The prediction of time to slope failure (TOF) is one of the most pivotal concerns for both geological risk researchers and practitioners. Conventional inverse velocity method (IVM), based on the analysis of displacement monitoring data, has become an effective method to solve this problem because it is easy to perform and the prediction results are generally acceptable. Practically, some limitations like random instrumental noise, environmental noise, and measurement error are ubiquitous factors hampered the reliability of the prediction. In this work, traditional IVM method and modified IVM with three different filters are respectively detected on velocity time series from an landslide event in an open-pit coal mine with the propose of improving, in retrospect, the accuracy of failure predictions. Simultaneously, the effects of noise on the appraisal of IVM graphics are also assessed and explanation. The results demonstrate that the sliding process of landslides can be divided into three signature stages based on the IVM. Noteworthily, the slope failure critical point occurs at the end of the progressive stage and generally coincides with a major acceleration event in which almost integrity of the slope is lost, transitioning to a linear trend ever since. Additionally, the short-term smoothing filter (SSF) and long-term smoothing filter (LSF) models can provide more accuracy and useful information about the probable failure time. Finally, with the intention of enhancing the feasible use of the method and supporting predetermined response plans, two-level alert procedures combing SSF and LSF are proposed.
