Evaluation of high-intensity rainfall observations from personal weather stations in the Netherlands

Abstract. Accurate rainfall observations with a high spatial and temporal resolution are key for hydrological applications, in particular for reliable flood forecasts. However, rain gauge networks operated by regional or national environmental agencies are often sparse and weather radars tend to underestimate rainfall. As a complementary source of information, rain gauges from personal weather stations (PWSs), which have a network density 100 times higher than dedicated rain gauge networks in the Netherlands, can be used. However, PWSs are prone to additional sources of error compared to dedicated gauges, because they are generally not installed and maintained according to international guidelines. Here, we quantitatively compare rainfall estimates obtained from PWSs against rainfall recorded by automatic weather stations (AWSs) from the Royal Netherlands Meteorological Institute (KNMI), over the 2018–2023 period, including a sample of 1760 individual rainfall events in the Netherlands. This sample consists of the 10 highest rainfall accumulations per season and accumulation interval (1, 3, 6 and 24 h) over a 6-year period. It was found that the average of a cluster of PWSs severely underestimate rainfall (around 36 % and 19 % for 1 h and 24 h intervals, respectively). Adjusting the data with the mean field bias correction factor of 1.24, as proposed by the PWSQC algorithm, reduces this underestimation to 21 % for 1 h intervals or almost reduces it to 0 for intervals of 3 h and longer. Largest correlation (0.83 and 0.83) and lowest coefficient of variation (0.15 and 0.18) were found during winter and autumn, respectively. We show that most PWSs are able to capture high rainfall intensities up to around 30 mm h^-1, indicating that these can be utilized for applications that require rainfall data with with a spatial resolution on the order of kilometers, such as for flood forecasting in small, fast responding catchments. However, PWSs severely underestimate (on average more than 50 %) rainfall events with return periods exceeding 10 or 50 years (above approximately 30 mm h^-1, which occurred in spring and summer. These underestimations are associated with large areal reduction factors, which can result in a reduction up to 17 % for 1 h events with a return period of 50 years. Additionally, this undercatch is likely due to the disproportional underestimation of tipping bucket rain gauges with increasing intensities. We recommend additional research on dynamic calibration of the tipping volumes to further improve this. Outliers during winter were likely caused by solid precipitation and can potentially be removed using a temperature module from the PWS.