Satellite based inference for disaster induced power outages using daily nighttime light observations

Abstract Power systems are vital to society, but remain vulnerable to natural disasters that cause widespread outages. Satellite-captured nightlights offer a global opportunity to monitor electricity disruptions. Prior work has typically relied on aggregating nightlight observations over multiple days to mitigate cloud-induced data gaps during disasters. While effective for this purpose, this aggregation obscures day-to-day dynamics needed for time-resolved outage analysis. Leveraging outage records from recent large-scale blackouts, we establish an empirical relationship between daily nightlight-intensity loss and power outages, accounting for variations in background radiance and moonlight conditions. We then operationalize this relationship as a two-stage modeling framework that first detects outage conditions and then estimates outage magnitude. We train and validate the framework using large-scale blackouts during Hurricanes Irma (2017), Michael (2018), Ida (2021), and Ian (2022), each affecting over a million customers. We further test generalizability using out-of-sample data from Public Safety Power Shutoffs in Humboldt County, California, and from Hurricane Fiona (2022) in Puerto Rico. We also apply the framework to recent earthquakes in Turkey (2023) and Myanmar (2025) to illustrate its use in regions without automated outage reporting. Overall, the framework relies on open-source, hazard-independent inputs and enables scalable monitoring of power disruptions in data-scarce settings.