The present work attempts to improve existing erosion models for ice crystal icing (ICI) simulations. In particular, prior comparisons between experiments and numerical simulations highlighted two major limitations of current erosion models. First, the erosion rate does not explicitly depend on the size of the impinging particles in existing models while the influence of the size parameter has been clearly demonstrated in experimental icing test campaigns. Second, erosion rates are systematically underestimated when considering accretion occurring at high liquid water to total water contents, i.e. large ice crystal melting ratios. The present work proposes to partially remedy these limitations by accounting for fragmentation effects of the eroding particles during their impact, resulting in an explicit size dependency of the erosion rate on ice crystal size. The second limitation is somewhat mitigated by accounting for the weakening of the ice layer due to entrapped liquid water. A third improvement is related to a more accurate description of geometrical effects for ice mass removal induced by tangential particle impacts. The resulting modifications in erosion rate prediction are then implemented in the icing suite IGLOO2D. Comparisons of numerical accretion simulations based on the improved erosion model with experimental data of the literature confirm quantitative improvements with respect to existing erosion models.