Spatiotemporal Dynamics, Nowcasting and Forecasting of COVID-19 in the United States

In response to the ongoing public health emergency of COVID-19, we investigate the disease dynamics to understand the spread of COVID-19 in the United States. In particular, we focus on the spatiotemporal dynamics of the disease, accounting for the control measures, environmental effects, socioeconomic factors, health service resources, and demographic conditions that vary from different counties. In the modeling of an epidemic, mathematical models are useful, however, pure mathematical modeling is deterministic, and only demonstrates the average behavior of the epidemic; thus, it is difficult to quantify the uncertainty. Instead, statistical models provide varieties of characterization of different types of errors. In this paper, we investigate the disease dynamics by working at the interface of theoretical models and empirical data by combining the advantages of mathematical and statistical models. We develop a novel nonparametric space-time disease transmission model for the epidemic data, and to study the spatial-temporal pattern in the spread of COVID-19 at the county level. The proposed methodology can be used to dissect the spatial structure and dynamics of spread, as well as to forecast how this outbreak may unfold through time and space in the future. To assess the uncertainty, projection bands are constructed from forecast paths obtained in bootstrap replications. A dashboard is established with multiple R shiny apps embedded to provide a 7-day forecast of the COVID-19 infection count and death count up to the county level, as well as a long-term projection of the next four months. The proposed method provides remarkably accurate short-term prediction results.