Scoring extreme events forecasts with the Continuous Ranked Probability Score
Ecole des Mines, Paris, France
The Continuous Ranked Probability Score (CRPS) is commonly used for scoring forecasts. In order to assess the quality at predicting very large values, we try to modify it and apply it to a Generalized Pareto Distribution, used to describe the behaviour of extremes in the Peak Over Threshold (POT) case. This preliminary work will be followed by an application to extreme events datasets, still to be found.
Hélène Benveniste, currently finishing her master degree at the Ecole des Mines in Paris, works since November with Philippe Naveau (Laboratoire des Sciences du Climat et de l'Environnement, France) and is staying until June in Colorado, working with Dan Cooley from Colorado State University and Rick Katz from NCAR.
Monday, March 26, 2012
Mesa Laboratory, Chapman Room
Lecture at 11:00am