Nonparametric error model for a high resolution satellite rainfall product

Abstract

Characterization of the error distribution of satellite rainfall product has important benefits in practical applications such as rainfall-runoff modeling. Most operational satellite rainfall products are, however, still deterministic and lack any estimate of their uncertainty. Given a high resolution satellite rainfall estimate, it is of interest to know the distribution of the actual rainfall. We develop a new nonparametric model that generates the distribution of actual rainfall values for any given satellite rainfall estimate. The model handles the conditional distribution as the mixture of a positive continuous distribution and a point mass at zero. We fitted and validated the model using rain gauge-adjusted ground-based radar rainfall (representing actual rainfall) and CMORPH satellite rainfall estimates (representing high-resolution satellite rainfall), available at a resolution of 0.25° × 0.25° and 3-hourly, over a domain of 6.25° × 6.25° in the southern United States where the radar rainfall products are considered to be of high quality. The modeling approach can be replicated in other regions.