Assessing the Performance of a Prognostic and a Diagnostic Cloud Scheme Using Single Column Model Simulations of TWP-ICE

The Tropical Warm Pool International Cloud Experiment (TWP-ICE) was a major field campaign held in the Darwin area in January and February 2006. One of the main aims of the experiment was to provide boundary conditions and validation data for modelling studies to help facilitate model development with a focus on tropical convection and clouds. The Darwin area experiences a wide array of convective systems consisting of active monsoon periods with typical maritime storms and break periods with more coastal and continental convection. The observations collected during TWP-ICE allow for a detailed evaluation of the ability of numerical models to simulate the evolution of tropical cloud systems and their effect on the environment. Global climate models must be able to represent cloud-scale processes and the feedbacks between clouds and the large-scale environment to ensure accurate projections of climate change. One of the aims of this study is to examine the different effects of the diagnostic and prognostic cloud schemes on the state variables of temperature and humidity. The model used in this study is the Australian Community Climate Earth System Simulator (ACCESS). The atmospheric component of ACCESS is the UK Met Office Unified Model (UM) and the focus of this study is to evaluate the performance of the atmospheric single column model (SCM) in tropical conditions associated with the monsoon and to identify areas for improvement in the model cloud and convection schemes.

The SCM produces generally reasonable representations of the TWP-ICE cloud fields, with the three different observed cloud regimes captured by the model. The new generation prognostic cloud scheme developed at the Met Office, PC2 (prognostic clouds, prognostic condensate), produces deeper clouds during the active monsoon phase than the diagnostic scheme, however both models simulate too much total cloud as is shown in the lower values of outgoing long wave radiation. The excessive cloud produces a cold bias in the levels between 7 and 15 km that is due to longwave radiative cooling. Both of the model runs dissipate cloud too soon after the mesoscale convective system event that occurred during the active period, more so for the diagnostic cloud scheme and this produces too much downwelling solar radiation at the surface for the initial times of the suppressed monsoon phase. The differences seen between the simulations are predominately due to the way in which the prognostic and diagnostic cloud schemes interact with the cumulus parameterization. In PC2 the convection scheme detrains condensate directly into the gridbox thereby allowing the stratiform cloud scheme to reflect details of the convective clouds. This differs from the diagnostic scheme where the detrained condensate evaporates and the radiative effect of the convective cloud is represented by a separate diagnostic cloud category. Both cloud schemes fail to maintain a thick enough anvil cloud during the suppressed phase resulting in too little radiative cooling and a warm bias at the heights of the anvil cloud. The final regime in TWP-ICE was the break period and while the observations during this time had many convective events occurring, these systems were smaller in scale than in the active period and characteristic of continental and coastal convection forced by sea breezes. Due to the nature of these convective cells the forcing data contain the ingredients to produce convection in the SCM. However, the model cloud fields are much longer lived than the observed clouds and not as deep.

The cloud area fraction or cloud cover is not the prognostic variable in PC2, instead it is the cloud volume fraction. To account for clouds not filling the gridbox in the vertical the model uses a diagnostic parameterization to calculate the cloud area fraction. A sensitivity experiment was conducted to test whether a cloud area fraction parameterization was needed for this case. The results from this simulation show that without using the diagnostic parameterization the OLR tends to agree better with observations. This is because in this case the SCM generally over predicts the cloud cover and by not applying the diagnostic scheme the cloud area fraction is not increased further. Significant sensitivity in the PC2 simulated low clouds is demonstrated when the shape of the distribution of water vapour is changed from a top-hat-like shape to a triangular distribution. The reduction that occurs in the low clouds when the triangular PDF is used is in better agreement with the observations. The prognostic scheme is able to simulate more variable cloud fields in agreement with the observations. Both model runs produce supersaturation with respect to ice as was observed in the upper levels during TWP-ICE. The lack of midlevel clouds associated with deep convection in the model has been reported in other studies and is exposed in the TWP-ICE results, however, the magnitude of the problem is obscured by the excessive midlevel convection that occurs for both cloud schemes during the suppressed monsoon phase.

Studies have suggested that the inability of many models to simulate realistic representations of the MJO may be caused by systematic diabatic heating profile errors. Temperature and moisture errors in the SCM simulations were seen to be the most pronounced during the suppressed and break periods. Other studies have identified the link between poor simulations of suppressed convection leading to unrealistic simulations of sub-seasonal variability in tropical convection, including the MJO, and TWP-ICE may provide a good case to study the model biases and make improvements in the model cloud and convection parameterizations. The GEWEX Cloud Systems Study Group is currently setting up a TWP-ICE intercomparison case for both SCMs and CRMs. This experiment will use the forcing and evaluation data set that was used in this study and the outcomes from the high resolution models will enable a more rigorous assessment of the link between the cloud and convection parameterizations in the SCM and the ability of the model to simulate tropical cloud systems. This is intended to build on the results reported in this study and lead to improvements in the physical parameterizations.