As mentioned in #1, autocorrelation is already rather advanced and maybe shouldn't be the first example for mixed models.
Using the ISIT data, it would be straightforward to implement a simple mixed effects model with gam( ... s(Station, bs = re) ...).
