So it is just a guesstimate based on what history has been recorded.
Saying they can model it pretty good is sort of giving themselves a pat on the back, thinking they are correct? Chances are good there was that 4 inch rain in 4 hours sometime in the past 10,000 years, but what they don't know is if it was right before we occupied a particular location and someone recorded it in some way (like 200-500 years ago) or if it was actually 9783 years ago.
rainfall records go back aways
plus they have live data now
so they record rainfall and river rise
for all levels of rainfall for all seasons
snowmelt and temp also factor in
with these data points they can build models
and tune based on actualities
they are pretty accurate predicting river rise after a storm
they can a peak and timeframe
they know the river level
they know the projected rise
and they know the ground contours and river bank elevations
it's the best they can do and it is continuously improving
risk accuracy is good
probability not so good