Spatial characteristics of precipitation fields and their relation with atmospheric predictors: Application to a stochastic downscaling model

User demand for fine-scale precipitation products grows, but running high-resolution atmospheric models remains a computationally intensive task, which is not affordable at present except for short-range forecasting over a limited area. Thus, a variety of users, in particular hydrologic modelers, rely on statistical downscaling for producing realistic, fine-scale, precipitation fields from coarser-scale fields. However, the question of how physically realistic the downscaled fields are remains a scientific challenge. In this study, we propose novel indices that quantify three important characteristics of a precipitation field: the small-scale variability, the anisotropy strength, and the anisotropy direction. The use of two different mat­hematical tools for defining these indices is investigated: a direction-dependent autocorrelation function, and a 2D-wavelet energy spectrum. The indices are then computed over 14 years of 6h-accumulated high-resolution precipitation analyses over the US, and their correlation with a variety of large-scale atmospheric variables is assessed. This experiment helps us identify the best predictors that control the small-scale variability and the anisotropy of precipitation, in the perspective of their use in a stochastic downscaling model. An example of such a model, based on the Gibbs sampling algorithm, is given in the work of Gagnon et al. [Journal of Hydrometeorology, 13(1), 2011]. Here, we present a variant that uses the new predictors, but also which includes a novel calibration procedure ensuring that the downscaled fields reproduce the spatial characteristics of the original fields.