A Higher-Order Singular Value Decomposition Tensor Emulator for Spatiotemporal Simulators

We introduce methodology to construct an emulator for environmental and ecological spatiotemporal processes that uses the higher-order singular value decomposition (HOSVD) as extension of (SVD) approaches emulation. Some important advantages method are it allows use a combination supervised learning methods (e.g., random forests Gaussian process regression) also prediction values at spatial locations time points were not used in training sample. The is demonstrated with two applications: first periodic solution shallow ice approximation partial differential equation from glaciology, second agent-based model collective animal movement. In both cases, we demonstrate combining different machine models accurate addition, case ability tensor successfully capture individual behavior space time. via real data example perform Bayesian inference order learn parameters governing behavior.