Impact of climate change on SARS-CoV-2 epidemic in China

Contribution rate analysis The Pearson correlation coefficient is typically used in correlation analyses to analyze linear correlations. Since a delay occurs between the initial SARS-CoV-2 exposure and infection and the effect of weather on the confirmed cases was not instantaneous, a lag occurred. The lag model was used to calculate the weather factors and GRs of the confirmed cases, then the cumulative effect model was used to calculate the contribution rate. A lagged variable affects the explanatory variable in the past. The lagged variables include the lagged explanatory and explained variables. The lagged variables were substituted into the regression model, i.e., the lagged variable model. The distribution lag model was selected to quantify the degree of influence of weather factors on SARS-CoV-2, and the general expression is shown in Eq (5). Yt ¼ a þ b0Xt þ b1Xt 1 þ b2Xt 2 þ � � � þ bsXt s þ ut ð5Þ where s is the lag length, t is the time, and each coefficient reflects the explanatory variable’s different degrees of influence of each lag value on the explained variable the multiplier effect. β0 is the short-term or immediate multiplier and represents the average size of the impact of a one-unit change in X on Y in the current period. βi is the delayed or dynamic multiplier, which represents the average size of the effect of a one-unit change in X on Y in past periods. Xs i bi is the long-term or total distribution multiplier, which represents the size of the full effect on Y due to the lag effect when X changes by one unit.

In the estimation process, multiple collinearities might be exhibited between explanatory variables, which can result in meaningless parameter estimation and failure to reveal X’s effect of each lag on the dependent variable. Therefore, polynomial distributed lag models, including the Almon distributed lag model [33, 34], were used to reduce the number of parameters to be estimated and smooth the lag coefficients. The six weather variables selected in this study were modeled with distribution lags, as shown in Eqs. (6), where λ is the SARS-CoV-2’s GR; i is the provincial administrative region; t is the specified time; l is the lag order (number of days); and UV, T, H, TP, WS, and SP are the weather variables. li t ¼ X l¼L l¼0 aUV l UVi;t l þ X l¼L l¼0 aT l Ti;t l þ X l¼L l¼0 aH l Hi;t l þ X l¼L l¼0 aTP l TPi;t l þX l¼L l¼0 aWS l WSi;t l þ X l¼L l¼0 aSP l SPi;t l þ y0Zit þ �it ð6Þ Spatial analysis Using spatial autocorrelation methods, spatial analysis was performed to investigate whether the SARS-CoV-2 epidemic and weather factors in each provincial administrative region exhib- ited spatial clustering characteristics. Moran’s I is a typical spatial autocorrelation analysis method with strong applicability that can test whether a statistically significant spatial distribu- tion exists and reveal the spatial process for generating this distribution [34], which can be classified into global and local. Global spatial autocorrelation assumes that the space is homo- geneous, where only one trend fills the entire region [35]. It can only detect whether the

research object has aggregated in the research region and cannot accurately determine the aggregation range. Therefore, local spatial autocorrelation is required to determine the specific aggregation region and range. PLOS ONE Impact of climate change on SARS-CoV-2 epidemic in China PLOS ONE | https://doi.org/10.1371/journal.pone.0285179 July 27, 2023 7 / 21