Use of a cluster analysis to investigate the relationship between large-scale dynamics and clouds

Poster PDF

Description

Cloud parameterizations are an attempt at statistically connecting large-scale dynamics to local cloud and precipitation properties. We investigate those relationships through the use of a clustering technique developed by Marchand and coauthors (2009, Journal of Climate) to classify regional atmospheric states. These atmospheric states are created by a neural network classifier acting on reanalysis data and are refined and shown to be statistically significant through the use of independent cloud radar data. Here we present results from the application of this method to regions surrounding the ARM sites at both Southern Great Plains and Darwin, Australia. In both cases we use ERA-Interim data and observations from the vertically pointed millimeter-wavelength cloud radars at the ARM sites. Having contemporaneous observations of atmospheric state and other atmospheric observables allows us to create distributions of observables associated with each state. Examples of observables include ground-based observations such as cloud occurrence, precipitation, and liquid-water path along with satellite retrievals of the same properties. The multi-year record allows us to investigate the seasonal and interannual variability of the dynamic states and the relationship between these states and larger-scale phenomena such as the Madden-Julian Oscillation (MJO). In addition, we can examine the diurnal cycle of the states, the duration of particular patterns (since we classify the atmospheric state several times a day), and the transition probability from any state to any other state. Our near-term goal is to apply this technique to climate model output to determine to what extent climate models can duplicate both the occurrence of the atmospheric states and the observed linkage between the dynamical states and associated hydrological properties.