The min_error calculated by residual_error_method includes time-invariant contributions from the obs error, representation error and model error. We also account for these terms in a time-variant way through other calculations (although addaveragingerror needs modifying - see Issue #42 ). Currently the min_error is added to the time-variant errors to yield the model-obs mismatch error, which means that errors are being double counted.
It would be better if this min_error was used as a floor below which the model-obs mismatch error (epsilon) never drops, and when the time-variant errors are larger than min_error we use those values (i.e. ignore min_error in those cases). That follow the approach taken in InTEM.
The min_error calculated by residual_error_method includes time-invariant contributions from the obs error, representation error and model error. We also account for these terms in a time-variant way through other calculations (although addaveragingerror needs modifying - see Issue #42 ). Currently the min_error is added to the time-variant errors to yield the model-obs mismatch error, which means that errors are being double counted.
It would be better if this min_error was used as a floor below which the model-obs mismatch error (epsilon) never drops, and when the time-variant errors are larger than min_error we use those values (i.e. ignore min_error in those cases). That follow the approach taken in InTEM.