The results of the random forest model were generally consistent with those of the linear multiple regression model (S2 Fig). The relative importance of the variables related to human mobility and host susceptibility to COVID-19 (elderly population, BCG vaccination effect, and malaria incidence) became predominant over time, whereas the relative importance of population density and the number of days from case onset decreased after March 2020. Moreover, additional analyses, which included the number of conducted COVID-19 tests as a covariate, revealed very similar patterns of regression coefficients, and their explanatory power (S3 Fig), i.e., the roles of climate, international human mobility, and host susceptibility to COVID-19, became more pronounced as the pandemic progressed. Therefore, the nonlinearity of epidemic and region-specific testing bias had no serious influence on identifying the environmental drivers shaping the present COVID-19 distribution.