Simulation

CHARLES A. LIN1, LEI WEN1, DIANE CHAUMONT2 AND

MICHEL BELAND3 1 Department of Atmospheric and Oceanic Sciences, and Global Environmental and Climate Change Centre, McGill University, 805 Sherbrooke Street W.,

Montreal (Quebec), H3A 2K6, Canada, 2Ouranos Consortium sur la climatologie regionale et l'adaptation aux changements climatiques, 550 Sherbrooke Street W., Montreal (Quebec), H3A 1B9, Canada, 3Reseau quebecois de calcul de haute performance, Universite de Montreal, C.P. 6128, succ. Centre-ville, Montreal (Quebec), H3C 3J7, Canada

15.1 INTRODUCTION

The atmosphere, ocean, and land surface are important components of the global hydrological cycle. Interactions in the coupled atmosphere-ocean-land surface system give rise to weather and climate variability. At the regional scale, the exchange of moisture and heat between the land surface and the atmosphere determines the low level atmospheric humidity and temperature fields, which in turn has an impact on regional weather and climate. The land surface is the interface between the atmosphere and the underlying hydrological regime; the latter is characterized by surface runoff, interflow, baseflow, soil moisture, and other hydrological variables. The importance of the land surface scheme (LSS) and its feedback to the atmosphere have long been recognized for regional climate. For example, Entekhabi et al. (1999), in their proposed agenda for land surface hydrology research, note that surface soil moisture is as important as sea surface temperature as boundary conditions for the climate system. In fact, they point out that precipitation extremes in the United States can be more strongly influenced by soil moisture fields than sea surface temperatures.

There have been less studies of the effect of LSSs on regional weather, in particular, precipitation. A reason is the perception that the atmosphere may not have enough time to react to significant surface flux changes, and thus LSSs are not as important for short-range precipitation forecasts. Recent studies show that this is not always the case. Entekhabi et al. (1999) report an improvement in the threat score of 36-h precipitation forecast by a weather prediction model when coupled to a more sophisticated LSS. The improvement is of the same order as a doubling of the resolution of the atmospheric model; a similar result was also obtained by Wen etal. (2000a). The latter evaluated the effects oftwo LSSs coupled to a mesoscale atmospheric model, and showed that the simulated precipitation over a 48-h period does depend on the LSS used. In addition, the

Climate and Hydrology in Mountain Areas. Edited by C. de Jong, D. Collins and R. Ranzi © 2005 John Wiley & Sons, Ltd partitioning of the simulated sensible and latent heat flux also changes. The sensitivity of simulated short-range precipitation to LSS is important for flood prediction because the single most uncertain variable in this case is the precipitation.

In addition to improving short-range precipitation, a mesoscale atmospheric model coupled to an LSS offers the potential of a longer lead time for flood prediction compared to other means of precipitation forecast. The displacement of rain cells obtained from radars based on Lagrangian advection offer a predictability lead time of up to 3h (Zawadzki etal. 1994); the use of composite radars could extend this up to 6 h (Germann and Zawadzki 2002). The predictability depends on the spatial scale considered. The predictability limit using precipitation from rain gauges is likely to be shorter than with radars because the gauge coverage is usually not sufficient. The representativeness of rain gauge data is a classic problem in hydrological applications. As an alternative, the use of precipitation from mesoscale models could extend the predictability limit to beyond several hours. For example, Wen etal. (2000a) showed that a reasonable 48-h simulation of precipitation is obtained with a mesoscale model coupled to an LSS when compared with observed values from rain gauges. This could potentially provide a longer lead time in flood forecast.

Other studies have examined the coupling of atmospheric and hydrological models. Georgakakos and Hudlow (1984) advocated the use of coupled cloud, soil moisture, and channel routing models for hydrological applications. They also proposed an assimilation procedure for ingesting rainfall and flow data to update the model state. The use of rainfall from such cloud models successfully extended the lead time of rainfall and flow prediction (Georgakakos 1986a, b). Operational applications have also been made (Bae et al. 1995; Georgakakos 2002). Vorosmarty etal. (1993) discuss different aspects of macroscale hydrological models, with an emphasis on linkage to atmospheric models. Some key problems in the coupling of these models are discussed and summarized in Schultz et al. (1995). Benoit etal. (2000) used one-way coupling to drive a hydrological model, with precipitation from a mesoscale atmospheric model. In this way, hydrological basins can be used as macrorain gauges to verify the precipitation from atmospheric models. Pietroniro et al. (2001) examined different levels of coupling, leading to a fully interactive coupled meteorological-hydrological model. Seuffert et al. (2002) used a coupled mesoscale atmospheric model and a land surface hydrological model to study regional weather. The atmospheric models have improved over the years in resolution and physical parameterizations, which will help improve precipitation forecast. This is important because precipitation is the single most uncertain parameter in flash flood prediction.

LSSs coupled to weather prediction models generally do not generate runoff, as lateral flow processes are not resolved in the LSS. As runoff is important for flood simulation, the LSS needs to be modified. Wen et al. (2000b) and Lin etal. (2002) did this using a field capacity threshold interflow generation in an LSS, which was subsequently coupled to a mesoscale atmospheric model. A router was then used to obtain a hydrograph. This approach would help answer the first priority science question posed by Entekhabi et al. (1999) in their agenda for land surface hydrology research: ''What are the mechanisms and pathways by which the coupling between surface hydrological systems and the overlying atmosphere modulate weather and climate variability?'' by determining the contributions of land-atmosphere coupling to severe precipitation and flood events.

In this paper, we report on a coupled meteoro-logical-hydrological modelling system consisting of a mesoscale atmospheric model, a runoff-generating LSS, and a router. The system is described in Section 15.2. Each component of the system is tested in an uncoupled stand-alone mode in Section 15.3, and the coupled model is applied to the flood that occurred in the Sague-nay region of Quebec, Canada in July 19-21, 1996, in Section 15.4. The conclusions are presented in the final section.

15.2 DESCRIPTION OF THE MODELLING SYSTEM

The meteorological-hydrological modelling system consists of a mesoscale atmospheric model, coupled interactively to an LSS that generates runoff, and a routing module that is run off-line. We now describe each of the components and the coupling procedure.

The atmospheric model is the Mesoscale Compressible Community Model (MC2, Benoit et al. 1997) developed in Environment Canada. MC2 is a limited area, high-resolution 3-dimensional regional model with compressible non-hydrostatic dynamics and it uses a semi-Lagrangian treatment of advection. The treatment of physical processes includes a planetary boundary layer based on turbulent kinetic energy (Benoit et al. 1989), short- and long-wave radiation schemes with interactive clouds, shallow convective parameterization and Sundqvist parameterization (Sundqvist et al. 1989; Yu et al. 1998) for resolved scale condensation and Kuo's scheme for subgrid scale convection (Kuo 1974).

MC2 is a community model that is well documented in the literature; it has been used extensively in different studies in Canada and abroad. We shall describe the LSS and routing module in more detail.

An LSS is needed to provide lower boundary conditions to MC2 over land areas. The LSS is a modified version of the Canadian Land Surface Scheme (CLASS; Verseghy 1991; Verseghy et al. 1993), modified to generate runoff for flood simulation. The standard version of CLASS is a physically based 1-dimensional column model designed to represent the average characteristics of a grid square. It resolves three soil layers, and uses a ''mosaic'' approach to treat surface vegetation in a grid square. Each square includes up to four surface types: bare soil, snow cover alone, vegetation cover alone, and both snow and vegetation cover. The vegetation itself has four types: needleleaf trees, broadleaf trees, crops, and grass. The surface energy budget is resolved for each of the four surface types when present. The sensible and latent heat flux fed back to the atmosphere are the sums over the four surface types weighted by their area fractions. For the soil column, the depth-averaged temperature and soil moisture content are calculated at each time step for each of the three layers. The soil moisture flux between layers is obtained using Darcy's Law for a 1-dimensional fluid. The Green-Ampt method is used for infiltration, and hence for surface runoff when precipitation exceeds infiltration capacity and the surface ponding limit is reached. The soil moisture flux at the bottom of the third layer (baseflow) is calculated using the third layer hydraulic conductivity modified by a drainage parameter that depends on soil type. This is generally a reasonable approximation as the variation of the deep soil water content is small over the small time steps (less than 30 min) used in CLASS.

It is well known that interflow is important for flood events in humid and semi-humid regions like eastern Canada. However, the standard version of CLASS has no treatment of interflow, and the baseflow treatment is not appropriate for flood simulation when the soil moisture is improperly initialized. It is thus necessary to modify CLASS. The mechanisms of runoff generation in humid and semi-humid regions are relatively well understood and can in principle be modelled with great detail for a point location, hydrological element, or small experimental catchment, if all the model parameters are measured and calibrated, and initial and boundary conditions provided. However, the application of a sophisticated runoff generation method over a large catchment is strongly limited by the lack of data for initialization, calibration, and verification. This is especially true of a coupled meteorological-hydrological modelling system with a mesoscale atmospheric model, as tens of thousands of surface grid point points are involved. Another consideration is that, as these coupled models are usually run on supercomputers, the fine tuning of model parameters, common in hydrological modelling, is not readily done. We thus keep the modifications to CLASS for runoff generation as simple as possible. Any newly introduced model parameter is either defined using the existing CLASS database or obtained with minimum tuning. Finally, we wish to point out that our modification to CLASS is to improve its hydrological response to short-term rain events. At present, we do not plan to use the coupled system for long-term simulations, as the predictability of regional mesoscale atmospheric models is usually limited to a few days.

Our major modification to CLASS consists of the use of a field capacity threshold to allow for interflow generation and the introduction of a reservoir at the bottom of the third CLASS soil layer that is similar to the drainage parameterization of Diimenil and Todini (1992). The baseflow drainage in the standard version of CLASS proceeds at a rate governed by a hydraulic conductivity modified by a drainage parameter. Our numerical experiments show that this can result in a large overestimation of the baseflow if the soil moisture content is initialized at too high a value. The initialization of soil moisture is problematic in meteorological and hydrological models. We introduce a reservoir at the bottom of the third soil layer to moderate the baseflow, thus circumventing the initialization problem. For the interflow, we take it from the portion of soil moisture in the first CLASS layer that exceeds field capacity; the primary interflow mechanism is saturated and unsaturated matrix flow enhanced by macropore flow near the surface. The interflow is removed from the first layer at a rate given by the horizontal hydraulic conductivity (KHI) modified by the local slope (s). According to Bear (1972), the ratio (ar) of KHI to the vertical hydraulic conductivity (KVI) ranges from about 10 to 100 near the surface. KVI is calculated in CLASS and thus the modified horizontal hydraulic conductivity can be obtained using ar. The product of ar and s is a tunable parameter. A similar approach is used by Soulis et al. (2000) to model interflow in CLASS. Any part of the interflow that is not removed at the end of each time step is added back to the soil moisture budget for the subsequent time step. The surface ponding limit (2 mm) in the standard version of CLASS was removed for surface runoff. The time step used in the coupled modelling system is dictated by the meteorological component, usually of the order of hundreds of seconds. Our experiments indicate that unless the precipitation rate is extremely large, the surface ponding limit is not met when interflow is generated as described earlier.

The coupling between MC2 and the modified version of CLASS proceeds as follows. Seven atmospheric variables are needed to drive CLASS: short-wave and long-wave radiation, wind, temperature, humidity and pressure at the surface, and precipitation. These variables are provided by the atmospheric model at each time step. CLASS in turn provides the surface sensible and latent heat fluxes to the atmosphere. The latter are calculated using bulk transfer coefficients, and depend on the surface wind speed and the temperature and humidity gradients at the surface. The time step of the coupled MC2/CLASS model is dictated by the smaller time step of the atmospheric model, which is of the order of tens of seconds. The fully interactive coupling between the atmosphere and the LSS is through the sensible and latent heat fluxes.

The off-line routing module in the modelling system is the geomorphological instantaneous unit hydrograph (GUH). The GUH is simple to use; all but one parameter depend upon the geomorphology of the basin, and can be determined directly either from 1:50,000 maps or from remotely sensed data using geographic information system (GIS) methods. The remaining parameter is the mean streamflow velocity. As GUH parameter estimation is relatively independent of rainfall-runoff data, the GUH is a promising tool for calculating flows in small- and medium-sized ungauged catchments. The GUH was first introduced by Rodriguez-Iturbe and Valdes (1979) to link the catchment hydrological response given by the instantaneous unit hydrograph with its geomorphological parameters. The catchment is assumed to be drained by a perfect Horton network. Much research has been published on this subject since its introduction. The principal assumption of the GUH was examined by Wen (1991), and Rodriguez-Iturbe and Rinaldo (1997) provided a review. To make the GUH more practical, Wen et al. (1988) derived a generalized GUH expression for a Horton-type basin of any order. The GUH so derived is a function of Horton's bifurcation ratio, stream length ratio, stream area ratio, length of the highest-order channel, and a dynamic parameter, which is the mean streamflow velocity in the catchment. The GUH can be time varying during the course of a storm, through the time-varying streamflow velocity. The geomorphological parameters can be determined directly from the basin geomorphology, whereas the streamflow velocity can be obtained using historical data or methods appropriate for ungauged catchments (Wen et al. 2001).

15.3 TESTING THE MODELLING SYSTEM IN UNCOUPLED MODE

We describe in this section the numerical experiments we have conducted to test each of the components of the modelling system in an uncoupled stand-alone mode. This is a necessary step in the development of a coupled modelling system. The mesoscale atmospheric model MC2 is tested by comparing the simulated precipitation with radar-retrieved and rain gauge values. The land surface scheme CLASS is tested using observed soil moisture data from two agricultural sites in Quebec, Canada, and observed sensible and latent heat flux values from the Mackenzie GEWEX Study (MAGS). Finally, the GUH is tested using data from flood events in China.

Yu et al. (1998) compared the precipitation simulated by MC2 with values retrieved from the McGill University radar, using two heavy rain cases in the Montreal region, Canada. The LSS used in MC2 is the force-restore scheme (Deardorff 1978), a diffusion-based scheme that is much simpler than CLASS. The goal is thus to test the atmospheric component itself. The first case (October 14-15, 1995) is used to calibrate model parameters, whereas the model is applied with no further tuning to the second case (November 8-9, 1996). The calibration takes into account uncertainty in the values of cloud parameters by varying them within reasonable ranges based on experimental measurements. MC2 is run in a self-nesting mode, with spatial resolution of 50, 18, and 6 km. The radar-retrieved precipitation is obtained from the Marshall Radar Observatory of McGill University, and is first corrected using rain gauge observations through a multiplicative factor. The mean precipitation rates for the two cases are simulated well over a domain of radius 100 km centered at Montreal, the area of coverage of the radar. The banded precipitation features are reproduced well. The model also successfully simulates the precipitation intensity, with the difference between MC2 and radar estimates being the same order as the uncertainty of the radar values, which is estimated to be about ±50%. Time correlation studies were also performed with a sliding time window. The results show that the model has phase errors, with the model precipitation taking place before or after the actual precipitation at a given location. The typical time shift increases from about ±1 to ±3 hours as the forecast lead time increases from 2 to 24 h.

MC2 with the force-restore LSS, denoted as MC2/force-restore, is then applied with no further tuning to the flash flood case that occurred in the Saguenay region of Quebec, Canada during July 19-21, 1996. Initial simulations were performed at a resolution of 35 and 10km (Yu etal. 1997). The simulated

48-hour accumulated precipitation is compared point by point with values from 46 stations in the Saguenay region. Regression lines from a least-square best fit show an improvement of the simulated precipitation with resolution. Qualitative evaluation of the rain distribution is performed by comparison with GOES satellite image of brightness temperature; the spatial structure compares well. Quantitative measures (mean bias and root-mean-square error) are also used to determine the quality of the simulation. The maximum accumulated precipitation over the model domain is 274.4 mm, compared with the simulated maximum of 205.0 mm at 10 km resolution. The bias improves from -23.5 to -14.3mm as the resolution is improved from 35 to 10 km. Similarly, the root-mean-square (RMS) error decreases from 58.3 to 48.6 mm. As the precipitation system is largely topographically forced, a higher resolution improves the topography in the model, and thus the quality of the precipitation simulation. The highest topography in the Saguenay region in the model domain reaches 1086 m. Our results show the importance of model resolution in simulating precipitation in mountainous areas.

We now turn to testing the land surface scheme CLASS in a stand-alone mode. The first test (Wen et al. 1998) examines the CLASS soil moisture regime that is important for runoff generation, through a comparison of the simulated soil moisture content with values retrieved from two experimental agricultural sites in Quebec, Canada, for the 1993 growing season (Figure 15.1). For depths of 0-0.45 m and 0-0.90m of the soil column, the simulated values are within one standard deviation of the measured values, indicating CLASS can reproduce the soil moisture variability over a relatively large depth. Wen et al. (1998) present an error analysis of the soil moisture measurements. The agreement in the surface layer (0-0.15 m) is not as good, but the major features are still captured. Observed precipitation and other atmospheric parameters from a meteorological station are used to drive the column model.

The second test of CLASS (Rodgers 2002) uses data from the Mackenzie GEWEX Study (MAGS),

4030

4030

40 30

40 30

Measured Simulated

1120

Measured Simulated

Julian day of 1993 Julian day of 1993

Figure 15.1 A comparison of the soil moisture content simulated by CLASS (solid line) with observed values (dots) for two agricultural sites in Quebec, Canada over the 1993 growing season. The left and right panels correspond to sites A and B respectively. For each site, the comparison is done for three depths (0-0.15, 0-0.45, and 0-0.90m) of the soil column. Taken from Wen L, Gallichand J, Viau AA, Delage Y, BenoitR (1998) Calibration of the CLASS model and its improvement under agricultural conditions. Trans ASAE 41: 1345-1351. Reproduced by permission of ASAE

19 23

19 23

19 23

80 w 100

19 23

500 450 ) 400 Trn 350

40 al

40 al

Figure 15.2 A comparison of simulated and observed hydrographs for two flood events (October 3-4, 1961; August 12-13, 1975) in the Hong Jia Ta catchment located in the Zhejiang Province of China. The rainfall is obtained from gauge measurements and the GUH is used for the simulated hydrographs

the principal Canadian contribution to the international GEWEX project. Observations taken during Canadian GEWEX Enhanced Study (CAGES) over the period July to September 1999 are used to evaluate the simulated sensible and latent heat fluxes. The field sites are located at two subarctic tundra sites in the Trail Valley Creek drainage basin in the Mackenzie River delta, Northwest Territories, Canada. CLASS is run in a stand-alone mode from July 2 to September 30, 1999, initialized with observations and driven with observed meteorological forcing variables. The simulated sensible and latent heat fluxes averaged over a diurnal cycle are compared with observed values, using different diagnostic statistics (mean bias error, RMS error, index of agreement). The CLASS forcing variables and the latent and sensible heat fluxes are all measured as part of CAGES. The results show the latent heat flux is underestimated by the model and the sensible heat flux is overestimated, with the mean bias and RMS errors of order 20-30 W/m2 and 40-60 W/m2 respectively. The magnitude of these errors can be reduced by at least 50% through further modifications to CLASS that take into account the organic soil nature of the site (Letts et al. 2000) and the local hummock hollow microtopography (Rodgers 2002). We will not focus on these modifications in this paper because they are site specific. For the purpose of coupled meteorological-hydrological modelling, we note that it may not be possible to reduce the errors in sensible and latent heat fluxes much below several tens of W/m2 because errors in the radiative fluxes themselves in atmospheric models are of this order (Feng 2001). The major source of error is the inadequate treatment of clouds.

The final component to be tested in a stand-alone mode is the generalized GUH used for routing. To do this, we use data from 252 flood events over 25 catchments from the Zhejiang Province of China. This province is located in China's southeastern coast, and extends from 27° 12' to 31°31'N latitude, and 118° to 123°E longitude. The 25 catchments are part of the Xin An-Jiang basin and range in area from 16.3 to 330 km2. All but one of the GUH parameters are obtained either from 1:50,000 maps or from remote sensing data using GIS methods. The remaining parameter, the mean streamflow velocity, is determined using methods discussed in Wen et al. (2001) for both gauged and ungauged catchments. Observed hydrographs are used only for model verification. An example of the results of a comparison of simulated and observed hydrographs is shown in Figure 15.2 for two flood events (October 3, 1961 and August 12, 1975) in the Hong Jia Ta catchment.

This is a fifth order Horton basin with a drainage area of 151km2. The agreement is good, and is typical of most of the 252 flood cases examined. The GUH is thus a good tool for flash flood routing.

15.4 APPLICATION OF THE COUPLED MODELLING SYSTEM TO THE 1996 SAGUENAY, QUEBEC FLOOD

We now describe two studies that use the modelling system to study the severe precipitation and flash flood events that occurred in July 19-21 in the Saguenay region of Quebec. The first (Wen etal. 2000a) focuses on the precipitation simulated by MC2 coupled to the standard version of CLASS. The second (Lin et al.

2002) uses MC2 coupled with the modified version of CLASS that allows for runoff generation as described in Section 15.2, together with the GUH run off-line to generate a hydrograph at the outlet of the Ha! Ha! Lake in the Saguenay region.

The synoptic situation that gave rise to the severe precipitation and flash flood is as follows. Heavy rain and early snowmelt in the spring of 1996 resulted in near saturated soil moisture. On July 18, 1996, a small low-pressure system with a central sea level pressure of 1002 mb, which had originated in southern Manitoba, formed and slowly deepened as it moved eastward. When the system reached the Saguenay region in Quebec, the system deepened rapidly, with a pressure drop of 20 mb. Over the next 48 h (July 19-21), the low pressure stalled over the Gaspe Peninsula of Quebec, resulting in intense precipitation. The observed maximum value of the 48-h accumulated precipitation was over 270 mm in the Saguenay region, several times larger than the maximum record of the past 120 years. The subsequent flooding led to a loss of life and widespread property damage.

Wen etal. (2000a) examined the precipitation of the Saguenay storm simulated by MC2 coupled to two LSSs:

force-restore and the standard version of CLASS. As mentioned earlier, the role of LSSs in long-term climate simulations is well recognized, but its role in short-range precipitation forecasts is less clear. Our results show that the impact of LSSs can be significant for the latter as well, especially in regions of complex vegetation variations. The resolution of the coupled models is 10 and 5 km. Figure 15.3 shows a comparison of the simulated 48-h accumulated precipitation with 46 rain gauge measurements for MC2/force-restore and MC2/CLASS in the Saguenay region. The best results are obtained with the more sophisticated land surface scheme (CLASS) at the higher resolution (5 km). In addition, MC2/CLASS at 10 km resolution performs better than MC2/force-restore at 5 km. This is due to the higher effective resolution of CLASS, as it resolves surface vegetation characteristics within a grid box, which is not the case for the force-restore LSS (Wen et al. 2000a).

Figure 15.4 shows the sensitivity of the simulated accumulated precipitation to surface vegetation characteristics. The results for four 10-km MC2/CLASS runs (actual vegetation, uniform coniferous trees, bare soil, grass) and MC2/force-restore are shown. The difference

Observed CLASS_5 km CLASS_10 km Force-restore_ 5 km Force-restore_ 10 km CLASS_5 km CLASS_10 km Force-restore_ 5 km Force-restore_ 10 km

Observed (mm)

Figure 15.3 A scatter plot of the 48-h accumulated precipitation (mm) from MC2/CLASS and MC2/force-restore at 10- and 5-km resolution versus observations (45° line) from 46 rain gauges in the Saguenay region in Quebec, Canada. Regression lines from a least-square best fit are also shown. From Wen L, Yu W, Lin CA, Beland M, Benoit R, Delage Y (2000a) The role of land surface schemes in short-range, high spatial resolution forecasts. Mon Weather Rev 128: 3605-3617. Reproduced by permission of American Meteorological Society

Observed (mm)

Figure 15.3 A scatter plot of the 48-h accumulated precipitation (mm) from MC2/CLASS and MC2/force-restore at 10- and 5-km resolution versus observations (45° line) from 46 rain gauges in the Saguenay region in Quebec, Canada. Regression lines from a least-square best fit are also shown. From Wen L, Yu W, Lin CA, Beland M, Benoit R, Delage Y (2000a) The role of land surface schemes in short-range, high spatial resolution forecasts. Mon Weather Rev 128: 3605-3617. Reproduced by permission of American Meteorological Society

Observed

CLASS_actual

CLASS_tree

CLASS_baresoil

Force-restore

CLASS_grass

CLASS_actual

CLASS_tree

CLASS_baresoil

Force-restore

CLASS_grass

Figure 15.4 A scatter plot of the 48-h accumulated precipitation (mm) obtained with different surface vegetation conditions versus observed values from 46 gauges in the Saguenay region in Quebec, Canada. Results from four MC2/CLASS runs with vegetation conditions of normal, uniform coniferous trees, grass, bare soil, and MC2/force-restore are shown. The resolution is 10 km. From Wen L, Yu W, Lin CA, Beland M, Benoit R, Delage Y (2000a) The role of land surface schemes in short-range, high spatial resolution forecasts. Mon Weather Rev 128: 3605-3617. Reproduced by permission of American Meteorological Society

Observed (mm)

Figure 15.4 A scatter plot of the 48-h accumulated precipitation (mm) obtained with different surface vegetation conditions versus observed values from 46 gauges in the Saguenay region in Quebec, Canada. Results from four MC2/CLASS runs with vegetation conditions of normal, uniform coniferous trees, grass, bare soil, and MC2/force-restore are shown. The resolution is 10 km. From Wen L, Yu W, Lin CA, Beland M, Benoit R, Delage Y (2000a) The role of land surface schemes in short-range, high spatial resolution forecasts. Mon Weather Rev 128: 3605-3617. Reproduced by permission of American Meteorological Society

between MC2/CLASS with actual vegetation conditions and coniferous trees is small. This is because for the actual conditions, mixed forest consists of coniferous and deciduous trees, and the former dominates in the Saguenay region. The difference between bare soil and force-restore is also small, as there is no physically based vegetation treatment in the force-restore scheme. We see from Figure 15.4 that there is sensitivity of the simulated precipitation to surface characteristics. In fact, this sensitivity is comparable to a doubling of spatial resolution (Figure 15.3). We also note that surface characteristics have a direct influence on sensible and latent heat flux, and only indirectly affects precipitation through cloud formation. Thus, the sensitivity of precipitation to the treatment of surface characteristics is lower compared to that of heat fluxes. In fact, Wen et al. (2000a) showed that the partition between sensible and latent heat fluxes is quite different between MC2/CLASS and MC2/force-restore, with their sums remaining about the same.

Finally, we describe the results using MC2 coupled to the modified version of CLASS that generates runoff.

The study area is the Ha! Ha! River basin with a drainage area of 567 km2 and is located in the mountainous and forested area of the Saguenay region. The basin extends about 45 km in the north-south direction and about 14 km in the east-west direction, with the Ha! Ha! River flowing from south to north. Figure 15.5 shows the study area and Table 15.1 shows a summary of the basin characteristics. We use our modelling system to reconstruct the hydrograph at the outlet of the Ha! Ha! Lake in the southern Ha! Ha! River basin. The southern basin has an area of 250km2 and is covered by six 10 x 10 km MC2 grid points. We first examine the precipitation simulated at these six grid points, and compare them with the nearest available gauge measurement located 20 km to the south (Station No. 7043713; Figure 15.6). The initial underestimation of the precipitation is due to the spin-up of the model. The overall agreement is very good, thus giving us confidence in the subsequent hydrograph calculation.

Figure 15.7 shows the hydrograph simulated at the outlet of the Ha! Ha! Lake. The effective precipitation from the six grid points in MC2 coupled to the modified

Figure 15.5 The Ha! Ha! River basin, with six model grid points (labelled 1 to 6) covering the southern basin. From LinCA, Wen L, BelandM, ChaumontD (2002) A coupled atmospheric-hydrological modeling study of the 1996 Ha! Ha! river basin flash flood in Quebec, Canada. Geophys Res Lett 29: 10.1029/2001GL013827. Reproduced by permission of American Geophysical Union

Figure 15.5 The Ha! Ha! River basin, with six model grid points (labelled 1 to 6) covering the southern basin. From LinCA, Wen L, BelandM, ChaumontD (2002) A coupled atmospheric-hydrological modeling study of the 1996 Ha! Ha! river basin flash flood in Quebec, Canada. Geophys Res Lett 29: 10.1029/2001GL013827. Reproduced by permission of American Geophysical Union

Table 15.1 Physical characteristics of the Ha! Ha! River basin

Catchment

Name of the basin/area Mountain range Elevation range of the entire catchment (m) Elevation range of individual sites (m) Latitude and longitude Area in km2 Geology % glacierised

Vegetation type (dominant) % forested

Mean Q at catchment outlet (mm/year)

Ha! Ha! River basin

Laurentians

0-1120

300-1120

Granite and gneissic rocks 100

Mixed coniferous 100 460

version of CLASS is used to drive the GUH routing module offline. The geomorphological parameters of the basin are obtained using GIS methods, and the streamflow velocity is determined using the method proposed by

Wen et al. (2001) for an ungauged catchment. The total runoff at the outlet is 312 million m3, which agrees with the runoff of 322 million m3 of another reconstructed hydrograph obtained with a different hydrological model and source of precipitation (Lapointe et al. 1998). There is a significant time difference between the time of peak precipitation and peak flow in their study (over 30 h) compared to our study (about 8h). Although the precipitation sources used are different, the peak precipitation and its timing are similar in the two studies. It is thus unclear why the time lag in Lapointe et al. (1998) is much longer than in Lin etal. (2002). However, our experience with flood events in basins in China of similar size and characteristics suggests the time lag between peak precipitation and peak flow should be much less than 30 h. There are no observed flow data available at the Ha! Ha! River basin for model verification. We chose this case for study, as it was a big flood event in Canada, ranking as the sixth most costly natural disaster event in the country. Our study, using a coupled model, offers a comparison study with Lapointe et al. (1998) for this case.

Figure 15.6 A comparison of the 48-h model precipitation from MC2/CLASS for the six grid points (Pri,..., Pr6) covering the southern Ha! Ha! River basin with observed values from the nearest gauge (No. 7043713) located 20 km to the south of the basin. From Lin CA, Wen L, Beiand M, Chaumont D (2002) A coupled atmospheric-hydrological modeling study of the 1996 Ha! Ha! river basin flash flood in Quebec, Canada. Geophys Res Lett 29: 10.1029/2001GL013827. Reproduced by permission of American Geophysical Union

Figure 15.6 A comparison of the 48-h model precipitation from MC2/CLASS for the six grid points (Pri,..., Pr6) covering the southern Ha! Ha! River basin with observed values from the nearest gauge (No. 7043713) located 20 km to the south of the basin. From Lin CA, Wen L, Beiand M, Chaumont D (2002) A coupled atmospheric-hydrological modeling study of the 1996 Ha! Ha! river basin flash flood in Quebec, Canada. Geophys Res Lett 29: 10.1029/2001GL013827. Reproduced by permission of American Geophysical Union

JE a

300 250 200 150 100 50 0

Precipitation MC2-CLASS/GUH Lapointe et al. (1998)

Precipitation MC2-CLASS/GUH Lapointe et al. (1998)

8 16 0 8 16 0 8 16 0 8 16 0 8 Local time (h), July 19-23, 1996

Figure 15.7 A comparison of two reconstructed hydrographs at the outlet of the Ha! Ha! Lake in the southern Ha! Ha! River basin. The solid line shows results from our coupled meteorological-hydrological modelling system, and the triangles are taken from the study by Lapointe et al. (1998). The simulation starts at 8 am local time on July 19. See text for further discussion. From Lin CA, Wen L, Beland M, Chaumont D (2002) A coupled atmospheric-hydrological modeling study of the 1996 Ha! Ha! river basin flash flood in Quebec, Canada. Geophys Res Lett 29: 10.1029/2001GL013827. Reproduced by permission of American Geophysical Union

n io tat ipi ci re pr

8 16 0 8 16 0 8 16 0 8 16 0 8 Local time (h), July 19-23, 1996

Figure 15.7 A comparison of two reconstructed hydrographs at the outlet of the Ha! Ha! Lake in the southern Ha! Ha! River basin. The solid line shows results from our coupled meteorological-hydrological modelling system, and the triangles are taken from the study by Lapointe et al. (1998). The simulation starts at 8 am local time on July 19. See text for further discussion. From Lin CA, Wen L, Beland M, Chaumont D (2002) A coupled atmospheric-hydrological modeling study of the 1996 Ha! Ha! river basin flash flood in Quebec, Canada. Geophys Res Lett 29: 10.1029/2001GL013827. Reproduced by permission of American Geophysical Union

15.5 CONCLUSIONS

We have summarized a series of studies that were undertaken to develop, test, and implement a coupled meteorological-hydrological modelling system for flood prediction. Precipitation forecasts from highresolution mesoscale atmospheric models are now developed to a stage where these might be sufficiently accurate to drive hydrological models for flood prediction. If this is the case, the prediction lead time can then be increased compared to other means of forecasting precipitation, such as using displacement algorithms with radar or rain gauge values. The potential increase in lead time is important as precipitation is the single most uncertain unknown in forecasting flash floods.

The modelling system consists of a mesoscale atmospheric model (MC2) coupled to a land surface scheme (CLASS) modified to generate runoff. The GUH is run off-line as a routing module to generate a hydrograph. The LSS is the interface between the atmosphere and the hydrological regime, and the coupling between the two is through the sensible and latent heat fluxes. CLASS is driven by seven atmospheric variables (short-wave and long-wave radiation, wind, temperature, humidity and pressure at the surface, and precipitation) furnished by the atmospheric model, and it returns the sensible and latent heat flux to the atmosphere, thus providing feedback and completing the coupling.

Each component of the modelling system is first tested in a stand-alone mode. The precipitation simulated by MC2 is evaluated using values retrieved from radar and rain gauges. The results for the accumulated precipitation show that the model errors are within the errors of the radar, and there is a phase error in the timing of the precipitation. The soil moisture content and sensible and latent heat fluxes simulated by CLASS in a stand-alone mode are compared with values from two agricultural sites in Quebec, Canada, and the Trail Valley Creek site in Northwest Territories, Canada. The latter is in the Mackenzie Basin, the study site of the Canadian GEWEX program. It is important to verify the accuracy of the sensible and latent heat fluxes, as they provide the feedback to the atmosphere from the LSS. These fluxes are simulated to an accuracy of several tens of W/m2, which is the limit of accuracy of regional atmospheric models, as the radiative fluxes have errors of this order. Finally, the GUH is tested using 252 flood events over 25 catchments from the Zhejiang Province of China.

After testing the model components in a stand-alone mode, we apply the coupled modelling system to the Saguenay flood that occurred in Quebec, Canada in July 19-21, 1996. A hydrograph at the Ha! Ha! Lake in the southern Ha! Ha! River basin is generated using the modelling system, which is compared with another reconstructed hydrograph published in the literature. No observed flow data during the flood are available for model verification. The precipitation used to derive the hydrograph is verified against station observations, and the results are good. We believe that the results of this proof-of-concept study of using a coupled meteorological-hydrological modelling system for flood simulation are encouraging, and the modelling system should be tested with additional flash flood cases for further verification.

REFERENCES

BaeDH, Georgakakos KP, Nanda SK (1995) Operational forecasting with real-time databases. ASCE J Hydraulics Div. 121: 49-60.

Bear J (1972) Dynamics of Fluids in Porous Media. Dover Publisher, New York.

Benoit R, Cote J, Mailhot J (1989) Inclusion of a TKE boundary layerparameterization in the Canadian regional finite-element model. Mon Weather Rev 117: 1726-1750.

Benoit R, Desgagne M, Pellerin P, Pellerin S, Chartier Y, Desjardins S (1997) The Canadian MC2: a semi-lagrangian, semi-implicit wide-band atmospheric model suited for fine-scale process studies and simulation, Mon Weather Rev 125: 2382-2415.

Benoit R, Pellerin P, Kouwen N, Ritchie H, Donaldson N, Joe P, Soulis ED (2000) Toward the use of coupled atmospheric and hydrologic models at regional scale. Mon Weather Rev 128: 1681-1706.

DeardorffJW (1978) Efficient prediction of ground surface temperature and moisture, with inclusion of a layer of vegetation. JGeophys Res 83(C4): 1889-1903.

Dumenil L, Todini E (1992) A rainfall-runoff scheme for use in the Hamburg climate model. In: O'Kane JP (ed) Advances in Theoretical Hydrology, A Tribute to James Dooge. Elsevier, Amsterdam, pp. 129-157.

Entekhabi D, Asrar GR, Betts AK, Beven KJ, Bras RL, Duffy CJ, Dunne T, Koster RD, Lettenmaier DP, McLaugh-linDB, Shuttleworth WJ, van GenuchtenMT, Wei MY, Wood EF (1999) An agenda for land surface hydrology research and a call for the second international hydrological decade. Bull Am Meteorol Soc 80: 2043-2058.

Feng J (2001) Solar radiation in the Mackenzie river basin: retrieval from satellite measurements and evaluation of atmospheric models. Ph.D. thesis, McGill University.

Georgakakos KP (1986a) A generalized stochastic hydrome-teorological model for flood and flash-flood forecasting, 1 formulation. Water Resour Res 22: 2083-2095.

Georgakakos KP (1986b) A generalized stochastic hydromete-orological model for flood and flash-flood forecasting, 2 case studies. Water Resour Res 22: 2096-2106.

Georgakakos KP (2002) US corporate technology transfer in hydrometeorology. J Hydroinformatics 4: 3-13.

Georgakakos KP, Hudlow MD (1984) Quantitative precipitation forecast techniques for use in hydrologic forecasting. Bull Am Meteorol Soc 65: 1186-1200.

Germann U, Zawadzki I (2002) Scale-dependence of the predictability of precipitation from continental radar images. Part I: Description of the methodology. Mon Weather Rev 130: 2859-2873.

Kuo HL (1974) Further studies of the parameterization of the influence of cumulus convection of large-scale flow. JAtmos Sci 31: 1232-1240.

Lapointe MF, Secretan Y, Driscoll SN, Bergeron N, Leclerc M (1998) Response of the Ha! Ha! river to the flood of July 1996 in the Saguenay region of Quebec: large-scale avulsion in a glaciated valley. Water Resour Res 34: 2383-2392.

Letts MG, Roulet NT, Comer NT, Skarupa MR, Verseghy DL (2000) Parameterization of peatland hydraulic properties for the Canadian land surface scheme. Atmos-Ocean 38: 141-160.

LinCA, Wen L, Beland M, ChaumontD (2002) A coupled atmospheric-hydrological modeling study of the 1996 Ha! Ha! river basin flash flood in Quebec, Canada. Geophys Res Lett 29: 10.1029/2001GL013827.

Pietroniro A, Soulis ED, Snelgrove K, KouwenN (2001) A framework for coupling atmospheric and hydrological models. In: Dolman AJ, Hall AJ, Kavvas ML, Oki T, Pomeroy JW (eds) Soil-Vegetation-Atmosphere Transfer Schemes and Large-Scale Hydrological Models. IAHS Publication No. 270, IAHS Press, pp. 27-34.

Rodgers D (2002) Validating Canadian land surface scheme heat fluxes under subarctic tundra conditions. M.Sc. thesis, Department of Atmospheric and Oceanic Sciences, McGill University.

Rodriguez-Iturbe I, Rinaldo A (1997) Fractal River Basins Chance and Self-Organization. Cambridge University Press.

Rodriguez-Iturbe I, Valdes JB (1979) The geomorphologic structure of hydrologic response. Water Resour Res 15: 1409-1420.

Schultz GA, Hornbogen M, Viterbo P, Noilhan J (1995) Coupling Larger-Scale Hydrological and Atmospheric Models. IAHS Special Publication No. 3, IAHS Press.

Seuffert G, Gross P, Simmer C, Wood EF (2002) The influence of hydrologic modeling on the predicted local weather: two-way coupling of a mesoscale weather prediction model and a land surface hydrologic model. J Hydrometeorol 3: 505-523.

Soulis ED, Snelgrove K, Kouwen N, Seglenieks F, Verseghy DL (2000) Towards closing the vertical water balance in Canadian atmospheric models: coupling of the land surface scheme CLASS with the distributed hydrological model WATFLOOD. Atmos.-Ocean 38: 251-269.

SundqvistH, Berge E, Kristjansson JE (1989) Condensation and cloud parameterization studies with a mesoscale numerical weather prediction model. Mon Weather Rev 117: 1641-1657.

Verseghy DL (1991) CLASS-a Canadian land surface scheme forGCMS, I. soil model. Int J Climatol 11: 111-133.

Verseghy DL, McFarlane NA, Lazare M (1993) CLASS-a Canadian land surface scheme for GCMS, II. vegetation model and coupled runs. Int J Climatol 13: 347-370.

Vorosmarty CJ, Gutowski WJ, Person M, Chen TC, Case D (1993) Linked atmosphere-hydrology models at the macro-scale. In: Wilkinson WB (ed) Macro-Scale Modeling of the Hydrosphere. IAHS Publication No. 214, pp. 3-27.

Wen K, LiQ, LuWX (1988) A generalized R-V geomor-phologic instantaneous unit hydrograph expression and its application (in Chinese). JHydrometeorol 45(3): 20-25.

Wen L (1991) The principle assumption of the GUH model. Master of Engineering Science Thesis, National University of Ireland, Galway.

Wen L, Gallichand J, Viau AA, Delage Y, BenoitR (1998) Calibration of the CLASS model and its improvement under agricultural conditions. Trans ASAE 41: 1345-1351.

Wen L, LinCA, LiQ, Wen K (2001) The geomorphological unit hydrograph-a practical tool for catchment flow computation. Technical Report R2001-70, CERCA (Centre de recherche en calcul applique), Montreal (Quebec).

Wen L, Lin CA, Yu W, Beland M, Benoit B, Delage Y (2000b) Land-atmosphere coupling: amodeling study of precipitation, sensible and latent heat flux, and runoff simulations. EOS Trans AGU 81(48): F30.

Wen L, Yu W, Lin CA, Beland M, Benoit R, Delage Y (2000a) The role of land surface schemes in short-range, high spatial resolution forecasts. Mon Weather Rev 128: 3605-3617.

Yu W, Lin CA, Benoit R (1997) High resolution simulation of the severe precipitation event over the Saguenay, Quebec region in July 1996. Geophys Res Lett 24: 1951-1954.

Yu W, LinCA, BenoitR, Zawadzki I (1998) High resolution model simulation of precipitation and evaluation with Doppler radar observation. Water Sci Tech 37: 179-186.

Zawadzki I, Morneau J, Laprise R (1994) Predictability of precipitation patterns: an operational approach. J Appl Meteorol 33: 1562-1571.

0 0

Post a comment