GCM Evaluation using Cloud Regimes

This section provides a brief description of my work into representation of convection in climate models. For more information, see my poster at the AMS 31st Conference on Hurricanes and Tropical Meteorology and the associated extended abstract. This is still work in progress.

The extensive deep convective and thick stratiform anvil clouds of the CD regime are produced in global climate models (GCMs), but it is unclear if they are indeed the result of organised deep convection or products of some other processes. This stems from the fact that the organisation of convection is not explicitly taken into account in GCMs, in which convection is represented through parametrisation schemes. In these schemes, the statistical effects of convection are quantified through relationships with the resolved variables, but there is no explicit consideration of convective organisation, assuming instead that such organisation will emerge from the resolved variables. Hence, despite the existence of clouds which in observation is an indication of organised deep convection, it is unknown if models can replicate organised deep convection and its associated properties.

Here, we examine the large-scale environment of the cloud regimes in model runs of the 'amip' experiments in the CMIP5 database. These regimes are derived in a method following Williams and Webb (2008), which is similar to that of Gordon et al. (2005). Instead of assigning the joint-histograms to a regime based on its 42-dimensional vector (7 CTP bins × 6 τ bins), the joint-histograms are reduced to a three-dimensional vector comprising its mean albedo, mean cloud top pressure and total cloud cover, normalised to a range of 0 to 1. They are then assigned their regime membership by comparing them with the equivalent reduction of the centroids of observed regimes. This approach provides a greater degree of tolerance in the identification of model cloud regimes.

We found that the model CD regimes have a lower precipitation and a value of vertical velocity that is too negative. Furthermore, we discovered that these two errors are related, such that a model which overestimates the ascending motion tends to produce rainfall that is closer to observation.

For more details, see the poster and extended abstract.