A robust estimator for the intensity of the Poisson point process of extreme weather events

AbstractThe article examines extreme temperature events defined as the exceedances of several high quantiles of temperature anomalies in regional climate model data over Germany as an example for the analysis of meteorological extremes using a two step-nonparametric approach. In the first step we estimate the intensities of the Poisson point processes of temperature extremes using a kernel estimator which so far has only rarely been used in climatology. Its advantages include robustness against model selection errors, simple computability and intuitive interpretability. In the second step we aggregate the pointwise intensity curves by means of functional cluster analysis to form regions, where the exceedance probabilities of the quantiles evolve similarly over time. A distinct gradient from Northwest to Southeast is found in the data with frequencies of exceedance of the 99% quantile of more than 1000% at the end of the 21st century compared to the control period 1961–2000