An analysis of spatial and temporal uncertainty propagation in agent-based models

Spatially explicit simulations of complex systems lead to inherent uncertainties in spatial outcomes. Visualizing the temporal propagation of spatial uncertainties is crucial to communicate the reliability of such models. However, the current Uncertainty Analyses (UAs) either consider spatial uncertainty at the end of model runs, or consider non-spatial uncertainties at different model states. To address this, we propose a Spatio-Temporal UA (ST-UA) approach to generate an uncertainty propagation index and visualize the temporal propagation of different uncertainty measures between two temporal model states. We select the total effects sensitivity measure (a Sobol index) for a sample application within the ST-UA approach. The application is the Tobacco Town ABM, a spatial model simulating smoking behaviours. We showcase the effect of the statistical distributions of wages and smoking rates on the propensity to buy cigarettes, which leads to the propagation of uncertainty in the number of purchased cigarettes by individuals. The findings highlight the usefulness of the ST-UA in (i) communicating the reliability of the spatial outcomes of the model; and (ii) guiding modellers towards the spatial areas with relatively high uncertainties at different temporal steps. This approach can be readily transferred to other application areas that are characterized with spatio-temporal uncertainty.This article is part of the theme issue 'Uncertainty quantification for healthcare and biological systems (Part 2)'.