Uncertainty quantification and spatial statistics

Group leader: Prof. Dr. David Ginsbourger

Spatial statistics essentially deals with the description and modelling of events occuring accross multivariate spaces, and has been used in a variety of domains from geosciences and beyond. Models from spatial statistics have been leveraged within the blossoming field of Uncertainty Quantification in order to reconcile observational and simulation data towards enhanced probabilistic predictions and targeted sequential data acquisition strategies. We investigate random field models as well as distance and kernel methods for prediction and sequential design algorithms dedicated to optimization, set estimation, and further inverse problems. Application domains include environmental geosciences, reliability and safety engineering, atmospheric sciences, agronomy, robotics, machine learning, and, increasingly, biomedical sciences.


The group is involved in a number of collaborations with national and international partners from various research fields, including geophysics (notably with Prof. Niklas Linde’s group at UNIL) and climate sciences with various colleagues from the Oeschger Center of Climate Change Research. Additionally, together with Prof. Ziegel, joint work in being conducted with the Federal Office for Meteorology and Climatology (Meteoswiss), with main foci on forecast postprocessing and warning evaluation. On a different note, a collaboration is ongoing with the Swiss Confederation’s centre of excellence for agricultural research (Agroscope). From the biomedical side, the group has been collaborating with the University Psychiatric Services (UPD), and a new collaboration is emerging with Inselspital and ISPM around statistical machine learning for endocrinological gynaecology. 

Selected Publications

  1. A. Gautier, D. Ginsbourger. Continuous logistic Gaussian random measure fields for spatial distributional modelling. [arXiv:2110.02876]
  2. C. Travelletti, D. Ginsbourger, N. Linde. Uncertainty Quantification and Experimental Design for large-scale linear Inverse Problems under Gaussian Process Priors. [arXiv:2109.03457]
  3. D. Ginsbourger and C. Schaerer. Fast calculation of Gaussian Process multiple-fold cross-validation residuals and their covariances. [arXiv:2101.03108]
  4. D. Azzimonti, D. Ginsbourger, C. Chevalier, J. Bect, Y. Richet. Adaptive Design of Experiments for Conservative Estimation of Excursion Sets. Technometrics 63(1): 13-26, 2021. [DOI:10.1080/00401706.2019.1693427]
  5. P. Buathong, D. Ginsbourger, and T. Krityakierne. Kernels over sets of finite sets using RKHS embeddings, with application to Bayesian (combinatorial) optimization. Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, PMLR 108, 2731-2741, 2020. [PMLR-proceedings]
  6. J. Bect, F. Bachoc, and D. Ginsbourger. A supermartingale approach to Gaussian process based sequential design of experiments. Bernoulli, 25(4A):2883-2919, 2019. [DOI:10.3150/18-BEJ1074]
  7. D. Ginsbourger, O. Roustant, and N. Durrande. On degeneracy and invariances of random fields paths with applications in Gaussian process modelling. Journal of Statistical Planning and Inference, 170:117–128, 2016. [DOI:10.1016/j.jspi.2015.10.002]

Group members