mescla.uncertainty.identifiability¶
Can these mixing ratios be computed at all?
Three failure modes recur, and all three produce confident-looking numbers rather than errors:
Uncertain end-member definition – undocumented variability, conceptual error.
Poor tracer choice – no contrast between sources.
Colinearity – end-members too close together in tracer space, making the solution unstable or unidentifiable.
Run identifiability_report() before believing any result.
Functions
Condition number of the end-member matrix. |
|
Pairwise distance between end-members, normalised by their overall spread. |
|
Run every identifiability check and return a pass / warn / fail table. |
|
Re-estimate with the end-member sigmas scaled up and down. |