Modeling and clustering of traffic flows time series in a flood prone area

Time series of traffic flows, recovered by mobile phone origin-destination signals, are used to monitor mobility and crowding in an area subject to flooding risk. We propose a time series model based on vector autoregressive with exogenous covariates combined to dynamic harmonic regression and a subsequent clustering procedure, aimed at obtaining groups of areas characterized by the common tendency to the occurrence of extreme events, that in this case study are extremely high incoming traffic flows