Evaluating Community Atmospheric Model (CAM) forecasts against satellite data along the GCSS Pacific cross-section

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 26 January 2011: 5:15 PM
Evaluating Community Atmospheric Model (CAM) forecasts against satellite data along the GCSS Pacific cross-section
608 (Washington State Convention Center)
Cecile Hannay, NCAR, Boulder, CO; and D. Williamson, R. Neale, J. Olson, and D. Shea

We use short-term forecasts along the GCSS Pacific cross-section to evaluate parameterizations in the Community Atmospheric Model versions 3, 4 and 5 and highlight major improvements in the recent releases.

Climate models are routinely validated against various time-mean statistics based on observations. However, it is difficult using this validation method to determine whether a reasonable climate simulation is achieved due to the correct interaction of processes or due to compensating errors, which are difficult to untangle in time-mean diagnostics. It is also problematic when trying to attribute any mean biases to a particular parameterized process. An innovative method to evaluate parameterizations in climate models is to use the Cloud-Associated Parameterization Testbed (CAPT) approach where the state of the atmosphere is initialized using realistic conditions and the model is run in a series of short-term forecasts. This approach allows a direct comparison of the parameterized variables (e.g. clouds, precipitation, radiation) with the time varying observations. Therefore, it is possible to gain insight into any parameterization deficiencies and to diagnose the processes responsible for the drift away from observed.

Our focus for this work is the GCSS cross-section case study region that runs from the stratocumulus regions off the coast of California, across the shallow convection dominated trade winds, to the deep convection regions of the ITCZ. It is particularly relevant for such a comparison because it includes several important cloud regimes and their interactions through the large-scale circulation.

Results are presented from JJA 2008 and take advantage of the availability of state-of-the-art analyses to initialize the model and a wealth of integrated observational datasets to evaluate the forecasts. The forecasts are initialized from analyses (ECMWF-YOTC and GMAO-MERRA) and the CAM is run in a number of 5-day forecasts throughout the whole season and evaluated against A-train, AIRS, TRMM, SSM/I, ISCCP and CERES satellite-based products.

In CAM, the mean forecast biases grow very quickly along the cross-section, and after 5 days, the error pattern is very similar to the mean climate error. Around the ITCZ, most of the temperature and moisture errors have developed after a single day reflecting the fast processes associated with deep convection. Over the stratocumulus region, the error grows more slowly and it takes 5 days before the mean forecast error reaches the amplitude of the mean climate error.

The CAM3 significantly overestimates the temperature along the Pacific cross-section. These warm biases are attributable to deep convection errors and their propagation to other cloud regimes. In CAM4, the local tropical errors are reduced and grow much more slowly as a result of a modified deep convection scheme. This leads to a dramatic improvement of the precipitation in the ITCZ region. In association with the adjusted atmospheric state, the biases are also reduced away from the region of significant deep convection (i.e. in the shallow cumulus and stratocumulus regimes). This clearly illustrates the interaction between the tropical convection and the large-scale circulation. Similar forecasts with CAM5, which includes a two-moment microphysics scheme, show further reduction in upper troposphere errors. Additionally, a more realistic representation of cloud-topped boundary layers deepens the boundary layer in stratocumulus region and produces a more accurate location for low-level clouds.