PREDICTING CHROMATIC POINT SPREAD FUNCTION USING MACHINE LEARNING

The Vera C. Rubin Observatory’s Legacy Survey of Space and Time (LSST) will probe the nature of dark energy by measuring weak gravitational lensing across billions of galaxies, enabling precise constraints on the growth of cosmic structure and the geometry of the Universe. The precision of LSST will require a greater understanding of the systematic biases than we have achieved to date. A leading systematic bias that is only partially corrected in current weak lensing measurements is due to chromatic effects in the atmospheric point spread function (PSF), including atmospheric differential chromatic refraction and wavelength-dependent seeing. In this thesis, we focus on the wavelength dependence of the size of the PSF due to atmospheric turbulence, which is called chromatic seeing. Since the effect of chromatic seeing on PSF size depends on the spectral energy distribution of the star or galaxy, we investigate how photometry in the six Rubin bands (ugrizy filters) can be used to predict biases between stellar and galactic PSF sizes due to chromatic seeing. The task of predicting a scalar quantity (PSF size in a band) using high-dimensional data (six-band photometry) can be tackled with machine learning. We implement a self-organizing map (SOM), an unsupervised machine learning algorithm, to identify correlations and patterns in simulated Rubin six-band photometric data and use it to predict chromatic seeing. For comparison, we implement an Extremely Randomized Trees Regressor (ETR), a supervised machine learning algorithm that builds on the simple decision tree by introducing ensemble learning and randomness. We find that the (supervised) ETR performs extremely well at predicting the effects of chromatic seeing on the PSF with simulated data, as seen in previous studies. Although the (unsupervised) SOM does not perform as well as the ETR, we show that it successfully captures patterns and relationships in six-band photometric data and does well predicting the effects of chromatic seeing on the PSF. This thesis demonstrates that the SOM is a viable method for mapping the relationships between six-band photometric data and predicting the effects of chromatic seeing on the PSF.