Deblurring DMSP nighttime lights: A new method using Gaussian filters and frequencies of illumination

A well known difficulty with the Defense Meteorological Satellite Program's nighttime lights series (DMSP-NTL 1992-2012) is that the images suffer from pervasive blurring, dubbed 'overglow' or 'blooming'. In this paper we devise a new method that significantly mitigates blurring. We assemble a sample of isolated light sources around the globe and discover that blurring is governed by a symmetric Gaussian point-spread function (PSF), but that the brightness of sources widens the PSF. To make sense of this, we recreate step-by-step the satellite's data collection and storage process, and discover an important fact: any pixel containing a light source will tend to be lit at least as often as its neighbors. This regularity provides a second filter on the data that allows us to calibrate the dimensions of the PSF to each part of the globe, each satellite, and each year. We generate a user-friendly, open-access MATLAB script that deblurs all DMSP-NTL images for all years, and we showcase the enhanced images for a sample of locations around the globe.