Task 2.3 Verification of extremes and multivariable indices
Task 2.3 will extend the existing evaluation strategy to extremes, and evaluate multivariable, userspecific indices for sectoral applications identified as relevant in Task 2.1 and implemented in Task 2.2., and evaluate the added value of improved initialization in this regard. In addition to standard verification, we will evaluate scores tailored to the tails of the predictive distribution using their decomposition into reliability and resolution (Bröcker, 2009; Bentzien and Friederichs, 2014). Deficiencies are addressed in Task 2.2 to improve reliability for the tails of the distributions as well. Since predictive skill for extreme events at the ICON grid level is obscured by high levels of random noise, we explore spatial aggregation methods ranging from spatial averages to typical spatial variability patterns for extremes (e.g., Szemkus and Friederichs, 2023+). Many user-specific indices (e.g., drought indices such as SPEI, renewable energy production indices, or human comfort indices) are derived from multiple variables. We implement multivariate scores for the evaluation to support the consistent calibration of complex indices in Task 2.2 and evaluate the relevance of dependence structure for the derivation of calibrated indices.
Because forecast quality is an important element of forecast communication and can change with "climate state," we evaluate forecasts as a function of specific initialization states, e.g., positive or negative NAO states, in a stratified verification approach (Richling, in preparation). To evaluate the added value of initialization and their respective enhancements in Task 1.2, we will follow the work of Glowienka-Hense et al. (2020). We will extend this approach to examine the added value of improvements in quantile predictions using, for example, relative entropy measures for quantiles based on the asymmetric Laplace distribution (Yu and Moyeed, 2001).
Contributors
University of Bonn
Priv. Doz. Dr. Petra Friederichs
MSc. Philipp Ertz
Freie Universität Berlin
Prof. Dr. Henning Rust
MSc. Andy Richling
MSc. Felix Fauer
Deutscher Wetterdienst
Dr. Clementine Dalelane
Dr. Andreas Paxian