BURGER: A Bottom‐Up Regionalization Approach for Global Sub‐Daily Intensity‐Duration‐Frequency Data

Intensity‐Duration‐Frequency (IDF) curves require accurate observations which are not available everywhere. To provide globally consistent IDF maps, we harness the accuracy of Global Sub‐Daily Rainfall (GSDR) gauge observations and combine this with the power of a random forest regression model to regionalize the parameters of the SMEV (Simplified Metastatistical Extreme Value) distribution. After regionalization, it is possible to compute intensities for any combination of return period and duration up to 24 hr. These regionalized intensities are named BURGER, the “Bottom Up Regionalized Global Extreme Rainfall” data set. Comparing intensities from BURGER against those obtained at GSDR stations shows overall good agreement as supported by a median percentage bias around 0% and an interquartile range between −5% and 5%. Errors increase with less frequent events, indicating a too light tail of regionalized intensities, and show marked regional variations. Intensities from simulations excluding station data in Japan and Germany deviate up to 15% from those obtained with the station data included. A benchmark with a remote sensing‐based IDF data set did not reveal structurally lower agreement in ungauged regions compared to gauged regions, suggesting a reliable transfer to ungauged areas. Comparing results with other IDF data sets shows that differences between the underlying methods and data hamper a robust benchmark. For instance, while at some GSDR stations NOAA data agrees with BURGER data, NOAA data hardly agrees with empirically derived intensities at other stations. This first bottom‐up approach to global IDF data yields promising results and insights warranting future improvements.