Choosing the correct cell resolution for kriging
|September 7, 2011||Posted by karmadsen under blog, GIS, Statistics, Uncertainty|
When constructing a raster from point data using kriging, how do you set the correct cell resolution? Cell resolution can have a significant effect on computation requirements, but assuming that memory and time are not a limiting resource, what is the ideal?
A desired cell size is often established before constructing a surface. Often, it is based on the intended use and detail required for the finished raster. Or, it may be determined by the physical realities of the given system. For example, cell size may be set to the thickness of the geological formation that the user wishes to capture.
A general rule of thumb is to take the width of height of the dataset (whichever is shortest), divide it by 250, and use that as the cell resolution.1 However, this only provides the default cell size. The final cell size will have to be informed by the results of kriging. Kriging software packages will return confidence intervals for cell predictions, assume that errors are normally distributed. Thus, the defensible cell size may have to be adjusted, along with other parameters, based on the desired confidence of prediction.
1. Maclean AL. (2010) Geographic Information Systems for Resources Management. Lecture 19: Introduction to Kriging. Available at: http://forest.mtu.edu/classes/fw5550/lectures/lecture19_kriging.pdf