(Street) Lights Will Guide You: Georeferencing Nighttime Astronaut Photography of Earth

Alex Stoken, Peter Ilhardt, Mark Lambert, Kenton Fisher

Abstract: Astronaut photography from the International Space Station provides the highest spatial resolution nighttime Earth observations imagery publicly available, offering up to a 150x increase in resolution over other freely accessible satellite data sources. Yet, this imagery is underutilized in science applications because it lacks the geolocation meta-data required for downstream analysis. We present Night-Match, a fast and accurate method for localizing and geo-rectifying nighttime astronaut photography. By combining street network data with daytime satellite imagery, we produce a reliable reference target for similarity detection via pairwise image matching. We curate and release the Astronaut Imagery Matching Subset - Night (AIMS-Night), a collection of 363 images and ground truth localizations, and benchmark our method against this set to establish a robust localization pipeline. Our method correctly localizes 81.8% of AIMS-Night, and can be quickly deployed on the over 2 million nighttime astronaut photographs to produce a high quality analysis-ready data product.