Bias correction of precipitation through Singularity Stochastic Removal: Because occurrences matter
Résumé
This study focuses on how the treatment of the rainfall occurrences in bias correction (BC) contexts may affect the resulting precipitation, both in terms of occurrence and intensity. Three methodologies are compared—the “direct approach” (DA), the “Threshold Adaptation” approach (TA), and the “Positive Approach” (Pos)—as well as a method called “Singularity Stochastic Removal” (SSR) specifically developed for precipitation, all based on the same adjustment technique. Unlike the three other models, SSR allows dealing in the same way with the situations where the precipitation model has too many wet days or not with respect to the reference data. SSR also avoids separating the correction of the occurrence from that of the intensity, which constitutes a flexible tool. First, the four approaches are applied to a historical regional climate model precipitation run. Evaluations are realized through occurrence‐ and intensity‐related criteria. Although SSR, DA, and Pos may be close to each other depending on the criterion, in general, SSR provides the best results when all criteria are accounted for. This is even more true when the classical assumption that “the model precipitation had too many wet days” does not hold. The BC methods are also intercompared over the 2071–2100 period. The different BC methods are in agreement with previous studies, with relatively equivalent evolutions from 1976–2005 to 2071–2100, although nuances are present from one BC method to another. As a global conclusion, the SSR method for precipitation is a good compromise to correct both occurrences and intensities.
Origine : Fichiers éditeurs autorisés sur une archive ouverte