Random Field Parameter Identification and Model Selection Using Time-Series PWP Data

To effectively characterize the spatial variability of soil properties using random field theory, it is crucial to determine the autocorrelation function (ACF) model and model parameters, includingthe scale of fluctuation ('). However, identifying these solely from massive monitoring data is challenging. This paper proposes a novel Bayesian model selectionframework to identifythe optimal ACFmodel among a set of candidate ACF models (i.e., single exponential(SNX), squared exponential (SQX), second-order Markov (SMK) and binary noise(BIN) ACFs) and to learn ' and spatiallyvarying soil parameters usingtime-series monitoring data of pore water pressure (PWP). The Bayesian updating with subset simulation (BUS)method is adopted to facilitate ACF model selection and parameter identificationwith massive monitoring data of PWP.The effectiveness of the proposed framework is validated using a real slope case in Hong Kong, China. The calculated results indicate that the SNX ACF model is identified as the most optimal ACF model. The epistemic uncertainties of ' and spatially varying saturated hydraulic conductivity (ks)can begreatly reduced by integrating the time-seriesmonitoring data of PWP.The PWP values predicted using the most optimal ACF model andthe posterior samples of parameters match wellwith the monitoring data of PWPduring validation.

Keywords: Slope, Bayesian model selection, Spatial variability, Autocorrelation function, Time-series monitoring data, Bayesian analysis.