The lock-down communities management

Abstract

The SARS-CoV-2 pandemic (COVID-19) has highlighted the need to better understand the dynamics and evolution of epidemics. Moreover, the unprecedented amount of information collected about it, at different levels of resolution, motivates a data-based modelling of pandemics. Besides, companies like Meta (Facebook) have made available co-location and movement data based on cell-phone tracking of their users. These data are completely anonymous, and can be used in an aggregated form to estimate probabilities of movement of people in between different areas, at different scales. Numerous studies in statistical physics have shown that it is possible to map an epidemic dynamic onto a network, whose nodes correspond to persons or groups, and links the contacts between them. To construct such a meta-population network, the gathering, analysis and reduction of mobility data are crucial to seize what are the main circulations happening on a territory.
Using people mobility data from Meta’s «Data for good» program and publicly available information on the Italian population from the National Institute of Statistics, we model the circulation of people in Italy during the two-year period (2020-2022) of the SARS-CoV-2 pandemic. We perform a spatial and a temporal clustering of the Italian movement network at the level of fluxes across provinces on a daily basis; the resulting partition in time successfully identifies the first two lockdowns without any prior information; similarly, the spatial clustering returns 11 to 23 clusters depending on the period («standard» mobility vs. lockdown). The spatial clusters largely overlap with partitions of the Country in macro-regions that have been previously suggested based on the infrastructure and the current commercial exchanges.
This work introduces and validates a data analysis pipeline that enable us: i) to assess the reliability of data obtained from a partial and potentially biased sample of the population (namely Facebook users sharing their location) in performing estimates of population mobility nation-wide; ii) to identify areas of a Country with well-defined mobility patterns, i.e, characteristic circulations on the territory; iii) to distinguish different patterns from one another, resolve them in time and find their optimal spatial extend. These results can thus represent a relevant step forward in the development of simulations, methods and strategies for the containment of future epidemic phenomena.