Towards a meaningful warning signal metric

  • Most papers do not give a statistical approach, but merely argue for the existence of a trend.
  • Kendall’s tau approach proved very unreliable. While the preliminary de-trending approach is unjustified in that the simulated data is known to be absent from trends, as a kind of windowed averaging it may sense as a way to detect changes.
  • Should distinguish between the gradual slow changes and change-point estimation. Perhaps change-point provides a more rigorous starting point.
  • need to predict the expected variance in the various warning signals, this entry