Satellites, Self-reports, and Submersion: Exposure to Floods in Bangladesh

摘要

A burgeoning "climate economy" literature (Dell, Jones, and Olken 2014) attempts to understand and project the economic impacts of anthropogenic climate change. This literature has largely focused on uncovering the effects of changes in temperature and precipitation on economic activity. Important papers have documented both short-run effects-for example, on agriculture (e.g., Schlenker and Roberts 2009), health (e.g., Deschenes and Greenstone 2011), and labor (e.g., Graff Zivin and Neidell 2013)- and long-run effects-for example, on economic growth (Dell, Jones, and Olken 2012), and education (e.g., Maccini and Yang 2009). This has been made possible by the availability of temperature and precipitation data with reasonable spatial and temporal resolution. Global climate change is not only expected to alter temperature, it is also projected to cause more powerful cyclones, greater coastal storm surges, and increased frequency and severity of flooding (Intergovernmental Panel on Climate Change 2014). The economics literature has made less progress in modeling the socioeco-nomic effects of these other phenomena expected to be associated with climate change, and which may have more intense, deleterious effects in the short run. A handful of recent papers have used physical science models to create such data (e.g., Anttila-Hughes and Hsiang 2012; Hsiang and Jina 2014). However, no similar effort has been made for flooding, a class of disaster that affects more people than any other (EM-DAT 2012). We describe the progress we have made in creating a time series of flood exposure derived from a new analysis of satellite data. We focus on the lower Ganges Delta, and the nation of Bangladesh in particular, one of the countries historically most affected by floods, and predicted to experience increasing flood severity due to climate change (Mirza2011). This paper makes two key contributions. First, we present new, objective long-run time series measures of floods that will allow us to study human behavioral responses to changes in the distribution of disaster events. In particular, the unexpected nature of the change may itself have productivity consequences separate from the occurrence of a disaster event. The socio-economic consequences of the disaster may therefore depend on the novelty factor, i.e., how much experience people already had in dealing with similar events in the past. This is a dimension of adaptation that is possible to study only with rich data on the variation in background frequency of exposure at the locations where those events occur. Our dataset does exactly this, giving accurate measures of both the long-term average and the short-term variation in exposure required to study adaptation. Second, we show that rainfall and self-reported exposure are weak proxies for true flood exposure. The most damaging floods are caused by rivers bursting their banks, generally caused by rainfall occurring over an entire river basin and not just directly above where flooding occurs. We demonstrate that flooding in districts in Bangladesh is not directly correlated with rainfall at those specific locations. Floods are a result of complex hydrology and this lack of correlation will likely hold in many locations around the world. We also show that self-reported exposure is a weak proxy for objective exposure, and that measurement error is likely to be correlated with important determinants of socioeconomic outcomes, in particular mean exposure to floods.