Designing Integrated Terrestrial and Non-terrestrial Networks via Machine Learning Models Over Riemannian Manifolds