Parametric models of diffusion MRI (dMRI) are extensively employed in both research and clinical contexts. These models yield parameters that characterize the local microstructural properties of biological tissues, derived through model fitting to real data. However, obtaining reliable parameter estimates remains challenging, primarily due to the influence of measurement noise. While several pioneering studies have addressed this issue by quantifying estimation uncertainty using bootstrap-based approaches, the present study shifts focus to the intrinsic characteristics of the models themselves, independent of measurement noise. To this end, we propose three novel metrics based on the geometry of the submanifold defined by the dMRI model within the signal space. For the models of diffusional kurtosis imaging (DKI) and neurite orientation dispersion and density imaging (NODDI), these metrics were evaluated at the typical parameter sets of brain tissues. The outcomes align with established insights in dMRI parameter estimation and suggest that the proposed metrics hold promise as new criteria for assessing the reliability of dMRI parameters.