Detecting Outliers in Ionospheric Correction Model for GNSS Precise Positioning

Abstract Global Navigation Satellite System fast precise positioning can be achieved with accurate ionospheric corrections computed from an adequate number of GNSS stations in a local region. Our previous study showed that using 15 well-distributed GNSS stations within a 5° latitude x 10° longitude region, a local regression model for ionospheric corrections can achieve a mean accuracy of 5 cm. In low-latitude regions, the presence of electron density gradients over short distances can lead to outliers in the map of ionospheric corrections and decrease its accuracy. In this study, we explored outlier detection in ionospheric correction mapping through statistical residuals during a four-month test in 2021. Our findings indicate that the residuals of the local ionospheric model conform to the Laplace distribution. To determine outliers, we use an empirical rule for the Laplace distribution, setting thresholds at µ ± 3b, µ ± 3.5b, and µ ± 5.8b for data retention rates of 95%, 97%, and 99.7%, respectively. Here, µ represents the location parameter, which corresponds to the median of data, and b is the scale parameter, calculated as the medium absolute deviation. We found that while removing outliers can improve model accuracy, it may result in unavailable prediction due to a lack of data. For example, applying a µ ± 3.5b threshold for outlier removal led to approximately 2.5% of recording time having no ionospheric corrections map in low-latitude regions, however, the local model has the potential to improve its mean accuracy by up to 50% for both low and mid-latitudes.