Regional modelling of the annual mean discharge and average low-flow discharge of the French Moselle sub-basins should improve our knowledge of its hydrosystem functioning. These models are part of the directives of the Water Act and Fishery Act and more recently of the European Framework Directive. Forty-nine sub-basins divided into three groups were used to calibrate and validate a regional model of the annual mean discharge and a regional model of low-flow discharge.
A two-parameter Weibull distribution was used to represent the regional models of the Moselle catchment. Results show a relative dependence between the regional models. A pseudo-dependence was observed between the regional annual mean discharge model and the regional average low-flow discharge model for the average years to the dry ones. This property allowed us to use a common simplified procedure, established from the low-flow episodic discharge measurements, to estimate the flow descriptors of an ungauged sub-basin, namely the median annual mean discharge and the median minimal daily low-flow discharge . For the low-flow regional model a second local descriptor was necessary, the characteristic time of low-flow discharges of the sub-basin (j), which allowed us to generalize the model for every duration (d). The QdF flow-duration-frequency concept, applied to the low-flow discharge, is independent of the selected frequency law and uses the observed convergence of the distributions of various durations (d). The operational character of these regional models depends strongly on the accuracy of the estimation of the flow descriptors of the studied sub-basin ( for the annual mean discharges and for the low-flow discharges). These flow descriptors were estimated according to two approaches: a traditional approach based on multiple regression and a simple approach based on a trend coefficient (k) calculated between low-flow episodic discharge measurements that are concomitant between the studied sub-basin (no or few observations) and the reference sub-basin (time series of continuous discharges). A choice of five low flow discharge measurements per year over the last twelve years (on average) was made. Knowing the flow descriptor of the reference sub-basin, the flow descriptor of the studied sub-basin was then deduced from the trend coefficient k. For we observe in the majority of cases a significant improvement of the estimation obtained by regression, especially a net reduction of the most important variations. A similarity of the surface classes between the studied sub-basin and the reference sub-basin is not required. Geographical proximity of the sub-basins seems to provide better results. The estimation of the median annual mean discharges using multiple regression was rather impressive. In addition, the trend coefficient k established for the low-flow discharges allowed a coherent estimation of and . This quite unexpected result suggests that the observed pseudo-dependence between regional models has a physical reality.
The "regionalization" technique detailed herein does not require a significant investment in flow measurements of the ungauged sub-basin compared to the national hydrometric network. It is an interesting (or complementary) alternative to the regionalization approach based on geostatistical methods that, for instance, identify the homogeneous hydrological vicinity of each sub-basin or take into account the structuring effect of the hydrographic network in the cartography of the flow descriptor.
Regionalization, annual mean discharges, low flow, statistical QdF modelling.
G. Galéa, Cemagref-Lyon, Unité de Recherche
Hydraulique, 3bis quai Chauveau 69336 Lyon Cedex 09 France