Variance constrained uncertainty quantification method of cloud microphysical property retrievals

Poster PDF

Description

Cloud microphysical properties attained from retrieval algorithms are critically important for evaluating model simulations of clouds and/or cloud parameterization. Previous studies have shown large uncertainties in existing retrieval products, including the Atmospheric Radiation Measurement Climate Research Facility baseline cloud microphysical properties (MICROBASE) product. In this study, we quantify the uncertainties in the key MICROBASE outputs (cloud liquid water content, cloud ice water content, cloud droplet effective radius, and cloud ice particle effective radius) by propagating the uncertainties in its input observations (temperature, equivalent radar reflectivity factor, and liquid water path) through the retrieval model. The uncertainties are calculated relative to the half-hour mean to facilitate the comparison with model outputs. Primary component analysis with variance constraint is applied to reduce the dimension of random variables and reveal the cross correlations in the input data, and hence making the large random sampling calculation computationally feasible. By greatly reducing the degree of freedom in the retrieval input (from 828 to about 10), our approach has the capability of attributing the uncertainties in the retrieval output to individual random input variables, which allows detailed sensitivity studies of MICROBASE algorithm and thus provides directions for improving observation instruments as well as strategies. In addition, we estimate the possibility range of the epistemic uncertainty in the retrieval algorithm associated to the parameters.