Insights about extreme precipitation in Limburg from crowdsourced data

23 juli 2021

Weather observations are the cornerstone of climatological and meteorological services. Crowdsourced data can complement the existing official KNMI network by providing observations with both a high spatial and temporal resolution. We examine precipitation on July 13th and 14th measured by the WOW-NL network to show how the acquisition of crowdsourced measurements can help KNMI to better understand extreme events.


On July 13th and 14th 2021, a stationary low-pressure system delivered record precipitation to the region covering parts of Germany, Belgium, Luxemburg and the Netherlands. Heavy rainfall was recorded in the south-east of the Netherlands (Limburg), leading to a serious flooding event. The extreme precipitation and flooding caused devastating impacts in the region, impacting lives, livelihoods, and damaging infrastructure. The heavy rainfall was well forecast from a meteorological perspective. KNMI issued a code yellow for extreme precipitation (> 50 mm in 24 hours) in the morning of the 12th. A code orange was issued in the evening of the 13th for precipitation exceeding 75 mm in 24 hours. Finally, because of the expected strong impact on society a code red was issued on the afternoon of the 14th. 

This was a large-scale event that straddled several countries and led to the observation of large precipitation amounts. Precipitation varies a lot in space and time, so even in a large-scale event like this, there are small regions where very extreme precipitation occurs, like the south-eastern edge of Limburg province. Extreme precipitation is strongly linked to increasing global temperatures, thus in future we can expect more extreme precipitation with subsequent impact for communities. 


Crowdsourced precipitation observations

Observations are fundamental to know the rainfall intensity and amount during the event. This information is essential for forecast verification and to improve future forecasts of extreme precipitation. Observations also help to put this event in a climatological context. For example, we can assess how extreme this precipitation event was compared to previous ones, or to carry out attribution studies in which we estimate the probability of this event being triggered by climate change. 

The KNMI has two networks acquiring ground-based measurements of precipitation. The first one is the official network of 35 automatic weather stations that measure precipitation every 10-minutes (‘AWS network’), whereas the second is a network of 300+ rain gauges at which volunteers measure precipitation once a day (‘volunteer network’). These are invaluable resources to get a timely picture of precipitation over the Netherlands, but the networks might not have sufficient spatial density to capture small-sized severe downpours, or they provide a single measurement that might not be informative of the intra-day rainfall dynamics.  

In this context, crowdsourced weather data might be a good alternative to fill these gaps. The KNMI is a partner of the Weather Observations Website (WOW-NL), an initiative promoted by the UK Met Office. In the Netherlands, a network of 500+ citizen weather stations monitor the weather and contribute observations to a central repository. The general public can upload their own weather observations and check the real-time measurements through the portal ( The images below (Figure 1a) show the precipitation from the low-pressure system in the afternoon of July 13th recorded by the KNMI radar. These figures also show the large number of WOW observations that were recording rainfall amounts in the south-east of the Netherlands on the 13th of July, with several WOW stations located directly in the region of the largest precipitation amounts. This is in stark contrast to the AWS network, which has only one station (Maastricht) in the region. 

A figure showing that there are many more WOW-NL stations in southern Limburg compared to the AWS network
Figure 1. a) The KNMI (top) and WOW (bottom) stations that were measuring precipitation on July 14th 2021. Uncalibrated radar observations are shown in the grey-red shading. b) Daily precipitation from the volunteer network on the 13th and 14th July 2021.
A figure showing that some WOW-NL stations recorded much more rainfall than the AWS station in Maastricht
Figure 2. Cumulative precipitation measurements at 32 WOW stations (coloured lines) and the only KNMI station (Maastricht AP, black line) located in south Limburg on July 13th and 14th 2021.

What can we learn about the precipitation in Limburg?

The official AWS weather station in south Limburg was not close enough to the centre of the low-pressure system to capture the most intense precipitation, which was located 20-30 km to the southeast. The cumulative precipitation amounts show that the official station measured around 25 and 40 mm of rainfall on the 13th and 14th, respectively (black line, Figure 2). WOW stations located more to the south-east measured daily precipitation amounts that were more than double the official station (coloured lines, Figure 2). 

It is important to note that here we are showing raw WOW-NL observations (Figure 2), that is, observations that have not been pre-processed, subjected to a quality assessment, or any statistical correction. At KNMI we are currently investigating how to apply quality controls to citizen science weather data to have a better understanding of the reliability of these observations (de Vos et al., 2019, Chen et al., 2021a, b). However, this preliminary visual comparison seems to be promising, since the daily values recorded by WOW-NL (30-100 mm of rainfall) recorded in southern Limburg are consistent with the official values from the volunteer network (Figure 1b).  

These visual representations suggest that the WOW-NL precipitation measurements are promising to complement the official observing network in future. The values match well with the daily volunteer network, but they provide a much higher temporal resolution. For example, we can also see large differences between the official station in Maastricht and the WOW-NL stations in the rainfall intensities. The relatively flat black line in Figure 2 indicates low hourly precipitation intensities in Maastricht, while the very steep green line (WOW-NL station starting with ‘141613f3’) shows that the maximum hourly precipitation intensity exceeds 50 mm/hour in the afternoon on the 13th of July. In this way, we can see the time and location of the largest precipitation intensities, which currently we can’t get from either of the KNMI ground-based networks at such high spatial or temporal resolution. The radar is one possible source of rainfall intensities with high spatial and temporal resolution. Unfortunately, there are large uncertainties associated with the rainfall amounts returned by the uncalibrated radar, particularly in mountainous regions and for higher rainfall intensities.  

An animation (visible here on vimeo) shows the hourly measurements collected by WOW-NL (coloured dots) and how they compare with the official KNMI station (coloured square). The WOW-NL network shows the precipitation starting in the east of Limburg around 16:00 UTC on July 13th and moving westwards through the evening. These intra-daily features are not visible in either the AWS or volunteer networks.  

Incorporating crowd sourced data such as WOW-NL measurements can provide a (near) real-time data stream that offers the ability to pinpoint more precisely location and timing of extreme precipitation events. This information is useful for all-level administrations to increase the preparedness to mitigate the negative effects of extreme rainfall over communities.  


Irene Garcia-Marti, Kirien Whan, Jan Willem Noteboom, and Peter Siegmund



  • de Vos, L. W., Leijnse, H., Overeem, A., & Uijlenhoet, R. (2019). Quality Control for Crowdsourced Personal Weather Stations to Enable Operational Rainfall Monitoring. Geophysical Research Letters, 46(15), 8820–8829. 
  • Chen, J., Saunders, K. & Whan, K. (2021a), ‘Quality control and bias correction of citizen science wind observations’, Quarterly Journal of the Royal Meteorological Society (under review). 
  • Chen, J., Saunders, K. & Whan, K. (2021b), ‘QCWind’,