Error and Uncertainty in GIS

Error and uncertainty are important topics to any discipline of science or engineering, but I have so far had little experience dealing with either in GIS. I have briefly encountered structures and concepts like confusion matrices, root mean square error, and Gaussian distribution, but have not delved deep into any of those concepts or been able to draw meaningful conclusions from them. In the classes I’ve taken up to this point, I feel as though the entire notion of error has so far been treated as a mystery, with the vague guideline that scientists strive to reduce error without artificially picking and choosing data. My GIS education has also focused almost exclusively on error and uncertainty in conception and representation, but not in analysis (Longley et al, 2005).

Geography and GIS have a somewhat unique relationship to error and uncertainty because of the complexity and subjectivity inherent in representations of space. Geographers share some of the responsibilities common to most scientific disciplines: to report error honestly, discuss uncertainty candidly, and do both of the above as productively as possible. However, what these general principles actually mean can change significantly depending on the application, and simply reporting abstract error metrics does little to contextualize or give meaning to the results of a study. From my inexperienced perspective, error metrics such as RMSE should be valued as analytical tools but treated with constant skepticism, and their use should not be taken for granted. This is perhaps as important to replicability as any other aspect of scientific study, since it often forms the basis for whether or not we consider a study to have been successfully replicated.

Addressing error and uncertainty in GIS falls under the slippery skill of ‘study design’, which seems to bear striking similarities to the field of ‘design’ more generally - it is not immediately obvious what makes good studies effective or how to craft an elegant experiment. Communication is vital to the entire process though, and putting time and effort into writing about uncertainty can play a significant role in effectively managing it. As little exposure to methods of measuring and dealing with error as I have had, I have never discussed or learned about how to write about uncertainty in a meaningful way, and I think this represents a significant gap in my education. Whether or not a study produces the desired results, it is vital to be able to communicate what the results mean in relation to the design and scope of the study. The challenge of spatial representation inherent to geography underscores this point, since subjective decisions and practical constraints related to how a phenomenon is mapped in GIS should be described clearly and defended thoroughly.

References: Longley, P. A., M. F. Goodchild, D. J. Maguire, and D. W. Rhind. 2008. Geographical information systems and science 2nd ed. Chichester: Wiley. (Ch 6: Uncertainty, pg 127-153).

Main Page