B0553
Title: D-vine GAM copula based quantile regression with application to ensemble postprocessing
Authors: David Jobst - University of Hildesheim (Germany) [presenting]
Juergen Gross - University of Hildesheim (Germany)
Annette Moeller - Bielefeld University (Germany)
Abstract: The D-vine (drawable vine) copula quantile regression (DVQR) is a powerful method in the field of ensemble postprocessing as it can automatically select important predictor variables from a large set, and is able to model complex nonlinear relationships among them. However, the current DVQR does not always explicitly and economically allow to take into account covariate effects, e.g. temporal or spatiotemporal effects. Therefore, an extension of the current DVQR is proposed, where the bivariate copulas are parametrized in the D-vine copula through Kendall's $\tau$ which is linked to covariates using generalized additive models (GAMs) and spline smoothing. Therefore, the introduced model is called GAM-DVQR. In a case study for the postprocessing of 2m surface temperature forecasts, a constant as well as a time-dependent Kendall's $\tau$ are investigated. The GAM-DVQR models are compared to the benchmark methods ensemble model output statistics (EMOS), its gradient-boosted extension (EMOS-GB) and DVQR. The results show, that the GAM-DVQR models identify time-dependent correlation as well as relevant predictor variables very well, and significantly outperform the state-of-the-art methods EMOS and EMOS-GB. Furthermore, using a static training period for GAM-DVQR yields a more sustainable model estimation in comparison to DVQR using a sliding training window.
