The impact of meteorological factors on the epidemic of COVID-19 remains controversial. Most of previous studies reported that cold and dry climate conditions were conducive to the transmission of COVID-19 [
24‐
26], while some showed that high temperature could not inhibit the transmission of COVID-19 [
27] or that there was no significant correlation between temperature and COVID-19 infection [
28,
29]. There are several reasons to explain these contradictory results. First, different research subjects may lead to different results, such as regional studies versus global studies. Second, many previous studies failed to cover all meteorological conditions. In addition, different research methods may lead to different results [
27]. In this study, we aimed to capture common patterns or discrepancies of the relationship between meteorological factors and the COVID-19 epidemic among individual countries. A climate-dependent epidemic model showed that meteorological variables were unlikely to be dominant transmission risk factors in the early stages of the COVID-19 pandemic due to the high population susceptibility [
30]. Besides, in order to minimize the influences of the variation of SARS-CoV-2 [
31] and to obtain the maximum range of meteorological data, we analyzed data from July 1, 2020, to June 30, 2021, in eight countries from four continents. Spearman’s correlation analysis showed that temperature and UV index were negatively correlated with COVID-19 prevalence in 7 and 8 countries, respectively. The correlation between relative humidity and COVID-19 prevalence showed positive correlation in 4 countries and negative correlation in 3 countries. Thus, we included the three meteorological factors (temperature, relative humidity, and UV index) in the DLNM for risk analysis, respectively. Our results showed a significant non-linear relationship between temperature and COVID-19 prevalence. Portugal, Greece, South Korea, and Japan are in the Northern Hemisphere and have similar latitudes. Three of them (Portugal, South Korea, and Japan) had a higher risk of COVID-19 infection at low temperature (< 5 ℃), and all the four countries also had a higher risk of infection at high temperature (> 25 ℃) with certain lag days. South Africa, Paraguay, and Uruguay are all in the Southern Hemisphere with similar latitudes, and their highest risk of infection occurred at around 15 ℃. In addition, there was a significant lag effect of temperature on COVID-19 prevalence, with the lag time for the occurrence of the highest RR longer than the estimated mean incubation period of COVID-19 in all the eight countries. We also proved that the pattern of risk effects of relative humidity on COVID-19 infection largely depended on the variation range of year-round relative humidity in countries. Lower relative humidity was associated with higher COVID-19 prevalence in countries with a wide range of relative humidity, while the relative risk of COVID-19 infection in high relative humidity could be high in countries with overall high relative humidity. Moreover, the lag effect of relative humidity generally lasts for a long time. Nevertheless, the non-linear effects of UV index on COVID-19 prevalence were polycentric and varied across countries. The potential reason could be that our UV index data only represented the outdoor air conditions, while epidemiological tracing reports indicated that the infection rate indoors was much higher than that outdoors [
32].
Since the first launch of COVID-19 vaccines in December 2020, around 65% of the world population has received at least one dose of a COVID-19 vaccine. The protective effect of vaccines can slow down the COVID-19 transmission [
33,
34]. Besides, the adjustment of government prevention and control policies has played a crucial role in containing the spread of COVID-19 [
35]. In addition, wearing masks and social distancing have a direct effect on controlling the spread of COVID-19 [
36,
37]. Therefore, our model included the related variables, such as cumulative vaccination rate, government response stringency index, face coverings policies, and Google mobility trends. Although we controlled the above factors, there are still some limitations. First, the number of DNCCs of COVID-19 notified by the official authorities may be omitted. Second, meteorological data were measured by remote sensing, which represented the average meteorological value of a country. The larger the country area, the less likely the meteorological data are accurate. Finally, we only included one meteorological factor at a time in constructing the DLNM, while the real exposure condition was a combination of various meteorological factors. As a result, the dominant meteorological factor cannot be identified.