Algorithms to enhance detection of landslide acceleration moment and time-to-failure forecast using time-series displacements

Landslide monitoring data are characterized by scatter that can make the velocity and acceleration of a landslide unclear. The presence of scatter can therefore influence the reliability of an early warning system (EWS) too. Data filters such as simple moving average (SMA) are commonly used to reduce scatter in the data and enhance the reliability of EWSs. Therefore, evaluating the adequacy of these filters to reproduce displacement charac-teristics representative of the landslide is important. Gaussian-weighted moving average (GWMA) and Savitzky-Golay (SG) filters are examined here against SMA. To this aim, a comprehensive numerical analysis of a synthetic database was carried out on accelerating scenarios to quantify the reliability of each filter to detect the onset of acceleration and forecast the failure time using the inverse velocity method. GWMA and SG applications in the synthetic scenarios reached reliability thresholds of 90% at 30% and 4% of the corresponding time by SMA, respectively, and provided a timelier capture of moment patterns. Specifically, these synthetic cases show the application of GMWA and SG improves failure time forecasting by 60 to 80% and 90 to 100%, respectively, compared to SMA depending on the amount of data used by the filter and the remaining time to failure. Additionally, nine failed cases (17 datasets) from the literature were examined after employing these three filters. Results of these cases show using alternatives to SMA would increase the accuracy of failure time forecasts by 60%.