A new Framework for Cumulus Parametrization - Ideas, Observations and first Results

Most cumulus parameterizations used in weather and climate models are build on the assumption that there is a unique small scale response to each possible large scale state. Nevertheless observations show a significant variability in cumulus activity for a given large scale state, especially with smaller domain size. This variability can be parameterized using Markov chains conditioned on the large-scale state to predict convective area fractions at cloud base.

We present a new framework for cumulus parameterization based on the use of the Markov chain model. The framework adopts a high-resolution grid (O(1km)) embedded in a climate model. For each climate model grid-point, the state of convection at each high-resolution grid-point is predicted using the statistical model. The so-derived sub-grid scale information can then be used in determining the characteristics and effects of convection in the climate model grid cell. We will demonstrate that the model can in principle account for many observed features of convection that conventional models fail to predict, such as the state of organization of convection. The use of the framework across multiple model resolutions will also be discussed.

In a first implementation of the framework, mid-tropospheric humidity and vertical velocity are used to predict the fraction of the grid-box covered by deep convection. A comparison with radar observations shows considerable skill of the statistical model in reproducing the observed behavior of both mean and variability of deep convective area fraction. Together with simple assumptions on density and vertical velocity at cloud base the new framework is used as a closure assumption in an existing convection parametrization. First results indicate significant improvements in the simulation of tropical variability, including but not limited to the MJO without deteriorating the mean climate.