Spatially Explicit Modeling of 2019-nCoV Epidemic Trend based on Mobile Phone Data in Mainland China

Xiaolin Zhu; Aiyin Zhang; Shuai Xu; Pengfei Jia; Xiaoyue Tan; Jiaqi Tian; Tao Wei; Zhenxian Quan; Jiali Yu

To examine the goodness of model fitting, we calculated R-squared and root mean square error (RMSE) between the fitting result and confirmed cases in each city (see representative examples in Figure 4 ). The median value of R-squared and RMSE across all cities is 0.96 and 1.20 respectively, indicating that our model can well fit the current spreading trend. We examined the fitted models with R-squared less than 0.7 (26 out of 306 cities) and found that all these cities have very small number (1-3) of confirmed cases and the number does not change in the past several days. It is reasonable to assume that these cities have completely controlled the spreading and no new infected cases will emerge in the future. Therefore, for these cities with R-squared less than 0.70, the 1 value of the city was set to zero. Transmission rate 2 estimated using Wuhan confirmed cases is 0.9, reflecting a high transmission around January 20. The transmission rate 1 , decay rate a, and initial infectious population 0 vary from city to city (Supplementary data Table 3 ). As the transmission rate among local residents in each city, 1 reflects the intensity of control measures adopted by each local government at the beginning of outbreak, as well as the awareness of citizens to take protective measures. For example, 1 in megacities such as Beijing, Shanghai, Guangzhou and Shenzhen are low ( Figure 5 .a) which may attribute to higher health literacy of their citizens 21 , although they have intensive traffic and population mobility. Decay rate a reflects the continuous effort input by each city to control the transmission of disease. It shows that decay rates of cities close to Wuhan is generally lower than other cities ( Figure 5 .b), suggesting the big challenge faced by these cities to control the disease spreading.