Climate and weather services rely, for a large part, on data obtained from in situ measurement stations. For this purpose, the Netherlands operates an official network of high-fidelity stations. However, over recent years, data has become available from multiple lower-fidelity observation networks. We are currently investigating how this data, of different fidelity levels, can be combined in a single service. As in multi-fidelity surrogate-based vehicle optimisation, Gaussian process regression (or Kriging) is a key method in our field of application, and in both vehicle design optimisation and meteorological services we are challenged by similar issues in the efficient application of such multi-fidelity surrogates. In this work, we focus on quality control, noise treatment, inclusion of high-resolution covariates, and improvement of the reliability of the predicted local uncertainty, all in the context of spatial regression of multi-fidelity data. DISCLAIMER: KNMI is continuously researching possible ways to improve services. The methods and results in this work should be considered as fit for research purposes, not as part of official KNMI services. After sufficient research, KNMI aims to make new methods operational as soon as possible.
Jouke de Baar, Irene Garcia-Marti. Recent improvements in spatial regression of climate data
Journal: NATOAVT-354 workshop on multi-fidelity methods for military vehicle design, Year: 2022, doi: http://dx.doi.org/10.13140/RG.2.2.25417.52325