Time-Varying Dependence Regimes for Multivariate Spatial Processes
Entropy · 07.11.2026 · 28 (11) · 1223
DOI: 10.3390/e28111223 ↗Kurz gefasst
Fachartikel in der Zeitschrift Entropy, online erschienen am 07.11.2026. Verfasst von Lucas Moura Faria e Silva und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Conselho Nacional de Desenvolvimento Científico e Tecnológico, Fundação Carlos Chagas Filho de Amparo à Pesquisa do Estado do Rio de Janeiro.
Abstract (Originalsprache)
Reconstructing and predicting multivariate environmental fields requires models that propagate information coherently across variables, space, and time, while allowing cross-dependence to change. We propose a Bayesian dynamic mixture of conditional Gaussian process models in which each component represents a distinct conditional dependence structure. Time-varying mixture weights governed by a Dirichlet evolution process allow the effective cross-covariance structure to adapt to changing dependence regimes, while predictive screening controls the number of candidate components. The framework is applied to weekly concentrations of nitrogen monoxide, nitrogen dioxide, PM10, and PM2.5 observed at 72 monitoring stations in the United Kingdom. Sparse conditional specifications retain the predictive accuracy of the full conditional model, while the mixture models better accommodate periods of elevated pollutant concentrations. The static mixture provides the best in-sample fit, whereas the dynamic mixture achieves the smallest out-of-sample RMSE, MAE, Energy Score, and Interval Score. Its posterior allocations also exhibit persistent dependence regimes, with structural changes associated with episodes of elevated pollution. These results demonstrate that dynamic mixtures of conditional Gaussian processes provide a parsimonious and interpretable framework for multivariate spatial prediction under temporally evolving dependence.
Förderung
- Conselho Nacional de Desenvolvimento Científico e Tecnológico
- Fundação Carlos Chagas Filho de Amparo à Pesquisa do Estado do Rio de Janeiro
Zitieren
APA
Silva, L. M. F. E., & Fonseca, T. C. O. D. (2026). Time-Varying Dependence Regimes for Multivariate Spatial Processes. Entropy, 28(11), 1223. https://doi.org/10.3390/e28111223
Vancouver
Silva LMFE, Fonseca TCOD. Time-Varying Dependence Regimes for Multivariate Spatial Processes. Entropy. 2026;28(11):1223. doi:10.3390/e28111223
BibTeX
@article{Silva2026,
author = {Silva, Lucas Moura Faria e and Fonseca, Thaís Cristina Oliveira da},
title = {Time-Varying Dependence Regimes for Multivariate Spatial Processes},
journal = {Entropy},
year = {2026},
volume = {28},
number = {11},
pages = {1223},
doi = {10.3390/e28111223}
}
EUVIMED