Crowdsourced Rainfall Data Cleaning, Validation, Visualisation, and Application (WOW)

Abstract

Rainfall observations from citizen scientists in Britain are shared from private automated weather stations (PAWS) to the Met Office Weather Observations Website (WOW). There is a reluctance by researchers and professionals to use the data which is ascribed to the relative difficulty in access, and an uncertainty in data quality. This research provides an assessment of the available data, metrics to assess data quality, examples of potential applications, and considers barriers and benefits to the provision of such data. Over 2,700 PAWS records were subject to statistical quality control (SQC), which flagged 2.5% of the dataset for removal representing 95% of the recorded rainfall total, indicating the propensity for PAWS records to include implausibly high rainfall. The post-SQC data were assessed against comparable datasets, establishing that some PAWS (40%) were generating good quality data all the time (based on SQC performance) whilst other records included intermittent faults (50%), and some were unreliable observers (10%). A systematic manual SQC process is presented, allowing the selection of good quality PAWS records for use in postevent analysis of convective rainfall events. When post-SQC’d rain observations from WOW and Official sources were compared with radar there was no statistically significant difference. This research concludes if Official data are accepted for use, then WOW data selected using the presented SQC methods are also of acceptable quality. The results of blending gauge observations with radar from four convective rainfall events are presented to show the value of WOW data in post-event analysis. Examples of improved delineation of the rainfall field and more accurate representations of rainfall than derived solely from radar are provided. This is further illustrated with an example of catchment modelling showing a rainfall dataset incorporating Official and WOW gauge data with radar improves model efficiency, as compared to using only gauge data or radar derived rainfall. Finally, the challenges of sustainable citizens science rainfall observation are addressed. A series of approaches for the instigation and promotion of citizen science are presented. Conclusions on the most effective way to ensure a rewarding experience for participants include self-selection combined with easy access to data sharing, timely data validation with feedback, and peer/technical support. This approach supports willing observers, promoting the generation of good quality reliable rainfall data.