How do scientists project climate change decades into the future—for example, 50 or 100 years from now? Among the most powerful tools for assessing future climate change are coupled atmosphere–ocean general circulation models, commonly referred to as global climate models. These models represent physical processes, atmospheric circulation, and the dynamic behaviour of Earth's atmosphere and oceans.
Future emissions scenarios
 |
Global climate models are run under different scenarios representing possible future greenhouse-gas emissions. These scenarios incorporate assumptions related to factors such as global economic growth, world population, technology, and even social awareness. The figure above illustrates several major scenarios and their corresponding influence on projected changes in global mean annual temperature.
Testing a model before using its projections
Before a global climate model is relied upon for future projections, its skill must be evaluated. The purpose of validation is to determine whether the model can reproduce climate information with an acceptable degree of reliability.
One approach is to run the model for a historical baseline period for which real observations are already available. The model simulates that past period, and researchers then compare statistical characteristics of its output with those of the observed data using appropriate tests and performance measures.
If the simulated and observed statistical characteristics agree sufficiently, the model's skill for the intended application is considered acceptable. Researchers can then run the model using a selected future greenhouse-gas-emissions scenario and evaluate the resulting climatic changes relative to the historical baseline.
The spatial-resolution problem
Global climate models typically simulate meteorological variables on relatively coarse grids. The source gives an approximate spatial scale of 100–300 kilometres. Using such coarse output directly as input to another model—for example, a hydrological model—can increase uncertainty when local or regional detail is required.
To obtain finer-scale information, researchers use downscaling. Downscaling methods are broadly divided into statistical and dynamical approaches. The original article notes that statistical techniques are widely used because they can be faster and simpler to implement than dynamical downscaling.