A method to derive satellite-based extreme precipitation return levels in poorly gauged areas

Precipitation frequency analysis based on satellite products is still limited by estimation errors and by the use of statistical methods inadequate for these products. However, when it comes to poorly gauged areas of the world, satellite products can be a vital source of information. We present here a new method to derive satellite-based estimates of extreme precipitation quantiles with long return period in poorly gauged areas. The method relies on the identification of relations between statistics of the satellite estimation error and errors in the parameters of a non-asymptotic extreme value distribution. We show an application of the method in three areas with diverse climatic conditions in Austria and in the South-eastern Mediterranean, showcasing results for different scenarios of rain gauge density. We find that simple linear relations can explain 35-90% of the variance in the error of the parameters of the non-asymptotic extreme value distribution. Using these relations, we derive estimates of extreme return levels with drastically reduced bias and dispersion with respect to the ones directly obtained from the satellite estimates.