The hindsight method was tried with following approach:

  • Smooth the temperature anomalies by centring the moving average of 13 consecutive monthly data
  • Define the period 1856-1992 as the learning period for which the best possible regression formula was sought.
  • Use the observations made between 1993 and 2013 to verify if the formula obtained for 1856-1992 enables the reproduction of later data.

The result is shown below:

ta reg verify
Regression analysis 1856 to 1982 and verification 1983 to 2013

Despite of the lesser correlation fitness due to a shorter dataset, this hindsight prediction has an unprecedented quality. Even the so-called pause since the end of last century is reproduced (slightly delayed).