Hydrological post-processing for predicting extreme quantiles

Hydrological post-processing using quantile regression algorithms constitutes a prime means of estimating the uncertainty hydrological predictions. Nonetheless, conventional large-sample theory for does not apply sufficiently far in tails probability distribution dependent variable. To overcome this limitation that could be crucial when interest lies on flood events, through extremal is introduced here extreme quantiles model’s responses. In summary, new method exploits properties Hill’s estimator from value to extrapolate regression’s predictions high quantiles. As proof concept, tested daily streamflow simulations provided by three process-based models 180 basins contiguous United States (CONUS) and further compared regression. With large-scale comparison, it demonstrated severely underestimates (at level 0.9999) regression, although both methods are equivalent at lower 0.9700). Moreover, shown that, same context, estimates predictive with efficiency is, average, study models.