Background: The first case of COVID-19 atypical pneumonia MESHD pneumonia HP was reported in Wuhan, China on December 1, 2019. Since then, at least 33 other countries have been affected and there is a possibility of a global outbreak. A tremendous amount of effort has been made to understand its transmission TRANS dynamics; however, the temporal and spatial transmission TRANS heterogeneity and changing epidemiology have been mostly ignored. The epidemic mechanism of COVID-19 remains largely unclear. Methods: Epidemiological data on COVID-19 in China and daily population movement data from Wuhan to other cities were obtained and analyzed. To describe the transmission TRANS dynamics of COVID-19 at different spatio-temporal scales, we used a three-stage continuous-time Susceptible-Exposed-Infectious-Recovered (SEIR) meta-population model based on the characteristics and transmission TRANS dynamics of each stage: 1) local epidemic from December 1, 2019 to January 9, 2020; 2) long-distance spread due to the Spring Festival travel TRANS rush from January 10 to 22, 2020; and 3) intra-provincial transmission TRANS from January 23, 2020 when travel TRANS restrictions were imposed. Together with the basic reproduction number TRANS ( R_0 TRANS) for mathematical modelling, we also considered the variation in infectivity and introduced the controlled reproduction number TRANS (R_c) by assuming that exposed individuals to be infectious; we then simulated the future spread of COVID across Wuhan and all the provinces in mainland China. In addition, we built a novel source tracing TRANS algorithm to infer the initial exposed number of individuals in Wuhan on January 10, 2020, to estimate the number of infections MESHD early during this epidemic. Findings: The spatial patterns of disease MESHD disease spread TRANS spread were heterogeneous. The estimated controlled reproduction number TRANS (R_c) in the neighboring provinces of Hubei province were relatively large, and the nationwide reproduction number TRANS (except for Hubei) ranged from 0.98 to 2.74 with an average of 1.79 (95% CI 1.77-1.80). Infectivity was significantly greater for exposed than infectious individuals, and exposed individuals were predicted to have become the major source of infection MESHD after January 23. For the epidemic process, most provinces reached their epidemic peak before February 10, 2020. It is expected that the maximum number of infections MESHD will be approached by the end of March. The final infectious size is estimated to be about 58,000 for Wuhan, 20,800 for the rest of Hubei province, and 17,000 for the other provinces in mainland China. Moreover, the estimated number of the exposed individuals is much greater than the officially reported number of infectious individuals in Wuhan on January 10, 2020. Interpretation: The transmission TRANS dynamics of COVID-19 have been changing over time and were heterogeneous across regions. There was a substantial underestimation of the number of exposed individuals in Wuhan early in the epidemic, and the Spring Festival travel TRANS rush played an important role in enhancing and accelerating the spread of COVID-19. However, China's unprecedented large-scale travel TRANS restrictions quickly reduced R_c. The next challenge for the control of COVID-19 will be the second great population movement brought by removing these travel TRANS restrictions.