Extreme weather

Cloudburst in Roskilde

In this blogpost we examine a marginal cloudburst event specifically focused on the Roskilde municipality in Denmark. The event occurred on the 26-28’th of August across the entirety of Roskilde city.

The cloudburst event in Roskilde

Recently, there have been an extreme cloudburst event in the Roskilde Municipality which turned out to correspond to a 228-year return period event causing overflow of severs, flooding of private apartments and erosion of the general floodplain.

Return period level events

But what does a 228-year return period event actually mean? Does this mean that it will be 228-years before another cloudburst event will occur or when will this occur again?

The short answer is no. A X-year return period event can occur multiple times a year, although the larger the return period the more unlikely it is to happen.

Thus a X-year return period event is a measure of the ‘unlikeliness’ of the event occurring rather than a expression of the reliable occurrence rate.

Rainfall event across Denmark

Below is a screenshot from DMI encapsulating the cloudburst events across Denmark during 26-28 of august. The worst affected areas are northern Jutland and Seeland.

Determining a return-period level event

But how does one then assign a return period level to a specific event? The following is a cooking recipe for inferring a given return period:

  • Step 1: Acquire historical time-series of measured rainfall amount
  • Step 2: Investigate time-series, ensure consistency in measurements
  • Step 3: Choose suitable extraction methodology for extracting extreme events from the time-series, methodologies include block maxima methods such as: Peaks-over-threshold, Annual Max Peaks, Average Annual Peaks and Partial Duration Series to mention a few.
  • Step 4: Rank extracted variate peaks according to a theoretical plotting position such as Gumbel, Gringorten, Hazen, Weibull, Cunnane or California.

(1)   \begin{equation*} latex P=\frac{r-b}{n+1-2b} \end{equation*}

Now you have assigned a probability to your measured and extracted events and are thus ready to examine the severity of your measured peaks. This is how we arrive at a 228-year return period level for the recent event experienced in Roskilde municipality.

For further analysis, it is possible to perform marginal model inference utilizing the generalized extreme value theorem.

Related read: Top ten earthquake misconceptions

For more information about extreme value theories. Freely available wiki resources include the following references:


Leave a Reply