Sensitivity of model climate to sampling configurations and the impact on the extreme forecast index

Ervin Zsoter*, Florian Pappenberger, David Richardson

*Corresponding author for this work

Research output: Contribution to journalArticle (Academic Journal)peer-review

18 Citations (Scopus)


The Extreme Forecast Index (EFI) compares a medium range ensemble forecast (ENS) to climatology to establish the severity of an event. In this paper, the large-scale influence of the design and configuration of this climatology on the climate distribution and the EFI is examined. The design factors considered are ensemble size (how many ensemble members), sampling frequency (how often per week), lead-time dependency (from which lead time to draw the climatology), and the sampling length (number of years). The results show that most of the climate configurations with sample size above 50-100 give a relatively good representation of the climate distribution between the 10th and 90th percentiles. For more extreme percentiles (for 1-10% and 90-99%) some care is needed to select the best configuration, whereas the extreme tails of the distribution cannot be well represented by any of the investigated configurations (up to a 14year, daily, 50 member climate), for which a bigger sample would be needed. Results show that the hindcast length is clearly the superior sampling factor in general, better than the sampling frequency and the ensemble size. In addition, the ensemble size has only very limited contribution to the non-extreme climate percentiles and the EFI, especially at short range.

Original languageEnglish
JournalMeteorological Applications
Publication statusAccepted/In press - 29 Apr 2014


  • Climate distribution
  • Ensemble forecasts
  • Extreme events
  • Model climatology
  • Sample size

Fingerprint Dive into the research topics of 'Sensitivity of model climate to sampling configurations and the impact on the extreme forecast index'. Together they form a unique fingerprint.

Cite this