Probabilistic Machine Learning for Short-Term Atmospheric Seeing Forecasting

Key points

• 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.1
• 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.1
• 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.1
• FloTS combines competitive forecasting accuracy with flexible uncertainty quantification and produces the most informative probabilistic forecasts.1

Sources (1)

• [1]Probabilistic Machine Learning for Short-Term Atmospheric Seeing Forecasting
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 21, 12:35 PM

  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.

Extractive summary: sentences quoted from the sources.