Students’ t-distribution modelling of LLS location errors

This paper investigates the statistical distribution that best describes Lightning Location System location errors. It is generally assumed that the bivariate Gaussian distribution is a good description but the presence of outliers calls for a heavier-tailed distribution. The bivariate Students' t-distribution is proposed. Using Maximum Likelihood Estimation, Mahalanobis quality-of-fit tests and ground-truth events to the Brixton tower, South Africa and Gaisberg tower, Austria, we are able to show that the bivariate Students' t-distribution is a better fit than the bivariate Gaussian distribution.