Regional climate models are prone to biases in precipitation that are problematic for use in impact models such as hydrology models. A large number of methods have already been proposed aimed at correcting various moments of the rainfall distribution. They all require that the model produce the same or a higher number of rain days than the observational data sets, which are usually gridded data sets. Models have traditionally met this condition because their spatial resolution was coarser than the observational grids. But recent climate simulations use higher resolution and the models are likely to systematically produce fewer rain days than the gridded observations.
In this study, model outputs from a simulation at 2 km resolution are compared with gridded and in situ observational data sets to determine whether the new scenario calls for revised methodologies. The gridded observations are found to be inadequate to correct the high-resolution model at daily timescales, because they are subjected to too frequent low intensity precipitation due to spatial averaging. A histogram equalisation bias correction method was adapted to the use of station, alleviating the problems associated with relative low-resolution observational grids. The wet-day frequency condition might not be satisfied for extremely dry biases, but the proposed approach substantially increases the applicability of bias correction to high-resolution models. The method is efficient at bias
correcting both seasonal and daily characteristic of precipitation, providing more accurate information that is crucial for impact assessment studies.
Figure 3: (a) Schematic of the bias correction proposed by Piani et al. (2010a). Mi is the intensity of an event in teh model and Oi is the intensity of an observed event with the same cumulative probability (CPmi) as defined by Fm and Fo, which are the cumulative probability functions for the model and the observations. (b) Schematic of adaptation of the bias-correction method using stations and regions.