BURGER: A bottom-up approach for global sub-daily intensity-duration-frequency data

Research led by Fathom’s Senior Hydrologist, Dr Jannis Hoch, and co-authored by Fathom scientists, describes Fathom’s new dataset, Bottom Up Regionalized Global Extreme Rainfall, or BURGER for short. It’s the first of its kind to be created from rainfall gauge data and the first to contain extreme estimates for durations as short as an hour. Here, Dr Hoch explains what BURGER is and what makes it different.

What is BURGER?

BURGER is Fathom’s latest rainfall extreme data set. It contains information about rainfall intensity for a given frequency and duration of a rainfall event. This information is available globally so it provides values in areas where no national datasets are openly accessible.

BURGER was created from quality-controlled rainfall gauge observation data, which provide the highest level of fidelity. This differs from other approaches which create rainfall extreme data from satellite observations. That difference inspired the name of Fathom’s new “Bottom Up Regionalized Global Extreme Rainfall dataset”, or BURGER. Since the most extreme rainfall events have a very short duration, BURGER focuses explicitly on events that range from one hour to 24 hours.

What differentiates BURGER?

Other global extreme rainfall data sets are created from remotely-sensed data. These data often struggle to capture peak rainfall accurately, which can lead to underestimation of rainfall extremes. Rainfall gauges, on the other hand, produce more reliable estimates*. To be able to produce a global dataset, BURGER used machine learning to regionalize from the gauge scale to the global scale.

Another major difference is the way extreme rainfall estimates are determined. Many other datasets, both global and national, make use of the widely applied Generalized Extreme Value (GEV) method. However, for BURGER, we used the recently developed Simplified Metastatistical Extreme Value (SMEV)* method. The main advantage of applying the SMEV is a reduced estimation uncertainty for shorter observation periods.

How accurate is BURGER?

Compared to rainfall estimates at rainfall gauges, BURGER data agrees well. At half of the stations, the error is above 0 %, the other half is below 0 %. At half of the stations the error is between -5 % and 5 %, which means the errors are nicely centered around 0. This indicates agreement between the BURGER simulation and observation. 

For less frequent, and therefore more extreme, events, BURGER tends to show underprediction. When removing some gauges from the regionalization, BURGER accuracy drops between around 10% and 20%, indicating that BURGER estimates are largely in the right ballpark. 

Preliminary work comparing global and national datasets indicates that BURGER agrees best with national datasets – not only with respect to rainfall intensities but also with respect to resulting pluvial (surface water) flood hazard. It is noteworthy that this agreement holds in areas without gauge observations that could be used to inform BURGER’s regionalization.

How is BURGER accessed?

The BURGER dataset is freely accessible via Zenodo under a CC BY-NC-SA 4.0 license.