Understanding and predicting the spatio-temporal spread of COVID-19 via integrating diffusive graph embedding and compartmental models.

In order to find useful intervention strategies for the novel coronavirus (COVID-19), it is vital to understand how the disease spreads. In this study, we address the modeling of COVID-19 spread across space and time, which facilitates understanding of the pandemic. We propose a hybrid data-driven learning approach to capture the mobility-related spreading mechanism of infectious diseases, utilizing multi-sourced mobility and attributed data. This study develops a visual analytic approach that identifies and depicts the strength of the transmission pathways of COVID-19 between areal units by integrating data-driven deep learning and compartmental epidemic models, thereby engaging stakeholders (e.g., public health officials, managers from transportation agencies) to make informed intervention decisions and enable public messaging. A case study in the state of Colorado, USA was performed to demonstrate the applicability of the proposed transmission modeling approach in understanding the spatio-temporal spread of COVID-19 at the neighborhood level. Transmission path maps are presented and analyzed, demonstrating their utility in evaluating the effects of mitigation measures. In addition, integrated embeddings also support daily prediction of infected cases and role analysis of each area unit during the transmission of the virus.