Datensatz

Zusammenfassung

Data cleaning is one of the most challenging steps in GNSS data analysis. Besides being time-consuming, failure in this process can lead to inaccurate and unreliable daily GNSS position and velocity estimates. To date, GNSS data cleaning was performed by finding positions that statistically differ from other positions without knowing the reason that is causing the outlier. However, we know that the degradation of data quality is one of the key factors influencing the quality of GNSS position estimates. We implemented a supervised machine-learning algorithm to predict, based on daily GPS data quality indicators of a GNSS station, if the daily GNSS data will be good enough to result in a reliable station position. To do so, we investigated the correlation between degraded GNSS data quality indicators and outliers in daily GNSS position time series. Six GNSS data quality indicators (number of observed versus expected observations in dual frequency, the lowest elevation cut-off observed, number of missing epochs, number of satellites, number of observations, and number of cycle slips) were employed to construct a predictive model that is able to detect outliers in daily position time series. The algorithm was trained using 12 years of position time series for 200+ GNSS stations and their associated daily data quality indicators. Based on this, we assessed the most relevant GNSS data quality indicators for detecting outliers in the GNSS position time series. We present the current development of this automated algorithm, the challenges we faced, and the preliminary results of this work.