-
Views
-
Cite
Cite
Thomas Opitz, Proposer of the vote of thanks to Healy et al. and contribution to the Discussion of ‘Inference for extreme spatial temperature events in a changing climate with application to Ireland’, Journal of the Royal Statistical Society Series C: Applied Statistics, Volume 74, Issue 2, March 2025, Pages 300–302, https://doi.org/10.1093/jrsssc/qlae087
Close - Share Icon Share
Extract
The authors are to be congratulated for providing several useful methodological innovations and raising important questions for the analysis of environmental extreme events against the backdrop of climate change. They develop a nonstationary generative statistical model for spatial temperature extremes, allowing for Monte-Carlo estimation of spatial risk measures. Based on the asymptotic theory for threshold exceedances of spatial risk functionals, the model combines data from irregularly spaced weather stations with simulations of a physical climate model on a regular spatial grid.
Their timely work addresses the general need for comprehensive nonstationary statistical assessment of the frequency, magnitude, and extent of extreme weather. This task is complex since temperature, the key variable of global warming, exhibits strongly heterogeneous trends across three-dimensional space and time. Numerical simulations from physical models provide a wealth of ‘big’ data but come with strong limitations: simulation is deterministic and not probabilistic, and is performed on relatively coarse spatial grids, i.e. not point-based at weather-station level; large biases of simulations with respect to the true climate are possible; computational cost is high and prevents simulating even moderately large numbers of full space-time chronicles and extreme-event catalogues. Instead, the proposed approach transfers information about spatial temperature heterogeneity in sparsely observed areas from physical simulations to a statistical model to obtain a point-based Stochastic Weather Generator (SWG) not suffering from such limitations. It showcases that SWGs are key tools to enhance data provided by physical simulations. They can be calibrated at low computational cost for various purposes: emulation of physical models, downscaling from gridded large-scale input data to point-based distributions, and stochastic simulation of large samples of rare events.
