Probabilistic Interpolation

Content for Wednesday, October 29, 2025

Last class we talked about deterministic interpolation to fill in gaps in our data by leveraging the first law of geography (closer things should be similar). These approaches use the distance between observed data point and the observations themselves to calculate the values that occur between the obseved points. These approaches are deterministic - there is no uncertainty or stochasticity in the values they produce. Today, we’ll look at probabilistic approaches to interpolate missing values based on statistical models built on asssumptions about first- and second-order stationarity and estimates of covariance in our observed data.

Resources

Technical Details

  • Ch. 14: Kriging in from Paula Moraga’s new book Spatial Statistics for Data Science: Theory and Practice with R gives a little gentler introduction to spatial neighbors specifically in the context of statistical models.

  • Chapter 12: Spatial Interpolation in Spatial Data Science by Edzer Pebesma and Roger Bivand provides some additional details for managing gstat and sf, but relies on stars a package we don’t use in this course.

Objectives

By the end of today you should be able to:

  • Differentiate deterministic from probabilistic interpolation

  • Recognize how first- and second-order properties relate to geostatistical interpolation

  • Define key components of a semi-variogram

  • Implement ordinary and universal krigging using gstat, sf, and terra

View all slides in new window Download PDF of all slides