I am working on Generalized additive models to study the effect of climate variables on the incidence of Malaria. The climate variables (Temperature, Rainfall, and Relative Humidity) are included in the model additively using penalized splines. I would like to know how I can compute the AR and AF for this setting when the exposure enters the model non-linearly. For the linear setting, it is easy as it only suffices to have the estimated parameter for that particular covariate. But for the case of non-linear function where the exposure is modeled using splines is not clear.