Item

Abstract

Comprehensive and robust statistical estimates of trends during heavy precipitation events are essential in understanding the impact of past and future climate changes in the hydrological cycle. However, methods commonly used in extreme value statistics (EVS) are often unable to detect significant trends, because of their methodologically motivated reduction of the sample size and strong assumptions regarding the underlying distribution. Here, we propose linear quantile regression (QR) as a complementary and robust alternative to estimating trends in heavy precipitation events. QR does not require any assumptions on the underlying distribution and is also able to estimate trends for the full span of the distribution without any reduction of the available data. As an example, we study here a very dense and homogenized data set of daily precipitation amounts over Germany for the period between 1951 and 2006 to compare the results of QR and the so-called block maxima approach, a classical method in EVS. Both methods indicate an overall increase in the intensity of heavy precipitation events. The strongest trends can be found in regions with an elevation of about 500 m above sea level. In turn, larger spatial clusters of moderate or even decreasing trends can only be found in Northeastern Germany. In conclusion, both methods show comparable results. QR, however, allows for a more flexible and comprehensive study of precipitation events.