Probabilistic machine learning predicts atmospheric seeing with uncertainty
Probabilistic Machine Learning for Short-Term Atmospheric Seeing Forecasting
Machine Learning
Summary
Atmospheric seeing refers to how much the atmosphere makes stars and other space objects look blurry, which affects telescopes and optical communications. This paper compares different machine learning methods to predict this blurriness up to two hours ahead. The authors find that deep learning models, especially one called FloTS, can not only predict seeing better than older methods but also provide useful estimates of how confident those predictions are. This can help plan observations or communications that rely on clear atmospheric conditions.
What this means in practice
- •For astronomical observatories: Use machine learning forecasts with uncertainty estimates to schedule telescope use more effectively under changing atmospheric conditions.
- •For optical communication engineers: Incorporate short-term seeing forecasts to adjust and optimize free-space optical communication systems in response to rapid atmospheric changes.
Tested on one dataset.
Authors
Mary-Joe Medlej, Rahul Srinivasan, Simon Prunet, Aziz Ziad, Christophe Giordano
Abstract
Accurate short-term forecasting of atmospheric seeing is of significant interest for astronomical observations and free-space optical communication systems, where rapid variations in optical turbulence can affect system performance and operational planning. In this work, we benchmark machine learning models for predicting atmospheric seeing up to two hours in advance, with a particular focus on probabilistic forecasting and uncertainty quantification. We compare two probabilistic approaches, Gaussian Processes (GPs) and a normalizing-flow- based model (FloTS), against deterministic recurrent neural network (RNN) and long short-term memory (LSTM) baselines. All models are trained and evaluated on seeing measurements from the Maunakea summit. We assess the models using both deterministic metrics, including the root mean square error and the Pearson correlation coefficient, and probabilistic metrics that evaluate the quality and calibration of predictive uncertainty estimates. The results show that deep learning approaches outperform standard recurrent networks for short-term forecasting, with LSTM and FloTS achieving the best overall predictive accuracy. While Gaussian Processes provide a useful probabilistic baseline, their predictive distributions are limited by the Gaussian assumption. FloTS combines competitive forecasting accuracy with flexible uncertainty quantification and produces the most informative probabilistic forecasts.