FAIL (the browser should render some flash content, not this).

The Result is Clear: A Weather Forecast Model has Trouble with Clouds

Ahlgrimm, M. and Forbes, R. 2012. The Impact of Low Clouds on Surface Shortwave Radiation in the ECMWF Model. Monthly Weather Review 140: 3783-3794.
One of the most difficult processes to represent in a model is clouds. While we understand what is needed for the large-scale basics of cloud formation, it is the micro-scale processes and character of clouds that are either poorly understood or not adequately represented. Early schemes in representing precipitation and cloud formation were based on the presence of adequate moisture and the existence of upward motion and instability. The presence of sufficient condensation or ice nuclei was assumed.

As cloud processes have been improved, these are typically done with an eye toward improving numerical weather forecasts. Then, these new parameterization procedures make their way into versions of the model that represent the General Circulation (GCMs), which are then used to derive climate change scenarios. Also, it is known that the general circulation is driven by what are called diabatic processes; they are sensible heating, radiative processes, and latent heating. Radiative processes, and especially their differences over the face of the Earth, are what help to drive the general circulation.

Ahlgrimm and Forbes (2012) investigate an irradiance bias in the European Centre for Medium Range Forecasting (ECMWF) GCM, especially in the Southern Great Plains of the USA. This GCM is built to generate daily numerical weather predictions. In one experiment, the authors compare measured radiation daily from 2004-2009 from the Atmospheric Radiation Measurement Site (ARM) in the Southern Great Plains (SGP). In the second, they compare these to observations of clouds, radiation, and the state of the atmosphere which is archived as the Climate Modeling Best Estimate product from 1997-2009. The authors selected 146 days where days when fair weather cumulus dominated the Southern Great Plains in this experiment. They compared these to the same days run using the ECMWF model initialized the day before and using the 18 - 42 hour forecasts.

The authors show that the mean model radiative bias compared to the observed product was approximately 23 W m-2. When the authors looked at the fair weather cumulus regime and studied the bias over the course of a day, they found that the bias in this regime was weak. This means that paramterizations for fair weather convective cloudiness are largely successful. When the authors attempted to classify clouds they found that the deep clouds (convective), thick mid-level clouds, and the low clouds as a group accounted for over 50% of the bias. As there was no way to account for high cloud and other phenomenon, the remaining bias was just called "residual." They also demonstrated that refinements in the cloud liquid water content and distributions could improve these parameterizations.

Finally, Alhgrimm and Forbes (2012) show that the biases are largest when the observations were cloudy and the model clear, or when the observations were overcast and the model produced broken skies. All other categories showed small biases which largely cancelled each other. As the authors conclude "it will be possible to carry out targeted sensitivity studies to improve cloud occurrence and radiative properties by examining the formulation of the shallow convection trigger, mass transport, and cloud microphysical properties." Additionally, it is obvious that if there is a net positive bias in surface shortwave down, it will result in the overestimation of temperatures in the two situations described above and overall. If these parameterizations, or others which produce warm biases, are used in climate models, the impact on generated climate scenarios would be a net surface warming.

Figure 1. Adapted from Fig. 1 in Ahlgrimm and Forbes (2012). The multiyear (2004-09) all-sky diurnal composite of surface irradiance. The black line is the ECMWF model, and the grey line is the observations from the CMBE product. Only daytime samples (modeled short wave down exceeding 1 W m-2) with good-quality coincident observations are included.

Archived 12 December 2012