Evaluating Forecasts for High-Impact Events Using Transformed Kernel Scores
SIAM/ASA Journal on Uncertainty Quantification · 30.09.2023 · 11 (3) · 906-940
DOI: 10.1137/22m1532184 ↗Kurz gefasst
Fachartikel in der Zeitschrift SIAM/ASA Journal on Uncertainty Quantification, online erschienen am 17.08.2023. Verfasst von Sam Allen und 2 weiteren Autor:innen. Die Arbeit wurde bisher 16-mal zitiert (Crossref) und nennt 42 Literaturangaben. Gefördert von: Swiss Federal Office of Meteorology and Climatology, Oeschger Centre for Climate Change Research.
Abstract (Originalsprache)
To account for uncertainties, forecasts for future events are commonly expressed in terms of probability distributions over the set of possible outcomes. To evaluate the quality of such forecasts, it is customary to employ proper scoring rules, which provide an objective framework with which to compare forecasters. When evaluating forecasts, it may be useful to emphasize outcomes that have a high impact on the forecast user. Weighted scoring rules have become a well-established tool to achieve this. However, while impacts may result from the interaction of events along multiple dimensions, weighted scores have been studied almost exclusively in the univariate case. This paper introduces weighted multivariate scoring rules, which can emphasize multivariate outcomes of interest when evaluating forecast performance. We demonstrate that the threshold-weighted continuous ranked probability score (twCRPS), arguably the most well-known weighted scoring rule, belongs to the class of kernel scores, and we use this to construct univariate and multivariate generalizations of the twCRPS. These generalizations include weighted versions of popular multivariate scoring rules, such as the energy score and variogram score. This result facilitates the introduction of weighted scoring rules that are appropriate in many practical applications, which also enjoy the theoretical framework associated with kernel scores. To illustrate the additional information that these novel weighted scoring rules provide, results are presented for a case study in which they are used to evaluate daily precipitation accumulation forecasts, with particular emphasis on events that could lead to flooding.
Förderung
- Swiss Federal Office of Meteorology and Climatology
- Oeschger Centre for Climate Change Research
Zitieren
APA
Allen, S., Ginsbourger, D., & Ziegel, J. (2023). Evaluating Forecasts for High-Impact Events Using Transformed Kernel Scores. SIAM/ASA Journal on Uncertainty Quantification, 11(3), 906-940. https://doi.org/10.1137/22m1532184
Vancouver
Allen S, Ginsbourger D, Ziegel J. Evaluating Forecasts for High-Impact Events Using Transformed Kernel Scores. SIAM/ASA Journal on Uncertainty Quantification. 2023;11(3):906-940. doi:10.1137/22m1532184
BibTeX
@article{Allen2023,
author = {Allen, Sam and Ginsbourger, David and Ziegel, Johanna},
title = {Evaluating Forecasts for High-Impact Events Using Transformed Kernel Scores},
journal = {SIAM/ASA Journal on Uncertainty Quantification},
year = {2023},
volume = {11},
number = {3},
pages = {906--940},
doi = {10.1137/22m1532184}
}
EUVIMED