Hot spot analysis is the problem of identifying statistically significant spatial clusters from an underlying data set. In this paper, we target the problem of hot spot analysis of massive spatio-temporal data, which raises the need for a parallel and scalable solution that operates on data distributed over a set of nodes. We propose an algorithm, called BigCAB, implemented in Spark, that solves the problem in a parallel and scalable way. Our experiments on real data representing taxi trips demonstrate both the efficiency as well as the nice scaling properties of our algorithm.