The identification of spatially constrained homogeneous clusters of Covid‐19 transmission in Italy

This paper introduces an approach to identify a set of spatially constrained homogeneous areas that are maximally homogeneous in terms of epidemic trends. The proposed hierarchical algorithm is based on the dynamic time warping distances between epidemic time trends, where units are constrained by a spatial proximity graph. Two different applications of this approach to Italy are presented, based on different data (number of positive tests and number of differential deaths with respect to previous years) and different observational units observed at different spatial scales (provinces and labour market areas). The provincial analysis was mainly used to divide the national territory into macro-areas with different contagion trends, while the more detailed partition was carried out only for the macro-areas with higher risk of transmission of the infection. Both applications, above all that related to labour market areas, show the existence of well-defined areas where the dynamics of growth of the infection have been strongly differentiated. The adoption of the same lockdown policy throughout the entire national territory has been therefore sub-optimal, highlighting once again the urgent need for local data-driven policies.