Methodology for Statistical Detection of Climate Change

Many detection or attribution studies are applications of the "optimal fingerprints" method, which is based on the optimization of a signal-to-noise ratio. One of the difficult point is that this method requires to know, or, in practice, to estimate the covariance matrix of the internal climate variability. In this work, a new adaptation of the "optimal fingerprints" method is presented for climate change detection. This adaptation is based on the use of a regularized estimation of the covariance matrix, that avoid to truncate it to a reduced dimension space. This technique presents two main advantages. On the one hand, under some acceptable statistical hypothesis, it can be shown to yield to a more powerful detection test. On the other hand, the covariance estimation is still efficient when the number of years used for covariance estimation is close to the number of observed time series. In order to validate the efficiency of the detection algorithm, it is first applied with pseudo-observations derived from transient regional climate change scenarios covering the 1960-2099 period. Then it is used to perform a detection study of anthropogenic climate change over France, analyzing homogenized temperature and rainfall series from 1900, produced at Météo France. The new methodology allows to estimate the covariance matrix only using part of the observation dataset. This new approach allows to confirm and reinforce the detection of an anthropogenic climate signal over the country.