Probabilistic identification of soil stratification

Abstract

Identification of soil stratification is vital to geotechnical structural design and construction where the soil layer, soil type and properties are necessary inputs. Although methods are available for classifying the soil profiling using measured cone penetration test (CPT) data, the identification of soil stratification at unsampled locations is still difficult due to significant variability of natural soil. The identification is further complicated by the considerable uncertainties in the CPT measurements and soil classification methods. This study aims to develop a probabilistic method to predict soil stratification at unsampled locations by explicitly filtering the uncertainties in soil classification systems. An established Kriging interpolation technique is used to estimate the CPT parameters which are further interpreted to identify the soil stratification. Equations are derived to quantify the degree of uncertainties reduced by this method. The approaches are illustrated using a database of 26 CPT tests recently sourced from a dike near Ballina, Australia. Results show that the majority of the uncertainties in the soil parameters are screened by a soil classification index. The remaining uncertainties are further filtered by the soil classification systems. A clear stratification with a high degree of confidence is obtained in both horizontal plane and vertical unsampled locations, which shows excellent agreement with the existing CPT tests. This study provides a methodology to clearly identify the soil strata and reduce the uncertainties in prediction of design properties, paving the way for a more cost-effective geotechnical design.