RainFARM: Rainfall Downscaling by a Filtered Autoregressive Model

Abstract

A method is introduced for stochastic rainfall downscaling that can be easily applied to the precipitation forecasts provided by meteorological models. Our approach, called the Rainfall Filtered Autoregressive Model (RainFARM), is based on the nonlinear transformation of a Gaussian random field, and it conserves the information present in the rainfall fields at larger scales. The procedure is tested on two radar-measured intense rainfall events, one at midlatitude and the other in the Tropics, and it is shown that the synthetic fields generated by RainFARM have small-scale statistical properties that are consistent with those of the measured precipitation fields. The application of the disaggregation procedure to an example meteorological forecast illustrates how the method can be implemented in operational practice.