A new ensemble format was implemented and introduced in May 2021, due to the IM-type limitations of the old one (as it required each branch to support all IM and all tectonic types).
The updated format solves this by adding an initial split based on the IM type, i.e. for each supported IM of the ensemble a sub-ensemble exists (referred to as IM-Ensemble from here on).
The figure above shows the input logic trees for each IM-type on the left, with each of the tree then further splitting based on tectonic type and the weights summing up to one for each tectonic type.
The drawing on the right, shows the Ensemble generated from the logic trees on the left. There is an initial split based on the IM-type and the branches for each IM-Ensemble are then flat (i.e. no further split
on tectonic type). This means that each branch must cover all tectonic types for the IM type of the parent IM-Ensemble. Weights of the branches of a IM-Ensemble sum to one.
Details
IMs:
The IMs exposed by the ensemble are determined based on the available IM-Ensembles
Stations:
The stations exposed/supported by the Ensemble is the intersection of the supported stations of all the IM-Ensembles that also exist in the ensemble station file.
And the stations supported by a specific IM-Ensemble is the intersection of the stations supported by its branches.
I.e. (Station file ∩ (IM-Ensemble-1-Stations ∩ IM-Ensemble-2-Stations ∩ ... ∩ IM-Ensemble-N-Stations) where IM-Ensemble-n-Stations = (Branch-1-Stations ∩ Branch-2-Stations ∩ .... ∩ Branch-n-Stations)
The Branch stations are the union of the supported stations of each IMDB.
I.e. (Branch-1-Stations = Leaf-1-Stations ∪ Leaf-2-Stations ∪ .... ∪ Leaf-n-Stations)