Nonparametric modelling of the relationship between wind characteristics and water level fluctuations in a large reservoir

The natural inflow at a site is a key variable for optimal management of water resources, particularly for hydroelectric production. For sites with dams and hydroelectric powerplants, this variable cannot be measured directly, and the water balance equation is used to determine the quantity of water a site receives on its surface during a certain period of time. However, several errors affect the natural inflows computed this way. One of the principal sources of uncertainty for large reservoirs is the nonrepresentativeness of water level because of the wind effect. To quantify the effect of wind on the reservoir surface, a nonparametric regression model was used to relate the water level differences between several stations located on the same reservoir and the characteristics of the wind (direction and intensity). The study showed that the nonparametric regression model substantially improves the knowledge of the water level differences between several stations when there is presence of wind. With this model, it is possible to characterize the types of wind affecting the reservoir and to establish validation strategies for the data. The studied reservoirs are Outardes-4 and Gouin, two large reservoirs located in the north of the province of Quebec, Canada.Original Abstract: L'apport naturel a un site est une variable cle pour la gestion optimale des ressources hydriques, en particulier pour la production hydroelectrique. Pour les bassins amenages d'ouvrages hydrauliques, cette variable ne peut pas etre mesuree directement, et l'equation du bilan hydrique est employee afin de determiner la quantite d'eau que recoit un bassin a sa surface durant une certaine periode de temps. Toutefois, plusieurs erreurs entachent l'apport naturel ainsi mesure. A une echelle journaliere, une des principales sources d'incertitude pour les grands reservoirs est la non-representativite du niveau mesure en raison de l'effet du vent. Afin de quantifier l'effet du vent a la surface du reservoir, un modele de regression non parametrique a ete utilise pour mettre en relation la difference de niveaux entre les limnimetres d'un meme reservoir et les caracteristiques du vent (direction et intensite). L'etude a demontre que le modele de regression non parametrique ameliore substantiellement la connaissance des differences de niveaux entre plusieurs limnimetres lorsqu'il y a presence de vent. Il permet aussi de caracteriser les types de vent qui affectent le plus un reservoir et d'etabl