Climate Change Attribution Using Empirical Decomposition of Climatic Data

It can be argued that estimation of periodic signals in climate data requires either spectral methods or very long time series, or both. For obtaining a very precise estimate of signal properties this may be true, but for estimating periodicity to within plus or minus a few years, this is not true. To illustrate, we replicate the procedure used in the main manuscript. We take the model from Fig. (1) as the target. A sample set is obtained by randomly adding white noise to this model with variance derived from Fig. (1b, (sd = 0.13°C). The new model is estimated for the period 1850 to 1950 and then extrapolated to 2009 (see Fig. Suppl 1). Repeating this 1000 times, the mean estimated period is 64.11 yr vs 64.93 for the target, with sd = 2.63 yr. Thus, clearly the proposed estimation method works sufficiently well even on a 100 yr calibration period.