Consider a monthly time series dataset for which we believe that the model Xt = (at + b)st + Zt is appropriate where st is a seasonal function i.e., st+d = st and Zt is white noise. What would be a way to difference the data in order to eliminate both trend and seasonality and why does it work?