Knowledge about soil behaviour in the near surface is fundamental to a broad range of military strategic research efforts. The prevailing mathematical representations of near-surface shear strength do not necessarily reflect the observed shear strengths. This paper presents an experimental procedure for testing self-supported, unconfined-drained specimens to investigate the shear strength of a poorly graded sand with the inferred unsaturated behaviour from typical effective stress expressions. The impact of soil suction is quantified from the soil–water characteristic curve obtained from a modified, unsaturated, one-dimensional oedometer device: Fredlund’s device. Seventy-seven unconfined tests illustrate that while individual or discrete states can be explained by effective stress relationships, those expressions do not accurately quantify the observed shear strength from one state to the next. Saturation failure is observed when the degree of saturation is between 70.5% and 76.5%, well below the assumed threshold. The results show that the dry shear resistance is a function of the initial density state, but as the degree of saturation increases, the shear resistance becomes independent of the initial density, and converges to the saturation failure. These findings expose the limitations of effective stress expressions and suggest a revised nonlinear, three-dimensional, unconfined Mohr failure envelope based on an applied external stress.

An energy-based laboratory-testing program was undertaken to investigate the effects of different testing methods, numerical model fits, and soil fabrics and densities on the soil-water retention curve (SWRC) using a poorly graded sand. Four different reconstitution energies and three saturation levels were used to generate different soil fabrics and structure within a narrow band of possible densities, as limited by the mechanical properties of the soil particles. Tests were performed using a “transient retention imbibition method” and a Fredlund device to develop a statistically representative laboratory SWRC. Testing results for the poorly graded sand indicate little aleatory variability in SWRC from the soil structure. The dominant source of data variability is a function of the epistemic uncertainty associated with the testing methods and fitting models but can be accounted for by a bounded mean SWRC. This bounding allows for the development of a laboratory “proxy” soil, representative of generalized sand SWRC behavior, for use as a hazard screening tool for modeling unsaturated sand behavior. The proxy soil SWRC is compared with other generalized SWRC models and independent SWRC field and laboratory tests, wherein the proxy soil SWRC yields significant increases in accuracy between the estimated and field SWRC behavior.