A Bayesian approach to upscaling and downscaling of aircraft measurements of ice particle counts and size distributions

This study addresses the issue of how to upscale cloud-sized in situ measurements of ice to yield realistic simulations of ice clouds for a variety of modeling studies. Aircraft measurements of ice particle counts along a 79-km zigzag path were collected in a Costa Rican cloud formed in the upper-level outflow from convection. These are then used to explore the applicability of Bayesian statistics to the problems of upscaling and downscaling. Using the 10-m particle counts, the analyses using Bayesian statistics provide estimates of the probability distribution function of all possible mean values corresponding to these counts. The statistical method of copulas is used to produce an extensive ensemble of estimates of these mean values, which are then combined to derive the probability density function (pdf) of mean values at 1-km resolution. These are found to compare very well to the observed 1-km particle counts when spatial correlation is included. The profiles of the observed and simulated mean counts along the flight path show similar features and have very similar statistical characteristics. However, because the observed and the simulated counts are both the results of stochastic processes, there is no way to upscale exactly to the observed profile. Each simulation is a unique realization of the stochastic processes, as are the observations themselves. These different realizations over all the different sizes can then be used to upscale particle size distributions over large areas.