Estimating time delays between irregularly sampled time series

Abstract

The time delay estimation between time series is a real-world problem in gravitational lensing, an area of astrophysics. Lensing is the most direct method of measuring the distribution of matter, which is often dark, and the accurate measurement of time delays set the scale to measure distances over cosmological scales. For our purposes, this means that we have to estimate a time delay between two or more noisy and irregularly sampled time series. Estimations have been made using statistical methods in the astrophysics literature, such as interpolation, dispersion analysis, discrete correlation function, Gaussian processes and Bayesian method, among others. Instead, this thesis proposes a kernel-based approach to estimating the time delay, which is inspired by kernel methods in the context of statistical and machine learning. Moreover, our methodology is evolved to perform model selection, regularisation and time delay estimation globally and simultaneously. Experimental results show that this approach is one of the most accurate methods for gaps (missing data) and distinct noise levels. Results on artificial and real data are shown.