long-term mean daily flow (214.8 m 3/s) as a threshold flow, 40 events were identified. Of these 40 events, 17 were

228

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

Table 3 Causes of peak flow generation and their contribution (causes: (1) rainfall events; (2) snowmelt events; (3) winter rain-on-snow events; (4) spring rain and snowmelt events; and (5) summer runoff events with contribution of glacier melt) Cause

Upper Campbell 1 3 4 Illecillewaet 2 4 5

Number of observations

Flow range (m 3/s)

Snowmelt contribution range (%)

Glacier contribution range (%)

Rainfall contribution range (%)

17 16 7

220–873 219–866 218–572

0–2 2–11 14–52

0 0 0

56–97 53–91 10–71

7 49 50

205–343 161–360 161–436

64–82 23–80 0–49

0–2 0–2 1–46

produced by rainfall, 16 by winter rain-on-snow and only seven by spring rain and snowmelt. It is evident from these results that most peak flows in the Upper Campbell watershed are produced by rainfall during the fall months, and winter rain-on-snow events. The spring rain and snowmelt events produced only one large peak discharge (572 m 3/s), which was also the annual maximum for the year 1985–1986. For this event, however, the contribution of rainfall runoff was 60% of the total peak flow. These results show that floods in the Upper Campbell watershed are mainly produced by rainfall runoff. The contribution of rainfall runoff for peaks above threshold ranged from 53 to 97% with the largest contribution during the maximum observed flood events. On the other hand, the contribution of snowmelt runoff was small and ranged from 2 to 11% of the total observed peak flow (Table 3). The above results could be explained by the prevailing weather systems. The majority of annual precipitation in the area of the Upper Campbell watershed is produced by frontal storms that are generated above North Pacific Ocean and travel eastward (Loukas, 1994). The first obstacles that they meet during their movement are the mountains of Vancouver Island and the Coastal Mountains of British Columbia. A large portion of atmospheric moisture falls, during the orographic lifting, in the form of high intensity rain. Even though rainfall is the main cause of floods, significant snow accumulations are observed at the higher elevations. The accumulated snowpack is highly transient and

0–1 0–36 0–60

partially melts during the mild winter months. Hence, the contribution of snowmelt runoff to the maximum annual peak flows and the flood flows is limited. The analysis of flood peaks was repeated for the Illecillewaet watershed and has shown that of the 20 observed maximum annual flows, 13 events were produced by spring rain and snowmelt runoff, four were generated by snowmelt runoff and only three were produced by summer runoff, when the glacier melt contributes, also, to maximum flow (Table 2). For the 13 spring rain and snowmelt events, the contribution of snowmelt runoff ranged from 23 to 69% of the total simulated flood peak, whereas the rainfall runoff contributed from 2 to 36% of the total peak flow. The snowmelt events are, by definition, generated only by snowmelt runoff. Finally, the summer runoff produced the largest annual maximum flow (436 m 3/s) for the observation period. For this particular event, the contribution of snowmelt runoff was only 6% of the total peak, whereas the contribution of glacier melt was 8%. However, this largest annual recorded flow was mainly produced by rainfall runoff, which contributed to 60% of the total peak flow. The second analysis of partial series of peak flows above threshold has shown that the majority of flood events was generated by spring rain and snowmelt runoff and summer runoff consisting of 49 and 50 events, respectively, whereas only seven flood events were produced by only snowmelt runoff (Table 3). The contribution of snowmelt runoff for the spring

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

flood events was high, ranging from 20 to 80%. However, the contribution of snowmelt runoff decreased for the summer flood events and ranged from 0 to 49%. It is also remarkable that the contribution of the rainfall, which in general contributes to low runoff into flood flows for summer events, increased to 60% for the highest recorded flow. The climate of the area of Illecillewaet watershed explains the results found in the previous analyses. The Illecillewaet basin is located in Inland British Columbia on the west slopes of the Rockies, about 500 km from the Pacific Coast. The climate is affected by moist air masses of the North Pacific Ocean, which move eastwards (Loukas and Quick, 1996). Owing to low air temperatures, most of the annual precipitation falls in the form of snow and because of the steep slopes and the orographic lifting of the air mass large accumulation of snow is observed at high elevations. For example, at the meteorological station at Revelstoke (443 m, a.m.s.l.), the mean annual snowfall is 445 mm and at the meteorological station at Mount Fidelity, which is located at an elevation of 1875 m, the mean annual snowfall is 1518 mm. The snowpack at the lower elevations usually melts completely at the end of April, but at the highest elevations there are permanent snowpacks as well as the Illecillewaet glacier. The rapid snowmelt at medium elevations during May and June can produce high flows. During these months, storms produce rainfall runoff, which is added to snowmelt runoff producing flood flows. In the summer months, the decrease of snowmelt runoff is partly balanced by the increase of glacier melt runoff. During this period intense convective storms can produce high runoff, which is added to snowmelt and glacier melt runoff, and in some cases, can generate very high flood flows. 5.2. Frequency analysis This study has shown that the flood flows in the two examined watersheds are generated by more than one cause. This finding has direct implications on flood frequency analysis. Specifically, Pitlick (1994) has shown in his analysis of floods in western USA that the shape of flood frequency curves is affected by the cause of flood generation mechanisms and not by the basin physiography and drainage area. Hence, if the annual flood series are generated by two or

229

more mechanisms, the flood sample should be treated as multiple population series (Watt et al., 1989). Examples of such analyses in Canada are provided by Stoddard and Watt (1970) for Southern Ontario, Waylen and Woo (1982) for British Columbia and Gingras and Adamowski (1993) for New Brunswick. In these studies, difficulties may arise in clearly associating certain peaks with either sub-series. The problem becomes more difficult when the drainage area of the watershed increases. In the present study, the U.B.C. watershed model’s ability, to separately compute the runoff produced by various physical processes, has become an easier way to identify the causes of flood peak. The flood frequency analysis of the annual maximum flood series for the Upper Campbell watershed should not be used for large return periods since only seven years of data exist. To overcome this problem, the Partial Duration Series (PDS) or Peaks Over Threshold (POT) approach, instead of the annual maximum flood series analysis has been adopted (McCuen et al., 1993; Birikundavyi and Rousselle, 1997). According to this approach the threshold level is selected by assuming that the number of exceedances in a year follows a Poisson distribution (El-Jabi et al., 1982). The Poisson parameter K is estimated by: Kˆ

M N

…2†

where M is the number of exceedances observed in N years of record. To select the threshold values the ratio, mean over variance, of the number of exceedances per year is plotted against the mean number of exceedances per year for different base levels (Fig. 6a). As the threshold level is raised, the Poisson distribution normally provides a better fit, i.e. as the threshold value is increased, K is reduced, and the ratio is stabilized around unity, which is the value of K for the Poisson distribution (Fig. 6a). With further increases of the threshold level, the ratio starts to fluctuate. According to this procedure the threshold level is found to be equal to the triple long-term mean daily streamflow (214.8 m 3/s), which corresponds to a value of 5.71 exceedances per year. To test the assumption that the Poisson distribution applies the cumulative theoretical Poisson distribution, the observed distribution is plotted (Fig. 6b) and the Kolmogorov–Smirnov test has shown that this

230

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

Fig. 6. Application of the POT method for Upper Campbell watershed: (a) selection of the threshold flow level; (b) comparison of the cumulative observed distribution and the theoretical Poisson distribution; and (c) comparison of the cumulative observed probability of the flows over threshold with the theoretical exponential distribution.

assumption cannot be rejected at the 5% significance level. It has also been shown (Todorovic and Zelanhasic, 1970; Todorovic, 1978; Ashkar and Rousselle, 1981) that exceedances within a season

or a year can be taken as belonging to an exponential distribution: F…x† ˆ 1 2 exp‰2b…x 2 x0 †Š; x . x0

…3†

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

231

Fig. 7. Non-homogeneous population of floods in Upper Campbell watershed divided into two homogeneous sub-populations: (a) rainfall; b) winter rain-on-snow; and (c) total population of floods.

where b is equal to:



M M X iˆ1

…xi 2 x0 †

…4†

The parameter b is calculated to be 0.0051. The cumulative exponential distribution function is plotted in Fig. 6c, along with a plot of the observed distribution. The hypothesis that an exponential distribution applies cannot be rejected at the 5% significance level using the Kolmogorov–Smirnov test.

232

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

The previous analysis of the causes of flood flows has shown that the majority of floods in the Upper Campbell watershed is produced by rainfall runoff and winter rain-on-snow events. Using the threshold value of flow, two sub-series of flood events have been compiled for the two main flood producing mechanisms. The two sub-series have been assumed to be independent and have been analyzed separately. The Extreme Value I (EVI) (Gumbel, 1958) distribution is fitted to the observed values of the two flood populations (Fig. 7a and 7b). Finally, the compound EVI distribution is calculated by: Tˆ

Ta Tb Ta 1 Tb 2 1

…5†

where T is the return period of a flood magnitude xT when it belongs to the compound EVI distribution, Ta is the return period if it belongs to rainfall flood series and Tb is the return period if it belongs to winter rain-on-snow flood series (Waylen and Woo, 1982). Furthermore, the EVI is fitted to the total population of floods without consideration of subseries (Fig. 7c). The results of the analysis indicate that the slopes of the fitted compound EVI distributions for the rainfall events and the rain-on-snow events are similar (Fig. 7a and 7b). As a consequence, there is small deviation between the compound EVI distribution and the fitted EVI (Fig. 7c). Application of the Kolmogorov–Smirnov test has shown that the two distributions are not significantly different at the 5% level. In the previous section, it has been shown that the rainfall runoff contributed 78–94% of the annual maximum flood and 53–97% for the flood peaks above the threshold for the rainfall and winter rainon-snow events. It is expected that the component of frequency distributions for rainfall, and for rain-onsnow events, would have similar shapes, since rainfall is the dominant component of flood peaks for both types of flood producing mechanisms. The flood frequency analysis in the Illecillewaet watershed has considered two component distributions, one from the snowmelt runoff (Fig. 8a) and one from the spring rain and snowmelt runoff (Fig. 8b). The analysis has shown that the two fitted component EVI distributions have similar shapes. As a result, the compound EVI distribution is not

significantly different, at the 5% level, from the fitted EVI distribution for the total annual flood series (Fig. 8c). This finding is in accordance with the previous results, since most of the flood events are generated mainly by snowmelt runoff. The snowmelt runoff accounted for 23–82% of the total peak flow, whereas the rainfall runoff during spring rain and snowmelt events contributed only to 0–36% of the total peak flow. Thus, it is evident that both component distributions should be mainly affected by the frequency distribution of snowmelt runoff. It should be mentioned that the largest recorded annual flood peak in Illecillewaet watershed deviates from the distribution (Fig. 8c). This particular event was mainly generated by rainfall runoff, which accounted for 60% of the total peak flow, whereas the snowmelt runoff and glacier melt runoff contributed only 6 and 8% of the total peak flow, respectively. Hence, it is reasonable that this event deviates from the distribution of the other events, which were mainly produced by snowmelt runoff. 5.3. Comparison of the results for the two study watersheds The results of the analysis for the coastal Upper Campbell watershed showed that most of the annual maximum flows and the peak flows over a threshold are generated during fall rainfall and winter rain-onsnow events. The contribution of rainfall runoff to peak flow, in either case, is very high taking its largest value during the maximum flood events. As a result, the flood frequency curves for the rainfall and the rain-on-snow events have similar shapes and can not be distinguished. On the other hand, the analysis for the interior Illecillewaet basin showed that most flood events are generated during the spring and summer months by intense snowmelt and combined rain and snowmelt. Snowmelt runoff contributes the largest percentages of peak flows, for both types of events, although rainfall runoff may contribute significant volumes of flow. The result is that the flood frequency curves for the spring snowmelt events and the spring rain and snowmelt events are similar. The above results may indicate that the response rainfall runoff and snowmelt runoff could be similar depending on the rate of snowmelt and the rainfall

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

233

Fig. 8. Non-homogeneous population of floods in Illecillewaet watershed divided into homogeneous sub-populations: (a) snowmelt; (b) spring rain and snowmelt; and (c) total population of floods.

intensity, respectively. As discussed previously, the ripe and moist snowpacks may melt at high rates that may be comparable to rainfall intensities occurring during snowmelt. However, this supposition needs to be thoroughly examined and studied.

6. Conclusions The results of this study indicate that the flood flows in the maritime basin of Upper Campbell are generated by autumn rainfall events and winter rain-onsnow events. The major component of runoff, in either

234

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235

case, is generated by rainfall, which has the largest contribution to the peak discharge. This result agrees well with the climate of the coastal region of British Columbia and the results of other studies in the Pacific Northwest. For example, Environment Canada (1982) analyzed the flow records of 58 basins in coastal British Columbia with nine or more years of record. The study concluded that 74% of the basins experienced floods that are predominantly rainfallinduced during the fall and winter months. In another study, Melone (1986) found that more than 90% of the maximum observed floods in coastal British Columbia and southeast Alaska occur during the fall and winter period. On the other hand, the flood flows in the inland basin of Illecillewaet are generated either by spring rain and snowmelt, or only snowmelt or summer runoff with the contribution of glacier melt events. The first two categories of events produce most of the annual maximum floods, whereas the summer runoff events can generate large floods mainly because of intense convective rainstorms. However, for most of the peak flows, snowmelt is the most significant and frequent component of flood runoff. The frequency analysis of flood flows for both examined watersheds has shown that the component distributions have similar slopes, since they are mainly affected either by the rainfall frequency distribution, such as in the case of Upper Campbell watershed, or by the snowmelt frequency distribution for the Illecillewaet basin. As a result the compound frequency distributions are not significantly different from the fitted theoretical EVI distributions to the total flood sample for both watersheds. Forest logging is a significant economic factor in British Columbia. The above findings may affect the forest management practices since logging influences the magnitude and timing of peak flows mainly in snow dominated areas, such as the area of Illecillewaet watershed. On the other hand, minimal effects of forest logging are anticipated in the maritime basin of Upper Campbell, where rainfall runoff is the prevailing flood producing mechanism. Furthermore, the estimation of design floods for hydrotechnical projects requires accurate prediction of flood recurrence intervals. Identification and determination of the causes of flood flows may improve flood frequency analysis. The precise classification

of peak flows according to the flood generation mechanisms requires detailed precipitation, temperature and snowmelt data, which are rarely available for medium and large mountainous watersheds. Use of a hydrological model, able to simulate reasonably well the observed hydrograph and the runoff of various physical types, with limited meteorological data, has been shown in this study, that can overcome the lack of data availability in time and space. The technique presented in this study can be used to identify the causative flood mechanisms and the types of flood events in large basins for flood frequency studies and to assist forest hydrologists in the identification of the effects of forest management practices on peak flows. Acknowledgements Dr. M.C. Quick provided the most recent version of the U.B.C. Watershed Model and the data from the two study watersheds. This assistance is gratefully acknowledged. The helpful comments of Dr. J. Pitlick and an anonymous reviewer are also gratefully acknowledged. References Archer, D.R., Bailey, J.O., Barrett, E.C., Greenhill, D., 1994. The potential of satellite remote sensing of snow over Great Britain in relation to cloud cover. Nordic Hydrol. 25, 39–52. Ashkar, F., Rousselle, J., 1981. Design discharge as a random variable: a risk study. Wat. Resour. Res. 17 (3), 577–591. Berris, S.M., Harr, R.D., 1987. Comparative snow accumulation and melt during rainfall in forested and clear-cut plots in the western Cascades of Oregon. Wat. Resour. Res. 23 (1), 135– 142. Birikundavyi, S., Rousselle, J., 1997. Use of partial duration series for single station and regional analysis of floods. J. Hydrol. 2 (2), 68–75. Brunengo, M.J., 1990. A method of modeling the frequency characteristics of daily snow amount for stochastic simulation of rain-on-snowmelt events, Western Snow Conference, 58, pp. 110–121. Dingman, S.L., 1994. Physical Hydrology, Prentice Hall, Englewood Cliffs, NJ. Dracup, J.A., Lee, K.S., Paulson Jr, E.G., 1980. On statistical characteristics of drought events. Wat. Resour. Res. 16 (2), 289–296. Dracup, J.A., Lee, K.S., Paulson Jr, E.G., 1980. On the definition of droughts. Wat. Resour. Res. 16 (2), 297–302. El-Jabi, N., Richard, D., Ashkar, F., Rouselle, J., 1982. Analyse

A. Loukas et al. / Journal of Hydrology 227 (2000) 218–235 stochastique des debits de crue pour la province de Quebec, Ministere de l’Environment, Quebec, Canada. Environment Canada, 1982. Magnitude of floods, British Columbia and Yukon territory, Water Survey of Canada, Ottawa, Canada. Gellens, D., 1991. Impact of a CO2-induced climatic change on river flow variability in three rivers in Belgium. Earth Surf. Proc. Land. 16, 619–625. Gingras, D., Adamowski, K., 1993. Homogeneous region based on annual flood generation mechanisms. Hydrol. Sci. J. 38 (2), 103–121. Gumbel, E.J., 1958. Statistics of Extremes, Columbia University Press, New York. Harr, R.D., 1981. Some characteristics and consequences of snowmelt rainfall in western Oregon. J. Hydrol. 53, 277–304. Harr, R.D., Harper, W.C., Krygier, J.T., Hsieh, F.S., 1975. Changes in storm hydrographs after road building and clearcutting in the Oregon coast range. Wat. Resour. Res. 11 (3), 436–444. Jones, J.A., Grant, G.E., 1996. Peak flow responses to clear-cutting and roads in small and large basins, western Cascades, Oregon. Wat. Resour. Res. 32 (4), 959–974. Kattelmann, R.C., 1985. Macropores in snowpacks of Sierra Nevada. Ann. Glaciol. 6, 272–273. Kattelmann, R.C., 1989. Spatial variability of snowpack outflow at a site in Sierra Nevada, USA. Ann. Glaciol. 13, 124–128. Loukas, A., 1994. Mountain precipitation analysis for the estimation of flood runoff in coastal British Columbia. Ph.D. thesis, University of British Columbia, Vancouver, Canada. Loukas, A., Quick, M.C., 1996. Effect of climate change on hydrologic regime of two climatically different watersheds. J. Hydrol. Engng. 1 (2), 77–87. MacDonald, L.H., Hoffman, J.A., 1995. Causes of peak flows in northwestern Montana and northeastern Idaho. Wat. Resour. Bull. 1, 49–95. McCuen, R.H., Johnson, P.A., Hromanka, T.V., 1993. Regionalized partial-duration balanced-hydrograph model. J. Irrig. Drain. Engng. 119 (6), 1036–1051. Megahan, W.F., 1983. Hydrologic effects of clearcutting and wildfire on steep granitic slopes in Idaho. Wat. Resour. Res. 19 (3), 811–819. Melone, A.M., 1986. Extreme floods in the pacific coastal region.

235

Ph.D. thesis, Department of Civil Engineering, University of British Columbia, Vancouver, Canada. Pitlick, J., 1994. Relation between peak flows, precipitation, and physiography for five mountainous regions in the western USA. J. Hydrol. 158, 219–240. Quick, M.C., 1995a. The U.B.C. Watershed Model. In: Singh, V.P. (Ed.). Computer models of watershed hydrology, Water Resources Publ, Littleton, CO. Quick, M.C., 1995b. U.B.C. Watershed Model Manual-Version 4.0. Department of Civil Engineering, University of British Columbia, Vancouver, B.C., Canada. Singh, P., Spitzbart, G., Hubl, H., Weinmeister, H.W., 1997. Hydrological response of snowpack under rain-on-snow events: a field study. J. Hydrol. 202, 1–20. Stoddard, R.B.L., Watt, W.E., 1970. Flood frequency prediction for intermediate drainage basins in southern Ontario, Deptartment of Civil Engineering, Queen’s University, Kingston, Ontario Research Report No. 66. Thomas, R.B., Megahan, W.F., 1998. Peak flow responses to clearcutting and roads in small and large basins, western Cascades, Oregon: a second opinion. Wat. Resour. Res. 34 (12), 3393– 3403. Todorovic, P., 1978. Stochastic models of floods. Wat. Resour. Res. 14 (2), 345–356. Todorovic, P., Zelanhasic, E., 1970. A stochastic model for flood analysis. Wat. Resour. Res. 6 (6), 1641–1648. Troendle, C.A., King, R.M., 1985. The effect of timber harvest on the Fool Creek watershed 30 years later. Wat. Resour. Res. 21 (12), 1915–1922. Watt, W.E., Lathem, K.W., Neill, C.R., Richards, T.L., Rousselle, J. (Eds.), 1989. Hydrology of floods in Canada: a guide to planning and design National Research Council, Ottawa, Ontario, Canada. Waylen, P., Woo, M.K., 1982. Prediction of annual floods generated by mixed processes. Wat. Resour. Res. 18 (4), 1283–1286. Wright, K.A., Sendak, K.H., Rick, R.M., Thomas, R.B., 1990. Logging effects on streamflow: storm runoff at Caspar Creek in northwestern California. Wat. Resour. Res. 26 (7), 1657– 1667.