statistical models


based on observed relationships between climate variables and the exposure unit are of considerable value in climate vulnerability and impact assessment. Statistical models are particularly valuable in situations where long time series or large spatial data sets are available that include the key climate and system response variables. Very often statistical models do not distinguish cause and effect within the structure of the model, and hence can be unreliable when extrapolating to new sites or conditions (e.g. future climates) that may differ markedly from those historical climates on which the model was based. However, statistical models can be constructed that explicitly incorporate knowledge of the causal relationships between variables, and such models provide a higher degree of confidence. Statistical models can, in general, be regarded as having a lower pedigree than process-based models. (UKCIP, 2003)