Modelling and managing uncertainty in soil maps

The course 'Modelling and managing uncertainty in soil maps' is currently under development.

For information contact Gerard Heuvelink

Course title: Modelling and managing uncertainty in soil maps
Course code: ISRIC-UNCERT
Subject areas (Curriculum): Statistics, Geostatistics, Soil Science
Duration:  1 or 2 days
Short description: Soil maps are not perfect, whether they are produced with traditional methods or with digital soil mapping techniques. The first part of this course addresses the accuracy assessment of soil maps by comparison with independent validation data. It explains how common accuracy measures such as map purity, mean error and root mean squared error can be estimated using sampling theory from statistics. The second part of the course focuses on the explicit modelling of soil map uncertainty using geostatistical models. These models represent the real world as a composite of a deterministic and stochastic component. Kriging uses information in point observations and covariates to reduce the stochastic component but cannot eliminate it entirely. It therefore results in probability distributions that are centered around the most likely value of the soil property or soil type, but that also characterise the uncertainty by means of the spread of the distribution.
Moderators: G.B.M. (Gerard) Heuvelink, T. (Tom) Hengl and B. (Bas) Kempen
Target audience:
  • Users of soil maps that acknowledge that soil maps are not perfect and wish to know how the associated accuracy can be quantified;
  • Producers of soil maps who want to check on the quality of the maps that they make and learn about geostatistical methods to explicitly incorporate soil map accuracy in their DSM models
  • Intermediate level of statistics (probability distribution, random variable, estimation, linear regression)
  • Familiarity with the R language for statistical computing is not required but will ease the computer practicals
Programme:  Module 1:
  • Lecture on use of sampling theory to estimate common accuracy measures for maps of soil types and properties
  • Computer practical on topics discussed during the lecture using real world data
Module 2:
  • Lecture on geostatistical models and kriging to explicitly characterize uncertainty in interpolated maps of soil types and properties
  • Computer practical on topics discussed during the lecture using real world data
Software / materials:  R base package and additional libraries; soil maps and datasets
Case studies:  
Minimum number of participants:  
Maximum number of participants:  
Further inquiries: Gerard Heuvelink