SLAC, Kavli 3rd Floor Conf. Room
Accurately modeling and interpolating the PSF across the full field of view (FoV) of LSST for a given exposure is crucial for making galaxy shape measurements that meet the requirements of cosmic shear. In current weak-lensing surveys, the PSF is estimated at each star location within a CCD chip and interpolated to the locations of the galaxies. For example, for the Dark Energy Survey Year 1 data, a second-order polynomial is used as the interpolation function for each PSF parameter. A limitation of treating each CCD chip separately is that spatial correlations larger than the chip size — due to optics and the atmosphere, for example — are not captured, leading to less sensitive and possibly less accurate modeling of the PSF at galaxy locations. In particular, PSF correlations that are not accurately modeled can lead to biased estimators of cosmic shear. In this presentation, we describe how Gaussian processes (GP) can be used to model the PSF over the full FoV, to capture spatial correlations on scales as large as the FoV itself. We show how we have implemented the GP so that the covariance function (the kernel) for the PSF parameters can be modeled quickly and robustly for the expected number of usable stars (~10^4). We present first interpolation results on realistic simulations of the atmosphere using GalSim.