Climate affects global patterns of COVID-19 early outbreak dynamics

Gentile Francesco Ficetola; Diego Rubolini

The best-fitting linear mixed model suggested that r 50 is non-linearly related to spatial variation in mean temperature of the outbreak month (Fig. 1A , Tables S2-S3). Growth rates peaked in regions with mean temperature of ~5°C during the outbreak month, and decreased both in warmer and colder climates (Fig. 1A , Table S3 ). The comparison of models with different 5 combinations of predictors confirmed temperature as the variable with the highest relative importance in explaining variation of r 50 (Table S1), and temperature was the only parameter included in the best-fitting model (Tables S2-S3 ). Temperature and humidity of the outbreak month showed a strong, positive relationship across regions (Fig. S1 ), thus they could not be included as predictors in the same model. When we repeated the analyses including humidity 10 instead of temperature, r 50 varied significantly and non-linearly with humidity, peaking at ~0.6-1.0 kPa (Fig. 1B, Tables S4-S5 ). The best model including humidity also showed slightly larger growth rates in countries with greater health expenditure (Table S5) , possibly because of more efficient early reporting and/or faster diagnosis of Covid-19 cases. Results were highly consistent if we calculated growth rates after minimum thresholds of 25 or 100 cases (r 25 and r 100 , 15 respectively) instead of 50 (Tables S3 and S5 ). Human population density and air pollution showed very limited relative importance values (always < 0.50; Table S1 ), suggesting that they play a relatively minor role in determining Covid-19 growth rates, at least at the coarse spatial scale of this study.