Comparison of simulated discharges

All three experiments are for the period Aug 1, 1981 to July 31, 1994 without intermediate reinitialization of the state variables. For the first initialisation, a spin-up method is applied: the first year is modeled twice, and then the state variables achieved with the first model run are used to initialise the second for the same year. This procedure enables us to start the simulation with a realistic distribution of soil temperature and moisture. The meteorological variables used are derived from three different interpolations with SAFRAN: one for the 1-km resolution altitudes, one for the 8-km cell altitudes and the third for the three subarea altitudes in each cell.

Because of its typical snowmelt regime, the discharge in the upper Durance catchment is generally very low in autumn and winter, and the main flood peaks occur during spring. The observed maximum of the simulation period (June 8,1983) is 94 m3 s-1 in Briancon and 374 m3 s-1 in

Figure 4.5 (Plate 7) Daily discharges in the upper Durance catchment, observed at the three gauging stations Briancon, L'Argentiere and La Clapiere for 1985/86 and modelled with different resolutions and subgrid parameterisations
Figure 4.5 (Plate 7) (continued)

La Clapiere (for L'Argentiere, observations are missing for this date). In summer, runoff remains low except when a severe storm affects the watershed.

As an example, Figure 4.5 shows the daily discharges for the three gauging stations located in the catchment for a typical year (1985/86). The small oscillations in discharge that are particularly noticeable at Briancon are due to a small dam upstream (Val Cerveyrette). The indices in the figure legend correspond with the experiments conducted in our study (Table 4.2).

Comparing the measured and simulated discharge for the three different experiments, the results for this particular year can be summarised as follows.

• The large snowmelt runoff peaks in spring are overestimated by all versions of the model; the rates of increase and decrease of simulated streamflow are too rapid.

• This overestimation is largest for Briancon and decreases with increasing catchment size (downstream).

• The overestimation is largest for the original 8-km simulation; the 1-km simulation is always better and thus the reference for the improvements achieved

Table 4.2 Indices that characterise the different datasets and experiments




Observed at the gauging station


Reference simulation/1-km resolution


Reference simulation/8-km resolution

8 km/Topography

Simulation with three separate altitude

subareas inside each 8-km cell

8 km/Forest

Simulation with 8-km resolution and

application of the forest climate model

8 km/Topography

Simulation with three altitude subareas

and Forest

inside each 8-km cell as well as

application of the forest climate model

in each altitude subarea

with the three subgrid parameterisation experiments concerning topography, forest climate and their combination.

• The overall error is remarkably reduced in all seasons for the subgrid topography experiment; notably the timing of the onset of snowmelt is better and the rising and recession limbs of the hydrographs are thus closer to the reference.

• The overall error is slightly reduced in all seasons for the forest climate experiment; the rates of increase and decrease of the simulated streamflow are still too rapid.

• The best results compared to the reference simulation are achieved with the combination of the subgrid topography and forest climate parameterisation schemes.

It is apparent that at the 8-km resolution the temporal dynamics of the elevation dependent melting pattern of the snow cover within a model cell cannot be correctly reproduced. With the consideration of the subgrid altitude parameterisation, a significant improvement concerning the snowmelt process is obtained. A smaller improvement is achieved with the forest climate model, but their combination leads to the best representation of the simulated snowmelt peak both in terms of its rising/recession and maximum value. The overestimation of the observed peak discharge, 45.3% with the 8-km resolution experiment for La Clapiere, is reduced to 27.5% (subgrid altitude), 29.2% (forest climate) and 14.1% (combination of the two). For Argentiere, the overestimation of 71.4% is reduced to 46.2, 48.2 and 31.3%, and for Briancon the overestimation of 110.2% is reduced to 55.5, 81.3 and 38.3%, respectively. The reference overestimations of the peak discharge, simulated with the 1-km resolution experiments, are 19.5, 30.5 and 47.9%, respectively. Similar results were obtained for all 14 years. In some ofthe years, the results of the combination experiment are even comparable with the 1-kmresolutionreference (e.g., La Clapiere 1985/86).

0 0

Post a comment