Abstract
It is important for engineers to understand the behaviour of the shear modulus of unsaturated soils (Gunsat) for the geotechnical analyses. Various models have been proposed for the estimation of Gunsat, but the accuracy of the estimated results from the different models needs to be evaluated for the different types of soils. In this paper, five mathematical models were selected to estimate Gunsat, and the results from those models were compared with 41 sets of experimental data from the published literature. Those sets of experimental data were divided into two groups, such as cohesive soils and cohesionless soils. Four statistical tools, such as the scatter plot, Taylor diagram, box plot, and the values of Root Mean Square Error and Corrected Akaike Information Criterion, were adopted for the evaluation of the performances of those models. Based on the results from the statistical analyses, it was observed that all five models could perform well for cohesive soils, but they had different performances for cohesionless soils. Consequently, two models were recommended for the cohesive soils, whereas three models were recommended for the cohesionless soils to represent the shear modulus function of unsaturated soils.