Estimating actual rainfall from satellite rainfall products

The lack of uncertainty measures in operational satellite rainfall (SR) products leads to a situation where users of the SR products know that there are significant errors in the products, but they have no quantitative information about the distribution of these errors. The authors propose a semiparametric model to characterize the conditional distribution of actual rainfall (AR) given measures from SR products. The model consists of two components: a conditional gamma density given each SR, and a smooth functional relationship between the gamma parameters and SR. The model is developed for monthly rainfall, estimated from a satellite with sampling frequency once a day, averaged over an area of 512 × 512 km2 in the Mississippi River basin. The conditional distribution results are more informative than deterministic SR products since the whole conditional distribution enables users to take appropriate actions according to their own risk assessments and cost/benefit analyses.