This paper presents three aspects of geotechnical engineering research that have been conducted throughout the author’s career and concludes with a brief treatment of the use of physical models in teaching. The first research topic deals with quantifying the large-scale spatial variability of the Keswick Clay in Adelaide by means of undrained shear strength data acquired from several private consulting companies and government departments, incorporating a large number of site investigations. The mathematical technique of geostatistics is used, and it is observed that kriging with a spherical model, with a range of influence of 1000 m, a nugget of 1500 kPa2, and a sill of 2500 kPa2, is able to generate good estimates of the undrained shear strength of the Keswick Clay that can be used for preliminary design purposes. Secondly, the ground improvement technique of rolling dynamic compaction (RDC) is examined in the field and in the laboratory, and numerically by means of artificial neural networks (ANNs). It is observed that RDC is able to improve the ground to depths in excess of 3 m, and the use of transparent soils in the laboratory provides useful insights regarding the influence of RDC on the subsurface profile. In addition, ANNs facilitate the development of reliable models for the prediction of the level of ground improvement due to RDC. The third and final research topic presented involves ground improvement on the Moon. It is a work-in-progress, and early results are presented in this fascinating and exciting endeavour. The paper concludes with a brief treatment of the use of three different physical models used in teaching. It is observed that incorporating demonstrations involving physical models in teaching is helpful for enhancing student learning and engagement.