Quality Control and Verification of Citizen Science Wind Observations

Abstract

Wind observations collected by citizen weather stations (CWS) are valuable for forecasting wind and issuing warnings, yet their quality is not guaranteed. Few people have worked on the quality of such wind data so far. In this thesis, we develop methods to improve the quality of wind data collected by CWS. The methods are applied to filter suspect observations and correct systematic biases, processes that are known as quality control and bias correction. We focus on the wind speed observations recorded by citizen weather stations in the province of Utrecht, the Netherlands, and our data is provided by the third-party platform, WOW-NL (wow.knmi.nl). WOW-NL allows users to upload and view their weather observations in real-time. This thesis consists of four parts: (1) pre-processing the raw data; (2) performing standard quality control that checks the internal consistency, the plausible range, and the temporal consistency of observations; (3) correcting the bias by empirical quantile mapping to reduce the errors mainly caused by low sensor heights; and (4) implementing spatial quality control that compares observations from neighboring stations, where the Earth mover's distance is introduced to select neighbors. More than one-third of low-quality stations are excluded from the study before the standard quality control, based on a lack of data completeness. About three-quarters of the remaining citizen wind speed observations pass all quality control tests. We compare the citizen science data with official data, and use statistical indicators such as Kolmogorov–Smirnov statistic and root mean square error to quantify the improvements in data quality after each step. Our results show that the bias correction substantially reduces the errors after the standard quality control, and the spatial quality control further improves the data such that it is comparable with official data. This study demonstrates that the citizen science wind data match well with official data after quality control and bias correction. This means that this data can potentially be used in many applications like analyzing localized extreme wind events that require denser observation networks than currently provided by the official stations.