Observation-Based Cloud and Precipitation Properties from Spaceborne Measurements using a Parallel Bayesian Retrieval Framework

Satellite measurements have long been used to provide information on cloud and precipitation properties from space. Because the measurements are indirect, and the models that convert physical cloud properties (e.g., particle number, diameter, and shape) to estimates of measured quantities (e.g., radar reflectivity, microwave brightness temperature, etc) are approximate, estimates of cloud and precipitation properties have uncertainty. Bayesian retrievals, which characterize sources of uncertainty in terms of their probability distributions, have become the state of the art for modern cloud retrievals. The advantage of a Bayesian framework is its potential to provide both the optimal estimate and an estimate of the uncertainty. However, it is difficult to properly quantify the uncertainty from all possible sources at once.

We have developed a parallel framework that enables the simultaneous estimation of the sources of uncertainty in cloud and precipitation property retrievals using a large ensemble of Bayesian retrieval solutions. A unique ensemble is run for each retrieved profile, and each member of the ensemble includes a different prior mean estimate, as well as modifications to unknown cloud microphysical parameters in the radiative transfer model. Computing statistics across the ensemble of solutions for a variety of different cloud types and scenes enables a more robust computation of uncertainty, and also an examination of its state dependence.