Probabilistic Learning on Manifolds for Statistical Surrogate Models in Uncertain Nano-to-Macro Stochastic Systems