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

introduction, we include the statistical analysis as a baseline, leverage the machine learning analysis for short-term prediction, and use the econometric results as an indication of long- term trends. SHAP values for meteorological and environmental factors other than UV tend to cluster very closely (see Fig 6), so their relative ranking points to relatively mild impact differences. Also, only pressure, temperature, UV, stringency, and containment factors show sufficient sta- tistical significance (p < 0.10) in the econometric results. Results for rainfall, absolute humid- ity, wind speed, short-wave irradiation, PM2.5, and PM10 where p > 0.10 can only be seen as weak indicators. UV emerges as the most impactful meteorological factor in COVID-19 transmission across all methods. More specifically, UV is negatively correlated with COVID-19 spread. This result is corroborated by overwhelming evidence that UV light can effectively kill SARS-CoV-2 and other coronaviruses [82–84]. Temperature is positively correlated with COVID-19 transmission in both the machine learning and econometric analyses. As discussed in the introduction, there is contrasting evi- dence about the role of temperature in the spread of COVID-19. The inverse correlation of UV and temperature with COVID-19 spread emerging in our analysis suggests that the inhib- iting factor in those studies where the temperature is negatively correlated with COVID-19 spread may not temperature per se, but rather the high UV that is often found in hotter cli- mates. This is a hypothesis that requires further inspection.

Absolute humidity is positively correlated with COVID-19 spread in the machine learning analysis and negatively correlated in the econometric analysis, though with insufficient statisti- cal significance. This is an interesting contrast as there is conflicting evidence in the literature about the role of absolute humidity in the spread of COVID-19, as discussed in the introduc- tion (S1 Table). Perhaps, the two analyses point to a different role that absolute humidity may play in the short and long term with reference to COVID-19 spread. This too is a hypothesis that requires further inspection. Table 6. Coefficients and relative rank describing the impact of climate factors on COVID-19 transmission across the three types of analysis carried out in this study. Statistical analysis (Spearman’s coefficient) Machine learning analysis (SHAP values) Econometric analysis (Panel data fixed effects model) Temperature 0:42 3 ð Þ 14:7 2 ð Þ � � �p > 0:10 5 ð Þ Absolute Humidity 0:39 4 ð Þ 11:6 4 ð Þ p > 0:10 7 ð Þ Pressure 0:19 8 ð Þ 14.5 (3) � � �p < 0:001 3 ð Þ Wind speed 0:21 6 ð Þ 10.4 (8) p > 0:10 6 ð Þ Rainfall 0:21 7 ð Þ 10:3 9 ð Þ p < 0:10 8 ð Þ Short-wave irradiation 0:55 2 ð Þ 10:9 5 ð Þ p > 0.10 (11) PM2.5 0:24 5 ð Þ 10.5 (7) p > 0:10 9 ð Þ PM10 0:17 9 ð Þ 10:8 6 ð Þ p > 0:10 10 ð Þ UV 0:56 1 ð Þ 18:1 1 ð Þ � � p < 0:001 4 ð Þ Stringency 8; 8 10 ð Þ � � �p < 0:001 2 ð Þ Containment 7:9 11 ð Þ � � �p < 0:001 1 ð Þ

Red fonts indicate negative correlations, blue positive correlations, and black undetermined polarity. The integers enclosed in parentheses describe relative rank (1 = highest, 11 = lowest). Stringency and containment results are not available in the statistical analysis. https://doi.org/10.1371/journal.pone.0273078.t006 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 21 / 27