Characterization of spatially varying soil properties using an innovative constraint seed method

Abstract

Soil properties are naturally varying in space. Random field model provides a powerful method for characterizing spatially varying soil properties, but it may not match the actual values at the measured locations since the spatial location information of site data is not fully utilized. This paper proposes an innovative Constraint Seed Method (CSM) for efficiently generating the conditional random field of soil properties based on available site data. It incorporates site data information to constrain the random seeds, which in return constrains the random fields. The obtained conditional random field are generally consistent with the observed values at the measured locations, and most observed and unobserved data points fall within the 95% confidence intervals due to spatial correlation. The standard deviations of updated predictions at the measured location can gradually converges to the standard deviations of measurement error, while the standard deviations of updated predictions at the unmeasured location also reduced due to the spatial correlation. Four geotechnical examples are utilized to illustrate the effectiveness of the proposed CSM. The CSM performs well across four geotechnical engineering problems that account for real site data, non-stationary characteristics, and geological uncertainties. The results indicate that the CSM can significantly reduce the global uncertainty of the site, especially with increasing observed data. Compared to other available methods, the CSM displays greater uncertainty reduction and higher accuracy while requiring less computational time. With the CSM, a more accurate characterization of soil properties can be obtained, which is essential for the geotechnical design and construction.