A Comparison of Gridded Population Data Products in Disaster Response

Abstract:

From global pandemics to localized earthquakes, effective disaster response hinges on accurate knowledge of how the affected population is spatially distributed. Yet many of the most vulnerable people on the planet live in countries that note only have limited resources for disaster response, but also often lack access to high-resolution population data to mount a response. In fact, many rapidly growing, lower to middle income countries have not conducted a reliable census in decades. The COVID-19 pandemic is hindering 2020 census efforts as well, even for high-income countries like the United States.

One way to counter this challenge and provide reliable fine-resolution population data is through the use of consistent and comparable gridded population datasets. However, there exists a range of gridded population datasets that rely on various data production approaches. These disaggregation methods range from simple areal weighting to more statically and ancillary-dependent techniques. The “fitness-for-use” considerations matter when trying to choose one gridded product over another. There is known pixel-level count variation across global gridded populations datasets and most datasets do not provide error metrics or confidence intervals. In addition, few studies to date have compared and contrasted gridded populations population estimates in a real-world disaster response scenario.

Here we estimate the affected population across globally comprehensive, gridded population datasets by the April 2015 Nepal earthquake. For each gridded population dataset, we determine the populations affected by intensity emanating from the earthquake’s epicenter, as well as each aftershock. We also determine how each dataset approximates both urban and rural populations impacted by the earthquake. Our objective is to assess how and why gridded population datasets vary in estimating populations in an actual disaster situation. This case study provides a starting point for learning how to harness these datasets in data-poor regions to improve disaster mitigation and response strategies for those most in need.