Try this analysis focusing on discharge

[ehmatthes]

The river height is visible, which is why I focused on it initially. But someone at USGS suggested trying the same analysis focusing on discharge instead of height. Height has an upper bound, so if the river is less than 2.5 feet from the max height, it can never go critical in this model.

If we can find critical values for the discharge rate and it gives a comparable true positive rate, this might be a more reliable model.