Why all the use of trend?

[seabbs]

https://github.com/kaitejohnson/cityforecasts/blob/d21df1531f02b6b3230f3200f64218fbe9b24336/models.R#L92

My expectation here is that we would need a season + location specific intercept. That sort of approaches seems to be taken in the examples so what does this mean?

My impression is that this is putting a random effect on the trend? I don't really get that

[kaitejohnson]

I think this is unnecessary, but in theory you could have a different mapping from your observed series to your latent trend (using trend_map()). In this case, I think it's overkill, since the groupings of the observations are the same groupings we want estimate in the latent process.

I just estimated a location specific intercept to start

[seabbs]

I just estimated a location specific intercept to start

and this is implicitly defined by trend? Its not clear to me that the trend mapping works like that but its also not clear to me how it works

[kaitejohnson]

I don't think its completely clear to me, but yes, on one of the tutorials I took these notes from the text (can now not seem to find it)

         # Use trend rather than series because can share same 
         # latent process model for multiple observation time series
         # everything happening on process model

[seabbs]

I still don't really understand this. What happens if you define the spline on variables in the model? Given the lack of clarity it makes me worry it is not doing what we think.