Simulations of unsteady char particle combustion rely on various models that are necessary in order to correctly predict the governing flow and combustion processes. These models, in turn, rely on model parameters, which are determined by experiments or small scale simulations and contain a certain level of uncertainty. It is therefore, essential to correctly determine the sensitivities of quantities of interest measured using such simulations, with respect to the existing parameters. In this study, a discrete adjoint algorithm is employed to extract sensitivities of various quantities of interest with respect to physical and model parameters. This adjoint framework bears a great advantage in cases where a large input space is analyzed, since a single forward and backward sweep provides sensitivity information with respect to all parameters of interest. Sensitivities are extracted for relevant quantities of interest, such as burning rate and particle temperature, and are then compared as free stream composition changes from air to oxy atmosphere. The evolution of sensitivities in time is shown to be dependent on the selected quantity of interest. Model sensitivities with respect to heterogeneous reaction parameters (oxidation of carbon, in particular) are shown to be the highest, whereas the sensitivities with respect to free stream composition are shown to be significantly lower.