‘Predictions are particularly difficult when referring to the future.’ (Mark Twain)
Several factors contribute to the uncertainty in climate projections.
Emissions scenarios: Climate change projections are based on scenarios of future demographic, economic, technological and political development and their impact on greenhouse gas emissions. Since mankind can follow multiple paths, we do not know for sure which scenario might become the future. This is the most important source of uncertainty for climate change projections. To assess different possible ‘futures’ a set of greenhouse gas emissions scenarios was developed (SRES scenarios). Often different scenarios are simulated to get an idea of what the future might be depending on which path is taken.
Model errors: Climate models are not perfect. Some interactions in the climate system might not be discovered yet and we can’t completely simulate all the things we currently know in a climate model. However, simulation results for past climate periods generally compare well with observations. Under the assumption that model errors differ from model to model and are in principle statistically distributed, a multi-model mean is often used to average out individual models errors.
Insufficient knowledge of the starting conditions: We can only determine the current situation of the climate system to a certain degree. Particularly the precise state of the deep ocean is not well known. Thus, climate simulations are often run in an ‘initial conditions ensemble’ mode. In this case, several simulations for historical times are started from different times in a control-run, which is forced by constant conditions for a long time.
Regional uncertainty: Global trends are relatively unambiguous, even though their magnitude differs. However, regional signals tend to differ from model to model. Additionally, different downscaling techniques (e.g. the use of different regional climate models) can lead to (mostly quantitatively, not qualitatively) different results. The approach is again to use ensembles of different models and downscaling techniques to estimate the possible regional climate change signal.