Machine Learning Based Prediction of Astronomical Seeing Using All-Sky Camera Images and Cloud Sensor Data

Astronomical seeing refers to how atmospheric turbulence degrades optical observations. This turbulence comes from temperature gradients, wind, humidity, and aerosols in the atmosphere. These seeing conditions don’t just affect astronomy they also impact space geodetic techniques like Satellite Laser Ranging (SLR) and Lunar Laser Ranging (LLR), which need stable atmospheric conditions to achieve millimeter-level precision. Traditional seeing measurements use dedicated instruments like Differential Image Motion Monitors (DIMMs). While these are accurate, they are expensive and operationally demanding, which limits how widely they can be deployed at geodetic and astronomical sites.

 

This study explores a cost-effective, data-driven approach to estimating and forecasting astronomical seeing. We do this by combining all-sky camera imagery with cloud sensor measurements from the SARAO Hartebeesthoek site in South Africa. Using machine learning methods, we aim to extract atmospheric turbulence indicators from diverse data sources. Following the CRISP-DM methodology, we have completed the business understanding, data understanding, and data preparation phases. This included aligning the timing of different datasets, performing quality control, and analyzing high-frequency image data alongside lower-resolution environmental sensor records.

 

Preliminary experiments using baseline models, including Multi-Layer Perceptrons, Random Forests, XGBoost, and Long Short-Term Memory (LSTM) networks, we have ran demonstrate encouraging capability in capturing nonlinear and temporal relationships between environmental conditions and observed seeing. Exploration of additional models, as well as efforts in uncertainty quantification and validation, are ongoing. 

 

The proposed approach aims to deliver near-real-time or short-term seeing estimates to support operational decision-making, improve scheduling efficiency, and enhance data quality for astronomical observations and emerging geodetic infrastructure, including the planned Lunar Laser Ranging facility in South Africa. By leveraging existing, lower-cost instrumentation, this framework offers a scalable and transferable solution for site characterization and operational support at current and future geodetic observatories.

Keywords: Astronomical seeing, Geodesy, Laser Ranging (SLR/LLR), Atmospheric turbulence, Site characterization