Maps are not what they seem: representing uncertainty in soil- property maps

Tomislav Hengl 1 and Norair Toomanian 2
1 European Commission, Directorate General JRC Institute for Environment and Sustainability Land Management & Natural Hazards Unit
TP 280, JRC Via E. Fermi 1, I-21020 Ispra (VA), Italy 
Tel.: + 39 0332 785535; Fax: + 39 0332 786394
tomislav.hengl@jrc.it
2 College of Agriculture, Isfahan University of Technology
Isfahan University of Technology, 84154 Isfahan, Iran
Tel.: + 98 311 3913360; Fax: + 98 311 3912254
norair@sepahan.iut.ac.ir

Abstract
The paper discusses use of static visualization techniques for representation of uncertainty in spatial prediction models illustrated with examples from soil mapping. The uncertainty of a prediction model, represented with the prediction error, is commonly ignored or only visualized separately from the predictions. Two techniques that can be used to visualize the uncertainty are colour mixing (whitening) and pixel mixing. In both cases, the uncertainty is coded with the white colour and quantitative values are coded with Hues. Additional hybrid static visualization technique (pixel mixing with simulations) that shows both the short-range variation and the overall uncertainty is described. Examples from a case study from Central Iran (42×71 km) were used to demonstrate  the possible applications and emphasize the importance of visualizing the uncertainty in maps. The soil predictions were made using 118 soil profiles and 16 predictors ranging from  terrain parameters to Landsat 7 bands. All variables were mapped using regression-kriging and grid resolution of 100 m. Final maps of texture fractions, EC and organic matter in topsoil were visualized using the whitening, pixel missing and pixel mixing combined with simulations. Visualization of uncertainty allows users to compare success of spatial prediction models for various variables. In this case study, the results showed that there can be quite some differences in the achieved precision of predictions for various variables and that some soil variables need to be collected with much higher inspection density to satisfy the required  precision. Visualization of uncertainty also allows users to dynamically improve the precision of predictions  by collecting additional samples. Links to scripts that the users can download and use to visualize their datasets are given.

Keywords: visualization, spatial prediction error, conditional simulations, regression kriging

In: Caetano, M. and Painho, M. (eds). Proceedings of the 7th International Symposium on Spatial Accuracy Assessment in Natural Resources and Environmental Sciences, 5 – 7 July 2006, Lisboa, Instituto Geográfico Português

AttachmentSize
Hengl2006accuracy.pdf2.5 MB