Cosmic-ray anisotropy: sensitivity of methods and implications for KASCADE data

We study the problem of measuring the anisotropy of high-energy cosmic rays across all angular scales. The limited field of view and non-uniform exposure of ground-based cosmic-ray experiments reduce their sensitivity to real anisotropy of cosmic-ray arrival directions. A widely used signature of anisotropy -- a dipole of the flux expansion over the right ascension -- provides very limited understanding of the underlying physics of cosmic-ray origin and propagation. In this study we test two other methods: the angular power spectrum and the autocorrelation function for sensitivity to possible medium- and small-scale anisotropies of cosmic-ray flux. We find that the autocorrelation function is the most sensitive estimator for an underlying physical anisotropy in most of the models tested, while the angular power spectrum can provide additional knowledge about cosmic-ray flux properties, when the anisotropy is strong enough. As a test of our findings, we apply these methods to $10\%$ sample of the KASCADE experiment public data. Namely, we consider all-particle set of events and sets of individual mass groups classified by a convolutional neural network. We find an indication of anisotropy at $> 2.5 \sigma$ level at $\sim 10^\circ$ angular scale for the iron nuclei mass group at $E \gtrsim 20$ PeV with both the angular power spectrum and the autocorrelation methods.