Semantics of passing lagging information

[mpartio]

For a while I was confused because when I was running read_eps_interpolate() for a lagged ensemble, it complained that some forecasts were not found and eventually the sqlite files for certain cycles were always empty. I did verify that the vfld-files for these cycles were present.

I had specified lagging counting from "parent" cycle forwards, but after I changed it to counting backwards all files were found correctly.

So I just want to confirm that my understanding is correct.

For example: Member 0 is starting at 0,3,6,...,21 Member 1 is starting at 1,4,7,...,22 Member 2 is starting at 2,5,8,...,23

Member 0 is the parent of member 1 and 2. Initially I specified the lags for members 1 and 2 to be c(1h,2h), but experimentally I found that the correct values are c(2h,1h).

I'm not sure how to interpret this. Data is read backwards from 00 cycle, so member 1 has 00-2 = 22, and member 2 has 00-1 = 23?

However, for a lagged forecast, we are interested in the child members that belong to the previous forecast, since the child members from the current parent will not be available yet.

https://andrew-met.github.io/cmeps.nb.html

[andrew-MET]

For creating lagged ensembles one would always look backwards in time. The confusion arises because CMEPS "creates" members forwards in time.