Ground improvement in the form of soil compaction plays a important part in reclamation projects. The development of the Cofra Roller Compaction (CRC), a non-circular impact roller, has proven to be valuable in these projects. However, the heterogeneity of the subsoil causes locally a non-uniform degree of compaction. Traditional compaction control tests are limited in measuring depth, expensive and cause time delay. Therefore, the Continuous Compaction Control (CCC) and Continuous Impact Response (CIR)method were developed in order to provide more real-time information of the compaction based on the response of the drum of the roller. The aim of this study is to develop a semi-empirical energy model which is based on the contact forces of the roller-soil interface as most CCC systems, but also uses field test data as was given in the CIR system to validate this model. The relevant parameters needed for this model were obtained from the field test conducted for the HES Hartel Tank Terminal project in Rotterdam. These included the impact acceleration, the cone resistance, the in situ density, the dynamic modulus and the dynamic plate load test velocity. Two methods are considered in this thesis and both aim to reproduce the measured values from the dynamic plate load test during the field test. The first method considers the acceleration signals and includes double numerical integration of these signals to obtain the displacement, while the other considers modelling the roller as a dynamic plate load test and obtaining the displacement from solving a 2-DOF spring-mass-damper system. However, after analysis of the motion of the roller, it was observed that due to the non-circular shape of the roller, a wedge effect was created where horizontal shearing forces caused loosening of the soil. This inhibited soil compaction up to 0.5 m depth. The impact acceleration signals were thus not representative of the soil compaction. Nonetheless, the DPL-Soil model was proven to be successful in correlating the soil settlement to the dynamic modulus. This study considers a silty sand, so further research should be carried out to obtain correlations for various soils. In order to develop the semi-empirical energy model, it is thus recommended to capture an accurate acceleration response. This can be done by placing accelerometers at a minimum depth of 0.5 m, replacing the 8G accelerometer with e.g. 16G accelerometer and increasing the sampling rate to at least 1000 Hz. In order to filter out the soil variability, a field test with the roller should be performed on a homogeneous sand without fines. Correlations can then be drawn again for the same field tests performed in this thesis. Finite Element Modelling (FEM) could be used to model the interaction between the non-circular shape of the lobe, the rolling motion and the soil. This might form a better correction method for the acceleration signals than those explained in this thesis. Low frequency geophones can be used to measure the velocity directly. This because low frequency data of the accelerometer should be removed and the roller works on a low frequency. The load imparted to the ground could also be measured directly by burying earth pressure cells at a minimum depth of 0.5m and at various depths to get a more accurate representation of the pressure distribution through the soil layers. By using other numerical integration methods such as Simpson’s 3/8 rule and Boole’s rule, the numerical accuracy of the displacement response of the roller could also be improved.