On the tuning of autoconversion parameterizations in climate models

Autoconversion is a highly nonlinear process, which is usually evaluated in global climate models (GCMs) from the mean in-cloud value of the liquid-water mixing ratio q(l)'. This biases the calculated autoconversion rate, and may explain why it usually seems to be necessary to reduce the autoconversion threshold to an unrealistically low value to obtain a realistic simulation in a GCM. Two versions of a threshold-dependent autoconversion parameterization are compared in the CSIRO GCM. In the standard (OLD) treatment, autoconversion occurs in a grid box whenever the mean in-cloud q(l)' exceeds the threshold q(crit), which is derived from a prescribed threshold cloud-droplet radius r(crit). In the modified (NEW) version, the assumed subgrid moisture distribution from the model's condensation scheme is applied in each grid box to determine the fraction of the cloudy area in which q(l)' > q(crit), and autoconversion occurs in this fraction only. Simulations are performed using both treatments, for present-day and preindustrial distributions of cloud-droplet concentration, and using different values for r(crit). Changing from the OLD to the NEW treatment means that r(crit) can be increased from 7.5 mu m to a more realistic 9.3 mu m, while maintaining the global-mean liquid-water path at about the same value. Simulations for preindustrial and present-day conditions are compared, to see whether the change of scheme alters the modeled cloud-lifetime effect. It is found that the NEW scheme with r(crit) = 9.3 mu m gives a 0.5 W m(-2) (62%) stronger cloud-lifetime effect than the OLD scheme with r(crit) = 7.5 mu m.