1) Spatial-scale incompatibility
The coarse horizontal and vertical resolution of the GCMs result in crude
estimations of the small-scale rainfall processes (Hewitson and Crane, 1992). For
example, convective cells have a characteristic scale of the order 50 km2. A
model grid box may be around 500 km2. Simulations are further hampered
when rainfall events are influenced by local topography. As a result, the rainfall
at a grid point is difficult to compare with the rainfall measured at a
meteorological station (von Storch et al., 1993).
2) Intermodel agreement
The quantitative agreement between different models is much better for
temperature than for rainfall. The first reason is that precipitation is changed
indirectly by a wider variety of different processes. Warming is primarily a direct
response to increased radiative heating. The second reason is that the simulated
changes in precipitation are relatively small compared with the natural
variations observed in time series, and are thus more difficult to detect. Also,
differences between models in CO2 doubling times, horizontal and vertical
resolution, correction of air-sea fluxes, and parametrisations lead to differences in
the GCM predictions.
