Regionalization of rainfall for broad‐scale modeling: An inverse approach

[1] A spatially consistent approach is used for the representation of rainfall at catchment scale for continuous rainfall-streamflow simulation by using inverse modeling. Representing rainfall data at every location as the product of the mean rainfall by the rainfall series reduced to unit average, it is shown that the regionalization of both terms should follow different ways for broad-scale modeling. Whereas the regionalization of the mean rainfall is based on its spatial continuity, it is demonstrated from the study of three French basins that are subject to different climates that the reduced rainfall data should be represented from a weighted sum of a small number of observed rainfall data (three to five) located both inside and outside the catchments to be as representative as possible at catchment scale. The reliability of peak flow modeling increases with the basin size as well as the return period of flood events provided that the rainfall is correctly regionalized, which is particularly important for real-time forecasting of rainfall and flow. This contradicts the widespread assumption that for the distributed rainfall-runoff model, the denser the network of rain samplers used as input in the models, the more accurate the broad-scale flood model will be.