Seeking guidance from active cloud observations to improve climate model subcolumn generators

Abstract

Our objective is to test and improve cloud subcolumn generators used for greater realism of scales in the radiation schemes and satellite simulators GCMs. For this purpose, we use as guidance water content fields from active observations by the CloudSat radar (CPR) and the CALIPSO lidar (CALIOP). Cloud products from active sensors while suffering significant sampling and coverage drawbacks have the advantage of resolving both horizontal and vertical variability which is what the generators are designed to produce. Our first order goal is to test the ability of the generators to deliver realistic 2D cloud extinction (cloud optical thickness) fields using, as in GCMs, limited domain-averaged information. Our reference 2D cloud extinction fields fully resolving horizontal (along the track of the satellites) and vertical variability come from combining CloudSat’s 2B-CWC-RVOD (liquid clouds) and CALIPSO-enhanced 2C-ICE (ice clouds) products. The combined fields were improved by introducing a simple scheme to fill liquid cloud extinction values identified as missing by comparing with coincident 2D (phase-specific) cloud masks provided by the CALIPSO-enhanced 2B-CLDCLASS-LIDAR CloudSat product. Our presentation will demonstrate the substantial improvements for low clouds brought by the filling scheme through comparisons with MODIS-Aqua cloud fraction distributions expressed in terms of joint cloud top pressure – cloud optical thickness histograms. Beyond global comparisons, the nature of the improvements become clearer when comparing mean joint histograms segregated by MODIS Cloud Regime (CR): improvement is by design superior for MODIS CRs dominated by low clouds. With the improved 2D extinction fields at hand, we test the skill of two subcolumn generators, one used in the COSP satellite simulator package, and one with more sophisticated cloud overlap implemented in the GEOS global model, to reproduce joint histograms that are statistically similar to the observed counterparts described above (as interpreted by COSP’s MODIS simulator). Our main comparison metrics are the Euclidean distance between observed and generator-produced global or near-global mean joint histograms, and the statistics of Euclidean distances calculated for individual scenes. One full year of data is used to assess whether the more sophisticated cloud generator produces clouds with greater realism in 2D cloud variability