Uncertainty, Error, and Provenance

GIS is always an imperfect conception of the real world, making some degree of uncertainty inevitable. Beyond conception, uncertainty is propagated and compounded during the processes of measurement, representation, and analysis. The presence of uncertainty is unavoidable and does not necessarily diminish the validity of geographic science, but it does influence the reproducibility and replicability of that science. Different researchers should be able to obtain similar results (within some degree of confidence) using the same data and experimental methods, and this is only possible if the sources of uncertainty in a study are properly measured and communicated.

My own experience with uncertainty in geographic research primarily comes from manual accuracy assessments of classification maps in remote sensing. During the process of classifying land cover, assessing randomly sampled points for accuracy by manual inspection of satellite imagery, and creating a confusion matrix to quantify errors of omission and commission, I learned just how inaccurate a classification map can be and how important it is to communicate that error rate.

However, this form of uncertainty measurement is relatively simple compared to the complexities of other types of geographic research. When social science is involved, uncertainty can arise from an increasing number of sources: Has location data been obfuscated to protect personal privacy? Have survey respondents answered questions truthfully? Is there bias in researcher/participant interactions? How arbitrary and vague are the boundaries of livelihood zones? Do the resolution size and square shape of a raster cell accurately reflect the conditions of the underlying space?

In all cases, geographers have a responsibility to communicate any sources of uncertainty and measure it whenever possible. This is a basic requirement to ensure transparency and trust in their results. Beyond ensuring trust in the communication of results, there are additional responsibilities if a researcher wishes for their work to be fully reproducible and replicable. There must be full provenance (i.e. data lineage and workflow history) published and easily accessible so that not only the work, but the uncertainty within that work, can be understood and exactly reproduced.

Here is more information about the class I’m taking, Open GIScience.

References:

• Longley, P. A., M. F. Goodchild, D. J. Maguire, and D. W. Rhind. 2008. Geographical information systems and science 2nd ed. Chichester: Wiley. Chapter 6: Uncertainty, (pages 127-153)
• Tullis, J. A., and B. Kar. 2021. Where Is the Provenance? Ethical Replicability and Reproducibility in GIScience and Its Critical Applications. Annals of the American Association of Geographers 111 (5):1318–1328. DOI:10.1080/24694452.2020.1806029