The Potential of Using Crowdsourced Automatic Weather Stations for Urban Rainfall Monitoring in Amsterdam

An increasing number of weather enthusiasts own weather stations that share their measurements in real time on online platforms. This results in a significant amount of ground measurements, especially dense in heavily populated areas like cities, that can easily be crowdsourced from such platforms. Even though the expected errors in measurements from these often low-cost weather sensors are large and metadata is often lacking, the sheer volume of available data makes this source worth investigating. So far, crowdsourced temperature measurements have been used to investigate the urban heat island effect in cities, but precipitation data from these stations has not received much attention. The rainfall measurements from amateur weather stations can provide more information on the spatial and temporal variability of rainfall, which is especially useful in areas with many impervious surfaces where water runoff is designed.

The performance of amateur weather stations on their ability to accurately measure rainfall has been investigated. A set of 64 automatic weather stations distributed over Amsterdam (The Netherlands) that have at least 3 months of precipitation measurements during one year are evaluated. Precipitation measurements from stations are compared to a merged radar-gauge precipitation product. Disregarding sudden jumps in station-measured precipitation, the rainfall sum over time was less than that derived from the corresponding radar pixel of the reference for most stations. The weather station measurements show more agreement with the reference when increasing the time steps and spatial averaging of nearby stations, at the cost of the resolution of rainfall monitoring.

An experimental set-up of 3 commonly used amateur weather stations near a high-quality reference rain gauge was conducted to investigate instrumental errors. An unexpected result of this study is that large errors are introduced by processes on the online platform, like rounding of measurement values and gaps in the time series due to bad connectivity. Obtaining the data in the most unprocessed form results in datasets with a significantly better correspondence with the reference. The instrumental errors are limited in comparison with other sources of error that affect amateur rainfall data.

The abundance and real-time availability of these amateur rainfall sensors make them a promising addition to existing rainfall sensors. Particularly when combined with radar data, this new source of information can be of significant benefit to high-resolution rainfall monitoring.