Consider intermediate between random effect of year and full DLNM

[Aariq]

Use DLNMs to identify important lag time windows (i.e. what we did in Global Change Bio) but rather than using DLNMs in the IPM, make vital rates models that include averages of SPEI in those periods as simple parametric predictors. E.g. something like surv ~ s(log_size_prev) + spei_wet_t2 where spei_wet_t2 is the average SPEI in the wet season two years prior to the census.

[Aariq]

this would still use parameter resampling, but would be a lot faster to iterate the IPM because calls to predict() would be faster.