posted on 2021-08-05, 11:16authored byLizheng Deng, Alister SmithAlister Smith, Neil Dixon, Hongyong Yuan
Knowledge of landslide displacement trends is important to understand risks and establish early warning trigger thresholds so that action can be taken to protect people and critical infrastructure. However, the availability of direct continuous displacement measurements is often limited due to relatively high costs. This has driven research to establish models that quantify relationships between landslide displacements and other measured parameters such as pore water pressures, rainfall and more recently acoustic emission (AE), so that displacement can be predicted, and hence made available at a lower cost. This paper describes an investigation of established machine learning models to predict displacements using time series measurements of AE and rainfall. Data from a case study site has been used to train models using measured displacements and then test to assess prediction accuracy. The LASSO-ELM model was shown to perform best. It was able to predict displacements to a mean absolute percentage error < 2.5% up to 60 days after the end of the training period, which is better than similar reported studies. Training a LASSO-ELM model using continuous high resolution AE measurements combined with rainfall data has potential to provide predicted displacement trends once direct measurement of displacement is no-longer available.
Funding
National Key Research and Development Program of China (No. 2018YFC0809900, No. 2018YFC0806900 and No. 2018YFC0807000), the Key Research and Development Program of Anhui Province (Grant No. S202104b11020044), and the Anhui Provincial Natural Science Foundation for Distinguished Young Scholars (Grant No. 1908085J22)
EPSRC Fellowship (Listening to Infrastructure, EP/P012493/1)
This paper was accepted for publication in the journal Engineering Geology and the definitive published version is available at https://doi.org/10.1016/j.enggeo.2021.106315.