Search for: All records

Abstract We propose a non‐stationary spatial model based on a normal‐inverse‐Wishart framework, conditioning on a set of nearest‐neighbors. The model, called nearest‐neighbor Gaussian process with random covariance matrices is developed for both univariate and multivariate spatial settings and allows for fully flexible covariance structures that impose no stationarity or isotropic restrictions. In addition, the model can handle duplicate observations and missing data. We consider an approach based on integrating out the spatial random effects that allows fast inference for the model parameters. We also consider a full hierarchical approach that leverages the sparse structures induced by the model to perform fast Monte Carlo computations. Strong computational efficiency is achieved by leveraging the adaptive localized structure of the model that allows for a high level of parallelization. We illustrate the performance of the model with univariate and bivariate simulations, as well as with observations from two stationary satellites consisting of albedo measurements. 

While many statistical approaches have tackled the problem of large spa- tial datasets, the issues arising from costly data movement and data stor- age have long been set aside. Having easy access to the data has been taken for granted and is now becoming an important bottleneck in the performance of statistical inference. As the availability of high resolution spatial data continues to grow, the need to develop efficient modeling techniques that leverage multi-processor and multi-storage capabilities is becoming a priority. To that end, the development of a distributed method to implement Nearest-Neighbor Gaussian Process (NNGP) models for spa- tial interpolation and inference for large datasets is of interest. The pro- posed framework retains the exact implementation of the NNGP while allowing for distributed or sequential computation of the posterior infer- ence. The method allows for any choice of grouping of the data whether it is at random or by region. As a result of this new method, the NNGP model can be implemented with an even split of the computation burden with minimum overload at the master node level.