As is customary in this field, Ghil and his colleagues used five decades of climate data and test predictions retrospectively. For example, they used climate data from 1950 to 1970 to make "forecasts" for January 1971, February 1971 and beyond and see how accurate the predictions were. They reported achieving higher accuracy in their predictions 16 months out than other scientists achieved in half that time.
Ik ben vooral benieuwd hoe het model dan scoort in de toekomst. Reanalysis is altijd heel nuttig, maar als je onbekende fysische of niet fysische parameters kan veranderen om je score hoog te krijgen (regressieananalyse). Dan geldt altijd: in het verleden behaalde resultaten bieden...
