We investigate the application of atomistic models of self-interstitial defects to the simulation of Rutherford backscattering-channeling (RBS-C) spectra in ion irradiated Si. By the comparison of simulated and experimental measurements, we verify the ability of different models, either of elementary interstitials or of small clusters, to reproduce experimental spectra measured under different alignment conditions in Si lightly damaged by ${\mathrm{Si}}^{+}$ ion implantation. A model system for RBS-C simulation is built by inserting a distribution of defects in a supercell with size of $\ensuremath{\sim}{10}^{6}$ atoms. The system is then structurally relaxed by the application of the classical environment-dependent interatomic potential (EDIP). After adjusting the defect distribution in order to fit the $〈100〉$ RBS-C spectrum, simulations are performed under the other alignment conditions investigated. The scattering factors of defects are then extracted from both experimental and simulated RBS-C spectra and compared. It is shown that the anisotropy of experimental damage is not compatible with a significant presence of random (incoherent) disorder, but can be reproduced by some of the defect models under consideration: the split-$〈110〉$ interstitial, the diinterstitial formed by the addition of an interstitial to the split-$〈110〉$ interstitial, and two different configurations of the four-interstitial aggregate; one formed by two close diinterstitials and the other by the aggregation of four split-$〈100〉$ interstitials. Due to the different $〈100〉$ scattering factors of the four defect configurations which are found to reproduce experimental spectra, there is an inherent uncertainty of a factor of 2 in the estimate of the amount of interstitials by $〈100〉$ RBS-C analysis. The agreement between simulations and experiments is remarkable, considering that the method makes use of physical, although empirical, models of defects, where the only adjustable parameter is the absolute concentration of interstitials.