Synthetic Spatiotemporal Covid19 Infection Counts to Assess Graph-Regularized Estimation of Multivariate Reproduction Numbers

The heavy impacts of Covid19 pandemic triggered significant research efforts to monitor the virus transmission. Several strategies were devised to estimate the reproduction number, quantifying the pandemic intensity, jointly along time and across territories while being robust to the limited quality of the reported Covid19 infection counts. However, because the true evolution of the pandemic intensity is unknown (lack of ground truth) estimation performance assessments and comparisons are impaired. The first contribution of this work is thus to design an original graph-based regularization strategy for the construction of spatially correlated synthetic ground truth reproduction number time series, further enabling the synthesis of realistic spatiotemporal infection counts. A second contribution consists in using such synthetic counts to compare the performance of several state-of-the-art reproduction number estimators, showing the superiority of multivariate estimation strategies compared to univariate procedures.