Quality Control for Crowdsourced Personal Weather Stations to Enable Operational Rainfall Monitoring

Automatic personal weather stations owned and maintained by weather enthusiasts provide spatially dense in situ measurements that are often collected and visualized in real time on online weather platforms. While the spatial and temporal resolution of this data source is high, its rainfall observations are prone to typical errors, currently preventing its large-scale, real-time application. This study proposes a quality control methodology consisting of four modules targeting these errors, applicable in real time without requiring auxiliary measurements. The quality control improves the overall accuracy of a year of hourly rainfall depths in Amsterdam to a bias of -11.3% (0.2% when a proxy for overall rainfall underestimation by personal weather stations is used), a Pearson correlation coefficient of 0.82, and a coefficient of variation of 2.70, while maintaining 88% of the original data set. Application on a national scale (average 1 station per similar to 10 km(2)) yields high-resolution nationwide rainfall maps, hence showing the great potential of personal weather stations for complementing existing often sparse traditional rain gauge networks. Plain Language Summary Rainfall measurements are needed for many applications, for example, water management and weather prediction. Especially for models describing urban drainage, the resolution of rainfall data should be high and dense networks of rain gauges are often lacking. However, many citizens own personal weather stations that share weather observations in real time on online platforms. Crowdsourcing measurements from these platforms provides rainfall information at high resolutions in both space and time, although they can contain many types of errors. We propose a quality control method that detects and filters typical errors in this data set using spatial consistency checks, requiring no additional measurements, and which is potentially applicable in real time. The method improves the accuracy of a 1-year data set of rainfall observations of all stations in the Amsterdam metropolitan area dramatically while removing only 12% of the raw measurements. Nationwide quality-controlled observations are used to successfully construct rainfall maps over the Netherlands. This shows that crowdsourced personal weather stations provide a valuable source of rainfall observations.