The Deep-Tree Approach: An Improved Parallel Matlab Implementation of the Multi-resolution Approximation for Massive Spatial Data on High-Performance Computing Systems
Published in: NCAR Technical Notes, 2021
Blake, L.R., Huang, H., Vanderwende, B., and Hammerling, D.M.: "The Deep-Tree Approach: An Improved Parallel Matlab Implementation of the Multi-resolution Approximation for Massive Spatial Data on High-Performance Computing Systems", NCAR Technical Note NCAR/TN-565+STR, DOI: 10.5065/pzzt-wj18, 2021.
Abstract
We implement a parallel version of the Multi-resolution Approximation (MRA) in Matlab for distributed computing environments. This implementation builds upon and improves our previous Shallow-Tree approach, facilitating computations with data sets on the order of 50 million observations while reducing execution times by up to 75%. We observe less sensitivity to computing cluster and model parameter settings allowing for more flexible specification.