Are Disaster Impact Estimates Distorted by Errors in Popular Night-Time Lights Data?

Abstract

Satellite-detected night lights data are widely used to evaluate economic impacts of disasters. Growing evidence from elsewhere in applied economics suggests that impact estimates are potentially distorted when popular Defense Meteorological Satellite Program (DMSP) night lights data are used. The low resolution DMSP sensor provides blurred and top-coded images compared to those from the newer and more precise Visible Infrared Imaging Radiometer Suite (VIIRS) images. Despite this, some disaster impact studies continue to use DMSP data, which have also been given a new lease of life through the use of harmonized series linking DMSP and VIIRS data to provide a longer sample. We examine whether use of DMSP data affects evaluations, using expected typhoon damages in the Philippines from 2012–19 as our case study. With DMSP data, negative impacts on economic activity from expected damages at the municipality level appear over 50% larger than when the VIIRS data are used. The DMSP data give the appearance of greater spatial autocorrelation in luminosity and through this channel may tend to spread apparent spatial impacts of shocks. Harmonized data that adjust VIIRS images to be like the DMSP data also have this spurious autocorrelation so researchers should be cautious in using these data for disaster assessments.