Does climate help modeling COVID-19 risk and to what extent?

PM10, stringency, and containment all show the same polarity in the machine learning and econometric analysis, though only for containment and stringency the econometric analysis yields statistical relevance (p < 0.001). Discrepancies between results in the statistical analysis and the machine learning and econometric analyses emphasize the difficulty in statistical analysis to deal with non-stationary processes and account for long terms trends. Perhaps, the most interesting discrepancies are those between machine learning and econometric analysis. As discussed in the introduction, machine learning is better equipped to take advantage of structural heterogeneity in training data to make short-term predictions, whereas econometric methods are better at capturing long-term trends [73]. Therefore, diverging results across the two methods may be indicative of short-term vs. long-term impacts. Conclusions Overall, disease susceptibility is the main factor driving the pandemic growth. Compliance with lockdown and restrictions policies and regulations and increased testing are the most effective strategies for disease control and COVID-19 spread prevention. For example, various studies have reported that interventions such as restrictions on mass gatherings, school clo- sures, and social distancing measures are strongly associated with a decrease in the COVID-19 transmission growth rate [33, 38, 39, 64, 85–87]. The correlation of COVID-19 transmission with climate factors provides a valuable complementary diagnostic that sheds light on the sea-

sonal characterization of the pandemic and helps refine measures to contain and prevent the spread of COVID-19. More specifically, weather forecasts could help predict new cycles of the pandemic and future outbreaks and thus contribute to the definition of ad-hoc measures that limit the economic impact of complete lockdowns. This study also extends the reach of earlier studies (S1 Table) on the relationship between COVID-19 transmission and climate factors by assessing how climate helps modeling COVID-19 through systematic validation using statisti- cal, feature importance, and econometric analyses. Such validation is crucial in proving which are the contributing factors and their relative magnitude and direction of change. Supporting information S1 File. (DOCX) S1 Fig. Statistical distribution of the share of population emerging from the sampling approach of meteorological conditions adopted in this study. On the left is a frequency his- togram of the population share covered by the sampling approach for each county. For most of the countries the share is above 75% and for only few of them is less than 20%. On the right side of the figure is a scatter plot of the share for each location considered in the study as a function of the overall population size. (TIF) S2 Fig. Summary diagram of the methodological approach. (TIF) S3 Fig. Feature importance summary plot for the north and south hemispheres. Mean absolute SHAP value (in log scale) of each variable showing the average impact on the model

output magnitude for the locations in the north (left) and south (left) hemispheres. (TIF) PLOS ONE Does climate help modeling COVID-19 risk and to what extent? PLOS ONE | https://doi.org/10.1371/journal.pone.0273078 September 7, 2022 22 / 27