Abstract
This paper discusses the software D-STEM as a statistical tool for the analysis and mapping of environmental space-time variables. The software is based on a flexible hierarchical space-time model which is able to deal with multiple variables, heterogeneous spatial supports, heterogeneous sampling networks and missing data. Model estimation is based on the expectation maximization algorithm and it can be performed using a distributed computing environment to reduce computing time when dealing with large data sets. The estimated model is eventually used to dynamically map the variables over the geographic region of interest. Three examples of increasing complexity illustrate usage and capabilities of D-STEM, both in terms of modeling and implementation, starting from a univariate model and arriving at a multivariate data fusion with tapering.
Highlights
The understanding of complex environmental phenomena usually requires the analysis of multiple variables observed over space and time, resulting in possibly large and complex data sets
When multivariate space-time data sets are considered, it is common to rely on statistical spatio-temporal models able to exploit the correlation across variables and to provide spacetime predictions over the geographic region of interest (Cressie and Wikle 2011)
This paper introduces the D-STEM software as a statistical tool for the analysis of environmental space-time data sets and the prediction, uncertainty included, of the observed variables
Summary
The understanding of complex environmental phenomena usually requires the analysis of multiple variables observed over space and time, resulting in possibly large and complex data sets. This paper introduces the D-STEM (distributed space time expectation maximization) software as a statistical tool for the analysis of environmental space-time data sets and the prediction, uncertainty included, of the observed variables. D-STEM: Analysis and Mapping of Environmental Space-Time Variables detailed in this paper by introducing three case studies of increasing complexity. Compared to the packages Stem and spTimer, D-STEM allows to estimate a larger class of univariate and multivariate hierarchical space-time models and it is optimized for large data sets.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have