Semi-parametric resampling with extremes

Nonparametric resampling methods such as Direct Sampling are powerful tools to simulate new datasets preserving important data features spatial patterns from observed while using only minimal assumptions. However, cannot generate extreme events beyond the range of values. We here propose value theory for stochastic processes extrapolate towards yet unobserved high quantiles. Original first enriched with values in tail region, and then classical algorithms applied data. In a approach enrichment that we label “naive resampling”, an independent sample marginal distribution keeping rank order point out inaccuracies this around most values, therefore develop second works many replicates. It is based on asymptotic representation through two stochastically components: magnitude variable, profile field describing variation. To data, fix target return levels resample magnitudes constrained range. use heatwave scenarios over France, daily temperature reanalysis training years 2010 2016.