Expressive GP Kernel Exploration

Background: Predictive modeling of atmospheric seeing requires approaches that can effectively capture complex, non-linear, and time-dependent phenomena, moving beyond traditional reactive compensation methods.

Question / Future Work: Explore the utility of more expressive Gaussian Process (GP) kernels that can capture multi-scale temporal structures in the seeing data while retaining the framework’s interpretability and calibration benefits, contrasting with the simple exponential kernel used in the current work.