Towards a data-driven model of the sky from low Earth orbit as observed by the Hubble Space Telescope

TL;DR

This paper develops a machine learning model to predict sky brightness observed by the Hubble Space Telescope in Low Earth Orbit, accounting for stray light sources like Earthshine, and demonstrates improved accuracy over existing physical models.

Contribution

The study introduces a data-driven machine learning approach to model and predict sky brightness in LEO, incorporating satellite weather data and outperforming traditional physical models.

Findings

Model predicts sky brightness within 3.9% accuracy.

Weather data correlates with sky brightness, aiding characterization.

Machine learning model outperforms physical Zodiacal light models.

Abstract

The sky observed by space telescopes in Low Earth Orbit (LEO) can be dominated by stray light from multiple sources including the Earth, Sun and Moon. This stray light presents a significant challenge to missions that aim to make a secure measurement of the Extragalactic Background Light (EBL). In this work we quantify the impact of stray light on sky observations made by the Hubble Space Telescope (HST) Advanced Camera for Surveys. By selecting on orbital parameters we successfully isolate images with sky that contain minimal and high levels of Earthshine. In addition, we find weather observations from CERES satellites correlates with the observed HST sky surface brightness indicating the value of incorporating such data to characterise the sky. Finally we present a machine learning model of the sky trained on the data used in this work to predict the total observed sky surface brightness. We demonstrate that our initial model is able to predict the total sky brightness under a range of conditions to within 3.9% of the true measured sky. Moreover, we find that the model matches the stray light-free observations better than current physical Zodiacal light models.