Astronomical Image Quality Prediction based on Environmental and Telescope Operating Conditions

TL;DR

This paper presents a machine learning model that predicts astronomical image quality based on environmental and operational data, enabling optimized telescope scheduling and dome vent control to improve image clarity and reduce costs.

Contribution

It introduces a data-driven probabilistic model for image quality prediction and demonstrates how vent control can enhance image quality and operational efficiency.

Findings

Improved image quality by 0.05-0.2 arc-seconds through vent control.

Reduced exposure time and costs by approximately 10-15%.

First application of machine learning for observatory scheduling optimization.

Abstract

Intelligent scheduling of the sequence of scientific exposures taken at ground-based astronomical observatories is massively challenging. Observing time is over-subscribed and atmospheric conditions are constantly changing. We propose to guide observatory scheduling using machine learning. Leveraging a 15-year archive of exposures, environmental, and operating conditions logged by the Canada-France-Hawaii Telescope, we construct a probabilistic data-driven model that accurately predicts image quality. We demonstrate that, by optimizing the opening and closing of twelve vents placed on the dome of the telescope, we can reduce dome-induced turbulence and improve telescope image quality by (0.05-0.2 arc-seconds). This translates to a reduction in exposure time (and hence cost) of $\sim 10-15\%$. Our study is the first step toward data-based optimization of the multi-million dollar operations of current and next-generation telescopes.