7.2.5 Evaluation of Models Through Intercomparison
Intercomparison of vegetation models usually involves comparing surface fluxes and their feedbacks. Henderson-Sellers et al. (2003), in comparing the surface fluxes among 20 models, report over an order of magnitude range among sensible fluxes of different models. However, recently developed models cluster more tightly. Irannejad et al. (2003) developed a statistical methodology to fit monthly fluxes from a large number of climate models to a simple linear statistical model, depending on factors such as monthly net radiation and surface relative humidity. Both the land and atmosphere models are major sources of uncertainty for feedbacks. Irannejad et al. find that coupled models agree more closely due to offsetting differences in the atmospheric and land models. Modelling studies have long reported that soil moisture can influence precipitation. Only recently, however, have there been attempts to quantify this coupling from a statistical viewpoint (Dirmeyer, 2001; Koster and Suarez, 2001; Koster et al., 2002; Reale and Dirmeyer, 2002; Reale et al., 2002; Koster et al., 2003; Koster and Suarez, 2004). Koster et al. (2004, 2006) and Guo et al. (2006) report on a new model intercomparison activity, the Global Land Atmosphere Coupling Experiment (GLACE), which compares among climate models differences in precipitation variability caused by interaction with soil moisture. Using an experimental protocol to generate ensembles of simulations with soil moisture that is either prescribed or interactive as it evolves in time, they report a wide range of differences between models (Figure 7.2). Lawrence and Slingo (2005) show that the relatively weak coupling strength of the Hadley Centre model results from its atmospheric component. There is yet little confidence in this feedback component of climate models and therefore its possible contribution to global warming (see Chapter 8).