Climatologists use the terms "detection" and "attribution" when using statistical methods to determine if an expected "signal" is present within highly variable ("noisy") climate observations. The signal in the statistical exercise often comes from a climate model simulation driven by changes in external forcing (for example, changes in greenhouse gas concentrations, sulphate aerosols, volcanoes, solar output, and/or land-use). The signal is a combination of the trend and spatial pattern change in some climate variable (usually surface air temperature). If this signal is statistically present in the observational climate data, we can claim to have detected the influence of the forcing in the observational data. If we can additionally rule out all other possible forcing combinations (using additional modelling experiments) as the cause, then we can claim to have attributed the observed change to a known combination of causal factors or, rarely, to a single factor.
The example shows plots of observed global mean annual temperature anomalies and anomalies from several climate models. The simulated temperature anomalies came from models driven by anthropogenic and natural forcing (pink line) and models driven only by natural forcing (blue line). The "fat" lines for the model data illustrate that, although they are in general agreement, model results cover a range of values reflecting differences in model formulation etc.. To a first approximation, this range can be interpreted as the error in the modelled climate. The example shows that we have detected the influence of human activity in the warming shown by the observations, and the differences between the plots of model data clearly attribute the warming to anthropogenic forcing. The temporary decline in temperatures in the 1940s simply demonstrates that human activity is not the only factor affecting climate.