AI and the World's Hard Problems · Earth & climate science · arXiv · 2026-10-22
Probabilistic Machine Learning for Short-Term Atmospheric Seeing Forecasting
The study benchmarks machine learning models that forecast atmospheric seeing up to two hours ahead, where rapid variations in optical turbulence affect astronomical observations and free-space optical communication. Gaussian Processes and the normalizing-flow-based FloTS were compared against deterministic RNN and LSTM baselines using seeing measurements from the Maunakea summit. Deep learning approaches outperformed standard recurrent networks, with LSTM and FloTS achieving the best overall accuracy, and while Gaussian Processes were limited by their Gaussian assumption, FloTS combined competitive accuracy with flexible uncertainty quantification to produce the most informative probabilistic forecasts.