Fig. 4.

Relative change in R₀ due to pollution variations observed in USA states. For each state included in the analysis, R₀ was predicted for the range of pollution values observed throughout all other states. Relative variation in R₀ was estimated through both A) Gradient Boost (GBoost) and B) Random Forest (RF) regressions, with the models trained as in Fig. 3C) Scatter plot of ΔR₀/R₀ predictions for GBoost and RF, with indicated Pearson’s correlation coefficient and P-value.