Natural disasters, remote sensing, and synthetic controls

Abstract

Satellite imagery has been used for decades to study changes on Earth’s surface and understand the mechanisms that have shaped it as we know it today. Moreover, substantial improvements in computing power and the increase of data available in recent years have boosted interest for this kind of research. Pixel-based composites of large areas are easily accessible today thanks to the Google Earth Engine platform[1]. These are being used to study the evolution of different ecosystems such as forests[2], as well as the frequency of wildfires. Furthermore, technological advances over the last decades have enabled to precisely monitor variations in extreme weather events[3]. These weather phenomena seem to be larger now in quantity and size due to the increase of climate volatility[4]. The consequences of natural hazards have been mostly studied by comparing pre- and post-disaster conditions, or simple pair-wise comparisons between affected and non-affected areas, rendering inaccurate estimates[5]. We are interested in developing a system that, by means of a synthetic control approach, will enable us to causally evaluate the effects of disturbances over areas of interest using satellite imagery. Resilience is another field of interest for the research community. The decrease in resilience of regions that are recurrently hit by these events might end up making certain places inhabitable. For example, extreme weather events already have their toll on life expectancy in the US[6]. Hence, large migrations may follow as a result in the long term.