Quels apports hydrologiques pour les modèles hydrauliques? Vers ...

Apart from these few examples, the majority of the studies in which a semidistributedmodel is applied do not distinguish between point and distributed105lateral inflows (Krysanova et al., 1999; Boyle et al., 2001; Onyando et al., 2003;Ajami et al., 2004; Zhang et al., 2004). When coupling RR models with a fullhydraulic model, this point becomes critical since the two types of injectionsare not equivalent when numerically solving the Saint-Venant equations: pointinflows introduce discontinuities in the mass conservation equation that can110lead to numerical instabilities. Conversely, uniformly distributed lateral inflowsfit well with the partial derivative equation system.tel-00392240, version 1 - 5 Jun 20091151201252.2 What Level of Spatial Aggregation Should Be Chosen?Many authors discuss the appropriate level of spatial discretisation in a RRmodelling context, i.e. the number of sub-catchments or the size of computationalgrid that should be used to maximise model performance. Refsgaard andKnudsen (1996), on three catchments in Zimbabwe ranging from 240 to 1090km 2 , as well as Ajami et al. (2004) on the Illinois basin at Watts (1645 km 2 ),did not find significant improvement when shifting from a lumped model toa semi-distributed model for catchment outlet simulations. With similar toolsand objectives, Boyle et al. (2001) reported a subtantial gain in performancewhen shifting from a lumped to a three-sub-catchment semi-distributed model,but no additional gain could be achieved with an eight-sub-catchment distributionon the Blue River catchment (1227 km 2 ). Butts et al. (2004) reportsimilar results based on coupling lumped RR models and different routingschemes including a fully dynamic model on the same catchment. This pointis important here since it suggests the existence of a plateau in model perfor-7