ECO covers the coordination of the Module E; it integrates the individual evaluation efforts and organizes scientific exchanges with the other modules and WPs in MiKlip II. Stakeholder/end-user interaction required for the orientation of the evaluation system to end-user needs will be ensured in cooperation with module D, involving individual work packages as needed.
Beside the coordination of Module E, ECO has following scientific contributions:
As bias and drift correction become more relevant for all validation activities, ECO will implement a drift correction framework based on generalized linear models (GLMs). The framework includes a drift correction which is initialization time dependent. It shall be developed and implemented into the Central Evaluation System (CES). Together with WP E-6 DROUGHTCLIP, the GLM-based framework will be extended to skewed and positive variables, to be particularly suited for precipitation (WP E-2 DAPAGLOCO, WP E-5 PROMISA), humidity (WP E-1 MOSQUITO) or wind. Furthermore, calibration of probabilistic forecasts to increase reliability is an important issue in MiKlip II. Different calibration approaches will be developed in Module E within WP E-8 PROVESIMAC and WP E-9 CALIBRATION. Together with bias and drift correction, calibration of forecast will contribute to general post-processing methodologies which will be coordinated by ECO. Predictive skill is assumed not only to vary with lead time but also with season and/or with the phase of low-frequency climate modes, like Atlantic multidecadal variability (AMV), Atlantic meridional overturning circulation (AMOC), and Pacific decadal oscillation (PDO). Skill estimation stratified along these influences will be a valuable contribution to the general skill estimation but also for processoriented validation. Therefore ECO will implement CES plug-ins for stratified verification.
Within the framework of ECO, a drift adjustment plug-in for the Central Evaluation System (CES) has been implemented, which is flexible in the choice of the correction method. Model drift can be estimated and adjusted parametrically by means of generalized linear models, or non-parametrically.
The sensitivity of the prediction skill and improvement of the parametric bias adjustment has been tested concerning cross validation (CV) with different block length.
Freie Universität Berlin, Institute for Meteorology
Prof. Dr. Uwe Ulbrich
Dr. Jens Grieger
Babian, S. | H.W. Rust, J. Grieger, K. Prömmel and U. Cubasch