Parallel Implementation of the Multi-resolution Approximation for Large-scale Spatial Gaussian Models in Python

Gaussian processes have become a standard tool for modeling large atmospheric data sets. Traditional methods used in this context, such as kriging, are computationally infeasible for many data sets encountered in practice due to their size. One solution to overcome this limitation is to employ the multi-resolution approximation (MRA), developed by. In this technical report, we describe a Python package containing a parallel implementation of the MRA algorithm for estimating the true value of the process from noisy data. We describe the algorithm and its implementation, apply it to several simulated data sets and report run times. We also include code examples that cover typical use cases of the algorithm as well as some diagnostic tools to verify its results.