This seems important but I don't understand the algorithm -- the data must be buried in an extreme level of noise
["Estimating temporal changes in seismic velocity using a Markov chain Monte Carlo approach"](https://academic.oup.com/gji/article/220/3/1791/5645243)