Spatio-temporal communities in networks of extreme rainfall events of the South Asian Monsoon

Climate networks have recently helped to unravel spatial patterns of extreme rainfall events (EREs). However, EREs remain challenging to analyse due to their inherent stochasticty involved in their local distribution and intensity. Here, we present a principled approach to identify regions of similar ERE dynamics with the idea that this will help developing ERE prediction schemes in a later study.
We use a probabilistic approach to quantify community structures and estimating the structural uncertainties involved in the community detection process. First, we use time-delayed event synchronization to construct a network of ERE teleconnections. Using a Bayesian hierarchical community detection algorithm based on the Stochastic Block Model (SBM) enables us to estimate the likelihood that spatial locations belong to the same community via a point-wise spatial density analysis.
We apply our method to the South Asian Monsoon system and reveal a latitudinal band-like structure of synchronous EREs, whose occurrence is consistent with the onset and withdrawal of the monsoon season. Moreover, we demonstrate that exceptionally strong synchronization is observed when a fast developing low pressure system over the South China Sea occurs, as demonstrated by climatologies of days with maximum synchronization.