False negatives in a flood extent map and what they mean for evacuation calls

A false negative on a flood extent map is water that's actually there but the map shows as dry. Not a bad pixel, not noise. Standing water the sensor looked right at and the delineation algorithm called land anyway. For a duty officer deciding which streets to pull people out of, that's the error that bites.
The opposite error, a false positive, calls dry ground wet. Annoying, but it fails safe: a crew drives out, finds the road passable, and moves on. A false negative fails the other way. It tells you a block is clear when it isn't, and nobody goes to check it because the map said no.
Where the water hides from the sensor
False negatives in satellite-derived flood maps trace back to three things, and they compound.
Shadow. A cloud edge, a levee, a line of buildings throwing shade across a channel at the moment of capture. Water under shadow reads dark in the optical bands the same way dry shaded pavement does, and at medium resolution (roughly 3 to 10 meter pixels) there isn't enough spectral separation to tell them apart reliably.
Canopy and vegetation. Flooded cropland with standing stalks, a flooded floodplain forest, backyard flooding under tree cover: the sensor is seeing leaves, not the water sitting underneath them. This is probably the single biggest source of missed flooding in automated extent maps, because vegetated floodplains are exactly where slow-rising river floods tend to sit longest.
Mixed pixels at the edge. A 10-meter pixel straddling a flooded ditch and a dry shoulder gets averaged. If the wet fraction is small, the pixel classifies dry. This mostly trims the edge of the extent rather than hiding a whole neighborhood, but edges are also where people decide whether their street counts.
None of this is a flaw unique to one product or one vendor. It's a property of passive optical remote sensing at medium resolution, event-driven cadence included. A cloud-free look only tells you what the sensor could see through, not what sat under a tree canopy or a shadow at that exact pass.
Reading the map for a call you have to make now
Three things a duty officer can do with that in mind:
Ground-truth the margins first. The bulk of a flood extent map from a cloud-free pass is solid. The vegetated floodplain fringe, the shadowed channel bends, and the tree-lined residential blocks are what need a second look, often from a spotter report, a gauge reading, or a quick pass by a crew already in the area.
Cross-check against known flood-prone vegetation and tree cover on your own base map. If a mapped extent stops cleanly at the edge of a wooded area that you know floods every time the river comes up, treat that boundary as suspect rather than final, and flag it in your brief rather than erasing it.
Timestamp the pass and say so out loud in the brief. A map built from imagery captured six hours before the current read is not wrong, it's just old. Water rises, water recedes, and a number on a map is a snapshot. Coordination needs to know which snapshot they're looking at before they stake an evacuation boundary on it.
Satellite-derived extent maps still beat hours of manual delineation from whatever imagery happened to come in. A tired analyst working one event alone makes the same kind of judgment calls under time pressure, often with less consistency between passes. Coordination gets more out of the extent map when they treat it as a strong first cut and mark the margins with ground truth as spotter reports and gauge readings come in through the shift.
That's the frame Rapid Flood Mapping builds its situation brief around: a numbered extent map as soon as a cloud-free look comes in, built to hand straight to response coordination rather than sit in an analyst's queue.
If your team is standing up a response right now, see how fast that brief turns around.