Using sub-hourly data for estimating the frequency and intensity of extreme rainfall events across Europe

Intensity-duration-frequency (IDF) curves based on high temporal resolutions are critical for applications within urban hydrology. However, such IDF curves rely on national rain gauge networks with low spatial resolution, and the methods for producing them vary from country to country. Recent advancements in the availability of rainfall data across Europe create new opportunities for generating IDF curves at a continental scale. Our overarching aim is to develop a scalable Machine Learning method for generating IDF curves across Europe and make the results available to the public, especially users of the Scalgo Live platform.

Our initial step is to create a target dataset based on gauged rainfall data. For this purpose, we compiled a dataset of gauged sub-hourly rainfall records from five European countries (Denmark, Germany, Norway, Poland and Sweden). More data will be added as they become available. We constructed annual maximum (AM) series of rainfall intensities for 15 durations ranging from 15 minutes to 7 days and fitted Generalized Extreme Value (GEV) distributions to the data.

While the location and scale parameters of the GEV distributions showed consistent spatial patterns overall, the shape parameter was highly variable, likely due to sampling uncertainty arising from the limited number of extreme observations in the tail of the distribution. The analysis revealed significant temporal non-stationarity in approximately 5% of the AM series and indicated systematic differences in the location parameter along the Danish-German border.

Future work will use the created target dataset to identify and develop a Machine Learning model that uses geographical and climatological covariates from publicly available datasets to predict the geographical variation of IDF parameters across Europe, enabling the generation of design rainfall in both gauged and ungauged areas.