Robust Gross-Error Detection in GNSS Coordinate Time Series by Fitting the Central Residual Distribution

Gross errors in Global Navigation Satellite System (GNSS) coordinate time series can bias velocity and deformation estimates. Conventional three-sigma screening is prone to masking because contaminated observations inflate the residual scale. This study evaluates AEC-NDF, which estimates scale from nested central-residual subsets and applies a three-sigma decision rule. The method was compared with conventional three-sigma and interquartile range (IQR) criteria using 2192-epoch synthetic series and 1000 paired Monte Carlo realizations under correct specification, omission of a 30-day periodic term, and earthquake-like deformation. Under correct specification, AEC-NDF achieved a recall of 99.42%, a precision of 96.93%, an F1 of 98.15%, and a false-positive rate of 0.353%, outperforming both baselines in F1. Omitting the 30-day term reduced recall and F1 to 45.09% and 61.69%, respectively. In earthquake-like simulations, AEC-NDF retained recall of 99.32% and F1 of 98.09% when coseismic and postseismic terms were modeled, but its F1 fell to 49.02% when both were omitted, below IQR (49.83%). Applications to YNMH and XIAG showed contrasting flagging, but independent labels were unavailable. AEC-NDF can reduce contamination-induced masking when deterministic signals are adequately modelled and should be used after known offsets and transients have been represented.