Stochastic parameterization of shallow cumulus convection estimated from high-resolution model data

Jesse Dorrestijn,Daan T Crommelin,A Pier Siebesma,Harm J J Jonker

doi:10.1007/s00162-012-0281-y

Abstract

In this paper, we report on the development of a methodology for stochastic parameterization of convective transport by shallow cumulus convection in weather and climate models. We construct a parameterization based on Large-Eddy Simulation (LES) data. These simulations resolve the turbulent fluxes of heat and moisture and are based on a typical case of non-precipitating shallow cumulus convection above sea in the trade-wind region. Using clustering, we determine a finite number of turbulent flux pairs for heat and moisture that are representative for the pairs of flux profiles observed in these simulations. In the stochastic parameterization scheme proposed here, the convection scheme jumps randomly between these pre-computed pairs of turbulent flux profiles. The transition probabilities are estimated from the LES data, and they are conditioned on the resolved-scale state in the model column. Hence, the stochastic parameterization is formulated as a data-inferred conditional Markov chain (CMC), where each state of the Markov chain corresponds to a pair of turbulent heat and moisture fluxes. The CMC parameterization is designed to emulate, in a statistical sense, the convective behaviour observed in the LES data. The CMC is tested in single-column model (SCM) experiments. The SCM is able to reproduce the ensemble spread of the temperature and humidity that was observed in the LES data. Furthermore, there is a good similarity between time series of the fractions of the discretized fluxes produced by SCM and observed in LES.

Highlights

The effect of clouds and convection on the large-scale atmospheric state is one of the major sources of uncertainty in weather and climate models
As we focus on shallow cumulus convection, we run Dutch Atmospheric LES (DALES) based on a non-precipitating shallow cumulus case as observed during the undisturbed phase of the Barbados Oceanographic and Meteorological Experiment (BOMEX) [10]
To construct the conditional Markov chain (CMC), we perform the three calculations mentioned at the start of Sect. 4: we determine Nα = 10 turbulent flux centroids (Fig. 4), consider Nμ = 10 resolved-scale states determined by the vertical profiles of θl and qt, and compute the 10 transition probability matrices P (i)

Summary

Introduction

The effect of clouds and convection on the large-scale atmospheric state is one of the major sources of uncertainty in weather and climate models. To resolve the convective dynamics realistically, a numerical model resolution of at least 100 m is required. Current operational numerical weather prediction (NWP) models are still far too coarse to resolve convection: global NWP models are approaching O(10 km) resolutions while high-resolution limited-area models operate at O(1 km) resolution. The atmospheric components of coupled climate models currently use resolutions of O(100 km) or more because of the long simulation time spans.

Objectives

Results

Conclusion