Analyzing dynamical circulations in the tropical tropopause layer through empirical predictions of cirrus cloud distributions

Thirty day back trajectories are calculated from the locations of CALIPSO cloud observations to obtain Lagrangian dry and cold points associated with each observation. These historical values are combined with "local" (at the location of the CALIPSO observation) temperature and specific humidity to predict cloud probability using multivariate polynomial regression.

We demonstrate that our statistical sample (seven seasons) is sufficient to retrieve the full nonlinear relationship between cloud probability and its predictors and that substantial information is lost in a purely linear analysis. The best cloud prediction is obtained by the two-variable combination of local temperature and humidity, which reflects the close relationship between clouds and relative humidity. However, single-variable predictions involving air parcel histories are better than those based solely on the individual local fields, indicating the existence of reliable dynamical information content within parcel trajectories.