A Path Towards Operational Uncertainty of Cloud Phase Identification

Description

Cloud phase state is a key piece of information in characterizing the impact of clouds on radiation and dynamics. Identifying cloud phase is also the first step towards deriving further information about hydrometeor mass, concentration, and size in remote sensing retrievals. While a variety of phase identification algorithms exist, they don’t have quantitative estimates of their uncertainty when run operationally so it is difficult to interpret disagreements. In this study, we use Bayesian Multivariate Classification to characterize the rigor of identifying cloud phase given different sets of information, including higher order moments of the Doppler spectra from the MicroARSCL VAP. We test this technique using MMCR and HSRL data during the MPACE field campaign at Barrow, and include comparisons with aircraft measurements as validation when possible. Three cases are identified to represent our training data of ice, liquid, and mixed-phase clouds. The results are promising with 97% of test data classified as expected, indicating a promising ability to separate cloud phase using a statistical approach. One of the difficulties of assigning uncertainties to remote sensing retrievals stems both from a lack of objective “truth” of hydrometeor properties, and insufficient independent information available to fully constrain the properties of cloud particles. This work will be aided greatly by routine flights at the NSA site scheduled for this summer.