Developing long-term homogenized climate Data sets Olivier Mestre Météo-France Ecole Nationale de la Météorologie Université Paul Sabatier, Toulouse
The introduction you ever dreamed of…
« State of fear » (Michael Crichton)
Homogenisation : why? Example of Pau temperature series 1912 : Lescar primary school 2007 : Pau-Uzein Airport
Pau: raw maximum temperatures (TX)
Homogenisation : a very old problem! « Comptes-rendus de l’Académie Royale des Sciences »
Usual method: relative homogeneity PRINCIPLE : removing the climatic signal to put into evidence artificial shifts in the series minus Tested series Reference series COMPARISON series
Shifts detection Dynamic programming algorithm + penalized likelihood Multiple comparisons of non-homogeneous series Metadata!
Shifts detection
Correction ANOVA model : correction of multiple non-homogenous series, provided change-point positions are well known. µiµi Climate factor + Station factor + Noise j1 j2 j3 j4 j5
Correction Climate signal estimation + Bias estimation in the station effects (monthly scale) Correction+reconstitution of missing data Absolutely no assumption is made concerning the evolution of the climate signal
Correction of Pau maximum temperatures « Before » « After »
Maximum temperatures : trends « Before » « After »
Developments in Homogenisation COST ACTION ES0601 : « Advances in HOmogenisation MEthods for climate series : an integrated approach » (HOME) Daily data homogenisation : study of extreme events
Requirements in terms of data digitization Fill the gaps and complete the target series as far as possible Quality control and homogenisation techniques require complete neighbouring series : digitize every data, not only target series! Metadata, station histories are as important as data itself Digitize metadata along with corresponding data!