README.md

In jae0/emei: Spatiotemporal models of variability

STMV

Space-Time Models of Variability (STMV) is a simple spatiotemporal modelling and prediction framework that addresses non-stationarity of spatiotemporal processes, that is where mean and variance varies across space and time. STMV comes from the tradition of Kriging with external drift, where external drift represents some externally imposed gradient that is first modelled at a global (domain-wide) level and then the residuals are modelled at a local (moving spatial window) level as some spatial autocorrelation process. This is, therefore, a divide and conquer approach to large space-time modelling and prediction problems. It is, therefore, an ad hoc hierarchical model. It is primarily a prediction-focussed approach, where predictions of overlapping locations are represented as variance-weighted spatial(-temporal) averages.

More specifically, external forcing ("drift") is treated separately (global modelled via MGCV, GLM or any other framework) from the random spatial processes. The random processes are subjected to a moving window that incrementally grow depending upon local estimates of spatial structure via autocorrelation functions. More refined results would be expected if Expectation-Maximization approaches or Gibbs sampling were to be used. However, due to computational load, this was not available when initially constructed (2010). This may change soon and I might update the approach, time permitting.

This package provides the core functions required for prediction and inference of STMV-type spatiotemporal models.

For a short self-contained example, see the test example here modelling ocean bottom depths.

Other examples can be found in the various projects in https://github.com/jae0/aegis.*, but these are more involved as they use much larger data sets and spatial/temporal domains.

Installation

To install, run the following:

  remotes::install_github( "jae0/aegis")  # helper functions
  remotes::install_github( "jae0/stmv")

You probably will want to have an Rprofile set up properly such as:

homedir = path.expand("~")
code_root = file.path( homedir, "bio" )   ### replace with correct path to the parent directory of your git-projects
data_root = file.path( homedir, "bio.data" )   ### replace with correct path to your data

require( aegis )
require( stmv )

jae0/emei documentation built on Jan. 25, 2024, 10:57 p.m.