Development of precipitation-data-based real-time slope stability estimation method considering the uncertainty of soil hydraulic properties

An autoregressive longand short-term memory (ARLSTM) model was applied to develop a real-time probabilistic slope stability estimation model for the engineered barrier system (EBS) of a near surface radioactive waste disposal facility. The effectiveness of the developed model was verified using actual data acquired from South Korea, including precipitation, soil moisture contents, and inclinometer time-series data. The precipitation and the factor of safety (FS) ensemble results were used as the input and output variables of the AR-LSTM model, respectively, where the FS ensemble results were calculated by the Taylor model, integrating the Mualem-van Genuchten soil water retention model with consideration of the multivariate statistics on the hydrophysical properties of the soil. The estimation accuracy of the AR-LSTM model was reasonable by showing high correlation coefficient (0.9468) and low root mean squared error (0.0070) values between the actual and estimated FS values. Moreover, a significant correlation was observed between the estimated FS ensemble results and displacement events recorded by the inclinometer sensor. All the results suggest the effectiveness of the developed model for the long-term integrity assurance of the EBS.