Non-stationarity between driver:response

[ericward-noaa]

This is probably beyond paper # 1, but I was thinking at some point it might be interesting to simulate threshold relationships that are changing slowly through time (either in response to the driver, or some other covariate). The estimation model would also have to change slightly -- from the 1-D smooth, to a 2-D smooth where time was a continuous or factor variable in the smooth.

See also this kind of approach https://peerj.com/articles/6876/