Spatial analysis of extreme hourly and daily precipitation return levels – Borrowing information over space in spite of complex topography

Knowledge of the magnitude of rare precipitation extremes is important for civil protection and infrastructure design. But the preparation of related datasets is methodologically complex. The estimation of, say, the 100-year return value at a location without measurement crucially depends on how much information can be drawn from instrumented locations nearby. Are there prospects for this “borrowing information over space”, when the climate of the region is spatially complex? In this presentation we illustrate that a method of spatial extreme value analysis has great potential, if it can be carefully configured and equipped with climatological knowledge. We also illustrate that spatial integration may be less efficient when the modelling is limited by poor observational coverage. In both cases, it is highly desirable to quantify uncertainties reliably.

Here, we focus on the domain of Switzerland, with a complex topography and climatology. A Bayesian Hierarchical Model, combining the GEV model for block maxima and the Gaussian Random Fields model for spatial dependence, is used to derive km-scale maps of return levels for 24-hour and 1-hour precipitation extremes. The observational coverage is very different between the two durations, with more than 400 stations over 60 years for the 24-hour, but less than 100 stations over 40 years for the 1-hour case. Accordingly, the model configuration for the data-rich case can involve numerous and verifiably informative covariates, whereas the modelling for the data-sparse case is left to be more scrimpy. Our results illustrate the difference in the information gain over space using a set of independent test stations. An additional finding of this study is, that efficient information gain over space is not a given from rich data alone, but depends on a careful model configuration. Advanced methods of data science do not replace a knowledgeable climatologist and patience with simple data exploration.