Identifying Fields of Cumulus in Satellite and HRRR Output to Improve Model Physics

Numerical weather prediction (NWP) mesoscale models have difficulty predicting the timing, location, and macro- and microphysical properties of clouds. Of particular interest are fair weather cumulus clouds because these clouds are smaller than the resolution of the model (i.e., the clouds are subgrid-scale in size). Furthermore, few studies have been conducted to quantify the ability for mesoscale models in predicting shallow cumulus cloud properties. This study proposes a novel method to use GOES-16 visible imagery for the purpose of verification and analysis of the high-resolution rapid refresh (HRRR) model’s cloud prediction capabilities.

The approach taken in this work is to identify fields of cumulus clouds that are free of other cloud types, and then compare the statistics from the GOES observations with the cloud field simulated by the HRRR. The development of new image processing techniques, which take advantage of the typical spatial patterns associated with large-scale cumulus fields, results in a rapid, accurate, and automated method of cumulus field identification within GOES image retrievals. The fitting of a spatial ellipse within identified cumulus regions makes for the trivial computation of an associated “field centroid” within latitude and longitude space. A similar approach is used to identify resolved cumulus fields in the HRRR through the analysis of liquid water path (LWP) distributions. Fields of cumulus clouds that ‘overlap’ between the observations and HRRR can them be compared statistically to see if the model is getting the correct cloud fraction and LWP distribution in each field. This approach, when combined with other datasets (e.g., downwelling shortwave radiative flux from the HRRR and observations and overall environmental conditions), provides better insights on how well the HRRR is able to simulate the shallow convective processes that lead to fair weather cumulus clouds.