Leveraging Official Weather Data to Fill the Temperature Data Missing in Citizen Weather Station (CWS) through ANN Method

Citizen weather stations (CWS) record outdoor temperatures hourly at a high density on a spatial scale, spatially complementing those recorded by traditional official weather stations, providing a prerequisite to understand outdoor temperature at a high spatiotemporal resolution over any time period. However, compared with the data from traditional weather official stations, the quality of CWS data is relatively low. To be more specific, the time series recorded by CWS tends to have many gaps due to the failure of sensors, connection issues of sensors, misuse of sensors and factors like that. To improve the reliability of CWS data, a statistically-based quality control (QC) is developed and introduced in previous studies [2].

Based on statistical analysis, the unreliable data can be deleted and some small gaps like missing one or two values can be filled by linear interpolation method. However, for those time series with longer gaps, for example, stations recording <19 hourly values per day (80%), or <80% of daily data per month, the linear interpolation is not valid for gap filling and all values during that period will be deleted through QC process. This leads to the waste of CWS data.

ML methods could be one answer for reducing this kind of waste. To solve this problem, this study focuses on dig up the spatial dependency between CWS data and weather data from official weather stations through ML method. Thus, the changes on time dimension can be captured and expressed through time series of official weather stations, which tends to be continuously, and differentiates from CWS data due to the urban morphology, population distribution, traffic flow and factors like that. Therefore, the gaps could be filled by inputting the time series from official weather stations during the corresponding time period.

.