Score-optimized copula coupling for multivariate ensemble post-processing
Abstract
Statistical post-processing can calibrate the individual components of an ensemble forecast but may also distort relationships between forecast dimensions. This paper proposes Score-Optimized Copula Coupling, which rearranges calibrated forecast samples according to an optimized dependence structure to restore potentially lost dependencies. We assess fixed and adaptively updated versions using wind-speed and temperature forecasts. In the variable-specific spatial case studies, all versions significantly outperform the strongest benchmarks, with improvements of 1.08-2.02 percent in energy score and 5.45-7.97 percent in variogram score. A third case study at 24 co-located stations shows that one optimized structure can represent both spatial and inter-variable relationships. The largest energy and variogram score improvements are 3.98 and 3.69 percent for spatial dependence and 1.16 and 1.06 percent for inter-variable dependence; all are statistically significant. Additional calibration terms improve covariance representation but can reduce gains in the other scores.
Related
- Vector Vine Copula Models for Multivariate Longitudinal Data
- Predicting continuous load-settlement trajectories of bored piles: A GA-optimized weighted ensemble framework with explainable AI
- Particle-Number-Preserving Quantum Signal Processing for Thermal Properties: Canonical-Ensemble Estimation and Grand-Canonical Reconstruction
- When is multivariate kriging worthwhile? Design geometry, parameter sharing and response coupling under heterotopic sampling
- Copula Operad and Copula Entropy
BibTeX
@misc{lakatos2026scoreoptimized,
title = {{Score-optimized copula coupling for multivariate ensemble post-processing}},
author = {Mária Lakatos},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.24591},
eprint = {2610.24591},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.24591},
}
RIS
TY - UNPB
TI - Score-optimized copula coupling for multivariate ensemble post-processing
AU - Mária Lakatos
PY - 2026
DA - 2026/10/23
PB - arXiv
DO - 10.48550/arxiv.2610.24591
UR - https://arxiv.org/abs/2610.24591
AB - Statistical post-processing can calibrate the individual components of an ensemble forecast but may also distort relationships between forecast dimensions. This paper proposes Score-Optimized Copula Coupling, which rearranges calibrated forecast samples according to an optimized dependence structure to restore potentially lost dependencies. We assess fixed and adaptively updated versions using wind-speed and temperature forecasts. In the variable-specific spatial case studies, all versions significantly outperform the strongest benchmarks, with improvements of 1.08-2.02 percent in energy score and 5.45-7.97 percent in variogram score. A third case study at 24 co-located stations shows that one optimized structure can represent both spatial and inter-variable relationships. The largest energy and variogram score improvements are 3.98 and 3.69 percent for spatial dependence and 1.16 and 1.06 percent for inter-variable dependence; all are statistically significant. Additional calibration terms improve covariance representation but can reduce gains in the other scores.
ER -