Some computer codes for numerical weather prediction (see the NWP chapter) are designed to forecast global climate. These codes, called GCMs (global climate models) have a somewhat-coarse 3-D arrangement of grid points filling the global atmosphere. At each grid point the governing equations for heat, momentum, mass, moisture, and various chemicals are repeatedly solved as the GCM iterates forward in time.
Other GCMs use spectral rather than grid-point numerics in the horizontal, where a sum of sine waves of various wavelengths are used to approximate the spatial structure of the climate. Regardless of the numerical scheme, the GCM is designed to forecast decades or even centuries into the future.
Such a long-duration forecast is computationally expensive. Improved climate forecasts are possible by coupling forecasts of the atmosphere, ocean, biosphere, cryosphere (ice) and soil — but this further increases computational expense. To speed up the forecast, compromises are made in the numerics (e.g., coarse grid spacing in the horizontal; reduced number of vertical layers), dynamics, or physics (e.g., simplified parameterizations of various feedbacks).
One disadvantage of coarse grid spacings is that some dynamical and physical phenomena (e.g., thunderstorms, thermals, clouds, etc.) become subgrid scale, meaning they are not directly resolvable. As an example, a GCM with horizontal grid spacing of 1° of latitude would be able to resolve phenomena as small as 7° of latitude (≈ 777 km); hence, it could not “see” individual thunderstorms or clusters of thun derstorms. Although unresolved, subgrid-scale thunderstorms still affect the global climate forecast. Therefore their effects must be approximated via parameterizations.
“We tend to predict impacts on people by assuming a static human society. But humans are adapting quickly, and time scales of human change are faster than the climate change. There’s no unique perfect temperature at which human societies operate. Run them a little warmer, and they’ll be fine; run them a little cooler, they’ll be fine. People are like roaches — hard to kill. And our technology is changing rapidly, and we’re getting richer, internationally and individually, at an absurd rate. We’re rapidly getting less and less sensitive to climatic variations.”
– David Keith (from an interview published in Discover, Sep 2005.)
Even for resolvable phenomena such as midlatitude cyclones, GCMs are designed to forecast the correct number of cyclones, although their locations and dates are usually incorrect (Table 21-5). The motivation is that the correct number of cyclones will transport the correct amount of heat, moisture and momentum when averaged over long time periods, thus increasing the likelihood that the climate simulation is reasonable.
|Table 21-5. Both OPMs (Operational Prognostic Models, for short-range daily forecasts) and GCMs (Global Climate Models, for long-range simulations) are NWP (Numerical Weather Prediction) models.|
|desired outcome||deterministic forecasts (of the actual weather)||forecasts that are statistically skillful (e.g., forecast the right number of cyclones, even though their locations and times are wrong)|
|horizontal grid spacing||10 m to 10 km||10 km to 100 km|
|boundary conditions||less important||more important|
|initial conditions||very important||less important|
|approximations||partially parameterized||extensively parameterized|
Other phenomena, such as solar input or deepsoil temperatures, are imposed as boundary conditions (BCs). By very careful prescription of these BCs, the GCM can be designed so that its numerical climate does not drift too far from the actual climate. This is often validated by running the GCM for past years and comparing the result to the observed climate, giving confidence that the same accuracy is possible for future years.
Initial conditions (ICs) are less important for GCMs, because the ability to forecast the exact location of individual cyclones decreases to nil within a week or so (see the Forces and Winds Chapter).
GCMs are often used to address what-if questions, such as “what if the concentration of CO2 were to double”, or “what if deforestation were to reduce vegetation coverage.” By comparing GCM runs with and without CO2 doubling or deforestation, we can estimate the effects of such changes on future climate. But such predictions carry a lot of uncertainty because of the many parameterizations and approximations used in GCMs.