Simon Scheider (Utrecht University) - Protogeography: validity of spatio-temporal models as a means to answer questions

Abstract: In modern data-driven geography, the question of validity is often reduced to a matter of ground truth data. However, for many spatio-temporal modeling purposes, such an account of validity is insufficient. Not only is ground truth often unavailable, data-driven approaches also fail to account for fictive modeling scenarios as well as for various purposes of modeling. Yet, the validity of geographic modeling is indispensible in the age of hallucinating AI. Valid methods make science ”scientific”, while invalid methods are bound to deliver falsehoods. Given this central role in science, it is somewhat surprising to find that validity remains a rather obscure concept. In this talk, I take a pragmatic (action-based) approach to the validity of spatio-temporal models, which I call Protogeography. I argue that spatio-temporal models are valid only by delivering a valid answer to a particular kind of spatio-temporal question, corresponding to its information purpose. Questions are about experiments including spatio-temporal controls and measures. The latter constitute the information purposes of models. Answers are valid if they fit this purpose and the underlying experiments are successful. I discuss a corresponding question-based grammar to formally distinguish among predictions, retrodictions, projections, and retrojective models. Based on various examples from geography, I show how the grammar can be used to understand the purposes and validity concerns about various spatio-temporal model applications.