Fig. 3

The panels display the estimated partial effect of temperature and humidity based on the inverse Gaussian generalized additive model: \( \begin{array}{l} \log \left({\mu}_t\right)={\alpha}_0+ \log \left({n}_t\right)+{f}_1\left({\mathrm{PM}}_{2.5, lag01}\right)I\left({\mathrm{flu}\ \mathrm{season}}_{\mathrm{t}}\right)+{f}_2\left({\mathrm{PM}}_{2.5, lag01}\right)I\left({\mathrm{nonflu}\ \mathrm{season}}_t\right)\\ {} + {f}_3\left({\mathrm{t}\mathrm{emperature}}_{\mathrm{t}}\right)+{f}_4\left({\mathrm{humidity}}_{\mathrm{t}}\right)+{f}_5\left({\mathrm{month}}_{\mathrm{t}}\right) + {\displaystyle {\sum}_k{\beta}_kI\left({\mathrm{year}}_{\mathrm{t}}=k\right).}\end{array} \) The x-axis tick labels in the panels represent the observed values temperature and humidity, respectively. The solid lines indicate the estimated log relative risk of ILI and the dashed lines indicate the corresponding 95 % confidence intervals