Need help interpreting time-series model (annual trend confounded)

This is a follow-up question of my OP here , with initial model and data. After learning that my spatial terms in te(CYR, Latitude, by = fSeason, k = c(5,3)) and s(fSite, bs="re") were actually strongly correlated, I reduced the temporal component to s(CYR, by = fSeason) + fSeason . The reduced model has some unaccounted for temporal auto-correlation, and since bam() , gam() , gamm() , and gamm4() can't model irregular spaced neg. binomial data that is auto-correlated (like in glmmTMB with the ou() function), correct me if I'm wrong, then I'm stuck with what I have now (which is fine). The auto-correlation doesn't seem that strong, even a simple random year effect s(fCYR, bs="re") takes care of it, but of course that term is highly correlated with my main year effect (with and without by= interaction). My question is: how do I interpret an annual trend that is mostly accounted for (goes away) with an auto-correlation structure? Do I simply replace the main s(CYR) term with a factor year random effect s(fCYR) and conclude there is no overall annual trend? I'd like to keep the non-significant fSeason component to show it had no strong signal in my dataset. The annual trend however has a cyclic pattern that I'm having trouble explaining. I'm wondering if its an artificat of the data collection frequency (5-7 days of surveys per season seperated my months of no data, glued together by the smooth s()). What would make sense to conclude from this model? Interpret as per usual and…