A Multi-Source Monitoring Data Driven Landslide Early Warning Model Using Unsupervised Machine Learning

The paper presents a landslide warning model dealing with multi-source monitoring using unsupervised machine learning methods, which include One-Class Support Vector Machine, Isolation Forest and Local Outlier Factor. Various types of monitoring data are comprehensively scored after wavelet transform and unsupervised machine learning. The accuracy of the model results is evaluated by comparing the score with the actual monitoring data changes. This will allow us to determine the accuracy and reliability of the output. The whole warning process was carried out using the Liannan landslide as an example, and the results showed that the performance of Isolation Forest was better than other algorithms with respect to comprehensive warning. As a result of its versatility in handling multiple sources of data, the model was selected for use in processing data from various channels. Subsequently, it underwent testing at an additional 7 landslide sites to evaluate its efficacy. The model demonstrated its capacity to provide quantifiable warning levels in response to deformation occurrences. In contrast to existing models, this model has advantages in monitoring new landslide sites because it can overcome the challenge of thresholds setting. Moreover, the integration of diverse parameters provides a partial solution to the problem of false alarms.