Multivariate manifold-valued curve regression in time

Fr\'echet global regression is extended to the context of bivariate curve stochastic processes with values in a Riemannian manifold. The proposed regression predictor arises as a reformulation of the standard least-squares parametric linear predictor in terms of a weighted Fr\'echet functional mean. Specifically, in our context, in this reformulation, the Euclidean distance is replaced by the integrated quadratic geodesic distance. The regression predictor is then obtained from the weighted Fr\'echet curve mean, lying in the time-varying geodesic submanifold, generated by the regressor process components involved in the time correlation range. The regularized Fr\'echet weights are computed in the time-varying tangent spaces. The weak-consistency of the regression predictor is proved. Model selection is also addressed. A simulation study is undertaken to illustrate the performance of the spherical curve variable selection algorithm proposed in a multivariate framework.