How well do we really know the world? Uncertainty in GIScience

Michael F Goodchild

doi:10.5311/josis.2020.20.664

Abstract

There are many reasons why geospatial data are not geography, but merely representations of it. Thus geospatial data will always leave their user uncertain about the true nature of the world. Over the past three decades uncertainty has become the focus of significant research in GIScience. This paper reviews the reasons for uncertainty, its various dimensions from measurement to modeling, visualization, and propagation. The later sections of the paper explore the implications of current trends, specifically data science, new data sources, and replicability, and the new questions these are posing for GIScience research in the coming years.

Highlights

As Alfred Korszybski put it [9], “the map is not the territory,” or in today’s digital era we might say that geospatial data are not geography, but merely a representation of it
Users of GIS and geospatial technologies in general are often reluctant to acknowledge uncertainty, perhaps expecting that results from a machine that operates to a precision of eight or sixteen decimal digits will be even more accurate than those obtained from analog maps
If uncertainty is present in the data and in the results of analysis, how much variation from one area to another is attributable to uncertainty, and how much to spatial heterogeneity? What is the role of place-based methods in this context, since they explicitly allow model parameters to vary from one area to another? Do we need a modified concept of replicability, call it weak replicability or weak generalizability, to accommodate the essential nature of research in GIScience?

Summary

Introduction

As Alfred Korszybski put it [9], “the map is not the territory,” or in today’s digital era we might say that geospatial data are not geography, but merely a representation of it (for a recent, broad review of the significance of this simple comment in science see [13]). GOODCHILD commonly assessed against the paper maps that were its source, rather than against the reality that they supposedly represent Work on this topic emphasized accuracy [5], that is, the measurement of the differences between a representation and the truth, often termed error. Users of GIS and geospatial technologies in general are often reluctant to acknowledge uncertainty, perhaps expecting that results from a machine that operates to a precision of eight or sixteen decimal digits will be even more accurate than those obtained from analog maps We see this every day in apps that report latitude and longitude to far more decimal places than is achievable even with the best measuring instruments, and pay no attention to the actual physical dimensions of the feature whose location is being reported. The remaining section of the paper looks forward, suggesting ways in which this situation might be improved in the coming years

Uncertainty in data science

New data sources

Replicability

Concluding remarks