MetroMan Making Bias Worse

[mikedurand]

For the first time, with this latest ML priors run, I can identify that MetroMan seems to be making the bias worse. Longer-term this issue needs more diagnosis, but in the short term, can address this with just a change to the hyper parameters on discharge uncertainty, making them more in line with the new MLpriors accuracy.

[mikedurand]

Sometimes this seems to come from ProcessPrior. The WSE data are noisy, and lead to large temporal variations in cross-sectional area, leading to an unavoidably high minimum area, a large mean area. This results in the mean of the prior estimated Q timeseries being much larger than the target value from the prior. E.g. reach 73140700361 for branch 16-discharge-timeseries-are-super-noisy.