Estimating the statistical uncertainty due to spatially correlated noise in interferometric images

TL;DR

This paper introduces a method to accurately estimate and simulate the statistical uncertainty caused by spatially correlated noise in interferometric images, improving the reliability of astronomical data analysis.

Contribution

It provides a comprehensive approach to measure, simulate, and incorporate noise correlations into uncertainty estimates and model fitting in interferometric imaging.

Findings

Ignoring noise correlation underestimates uncertainties

The method accurately characterizes noise autocorrelation functions

Proper accounting prevents false detections

Abstract

Interferometers (e.g. ALMA and NOEMA) allow us to obtain the detailed brightness distribution of astronomical sources in 3 dimensions (R.A., Dec., frequency). However, the spatial correlation of the noise makes it difficult to evaluate the statistical uncertainty of the measured quantities and the statistical significance of the results obtained. The noise correlation properties in the interferometric image are fully characterized and easily measured by the noise autocorrelation function (ACF). We present the method for (1) estimating the statistical uncertainty due to the correlated noise in the spatially integrated flux and spectra directly, (2) simulating the correlated noise to perform a Monte Carlo simulation in image analyses, and (3) constructing the covariance matrix and chi-square $\chi^2$ distribution to be used when fitting a model to an image with spatially correlated noise, based on the measured noise ACF. We demonstrate example applications to scientific data showing that ignoring noise correlation can lead to significant underestimation of statistical uncertainty of the results and false detections/interpretations.