Computational Developments and Applications of the Multi-Resolution Approximation for Massive Spatial Data
Date:
Virtual presentation (session 131 - Methods for Spatial, Temporal, and Spatio-Temporal Data) at the Joint Statistical Meeting 2021 (online).
Abstract
We present recent developments of computational implementations and applications of the Multi-Resolution Approximation spatial model. The MRA tailors towards modeling situations where traditional techniques are no longer computationally feasible due to constraints imposed by the data input size. Prediction and parameter inference are instead performed via a basis function representation of a Gaussian Process. In the first part of the talk, we introduce a parallel version of the MRA in MATLAB for distributed computing environments. This implementation builds upon and improves a previous approach, facilitating computations with data sets on the order of 50 million observations while reducing execution times by up to 75%. The second portion of the talk provides a first glance at extending the MRA to model nonstationarity at global scale. A basis function representation of covariance function parameters allows us to capture nuanced spatial dependence across large domains. We apply this methodology to global measurements of sea surface temperature data from the Moderate Resolution Imaging Spectroradiometer on board NASA’s Aqua satellite.