When Space Beats Time: A Proof of Concept with Hurricane Dean

Abstract

In this research, we present an empirical case study to illustrate the new framework called “space beats time” (SBT). SBT is rooted in the expectation that predictions based on temporal autocorrelation typically outperform predictions based on spatial autocorrelation, except in the aftermath of abrupt disruptive events. Following such disruption scenarios, space is likely to outperform time, albeit often for a brief post event period. We illustrate the SBT concept by assessing the impact of Hurricane Dean on vegetation greenness using a remotely sensed spatiotemporal data series. We predict the normalized difference vegetation index (NDVI) using separate temporal-only and spatial-only models without the aid of covariates. We then compare each prediction model’s performance before and after the hurricane event. Results suggest that SBT expected behaviors are valid in general terms but that some issues require attention. Our case study shows conspicuous SBT effects in the aftermath of the hurricane event in question, including increased performance in the geographic areas where the hurricane impact was more severe. In addition, we unexpectedly find that a more limited SBT pattern is present before the hurricane. This unanticipated pattern suggests that the presence of SBT features in an empirical study may vary, depending on the strength of a disruptive event as well as on the ability of a dataset and proxy variable to capture a disruptive event and its effects.