Changing state during secondary sampling period

[sixtystrat]

We are doing a RD Multi-state analysis of black bears to evaluate a major flooding event on surival and transition probabilities. We are comparing transition rates from a flooded area to a non-flooded area before, during, and after the flood. The problem we are having is that a proportion (9 of 109) of the bears change states (flooded to non-flooded and vice versa) during the secondary sampling period (a violation of assumptions). This is not really an unobservable or misidentified state because we know what the states are, it is just that they change during a period when they are not supposed to. How do we best handle this in our analysis?