What does “dB” mean?
Radar images are measured in decibels (dB). It is a scale that works by multiplying instead of adding, so small numbers hide big changes.
Every +3 dB roughly doubles the signal. Every −3 dB roughly halves it.
How radar sees water
NISAR sends a radar signal to the ground and measures how much comes back. Pick a surface to see what happens to the signal.
Signal that comes back
Typical example values, for illustration only. They are not measurements from this study.
Calm water bounces the signal away, so it looks dark. Rough ground, rubble and buildings send more back, so they look bright. AAB looks for places that turned dark between two dates.
About half the signal
The bars use the same log-style scale as dB, so equal steps look equal.
Five steps
Choose a step to read what happens, in simple words and in detail.
Pick a matching pair.
Both scenes come from the same satellite track, frame and viewing direction (Track 84, Frame 19, ascending). Different viewing angles make steep hills cast different shadows, which can look like change, so mixed pairs are rejected.
IN SIMPLE WORDSTake both photos from the same spot. If you take one photo from the left and one from the right, the shadows move, and it looks like the land changed when it did not.
Scene A · same track, frame, direction=Scene B · same track, frame, direction→Valid pair ✓
Prepare the images.
We use terrain-corrected radar data (NISAR GCOV, HH polarisation), mask out invalid pixels, and convert to decibels (dB). On this scale, a 3 dB change means the signal roughly halved or doubled.
IN SIMPLE WORDSThe raw radar data is tidied up first. It is placed correctly on the map, pixels with no usable data are ignored, and the numbers are put on the dB scale so they are easy to compare.
Terrain-corrected data→Invalid pixels masked→Converted to dB
Compare the two dates.
For every 10 m pixel, we subtract the “before” value from the “after” value.
IN SIMPLE WORDSFor each small square on the ground, we ask one question: is it darker or brighter now than before? A negative answer means darker. A positive answer means brighter.
after − before = changee.g.−18 − (−9) = −9 dBGot darker
The numbers above are an example. You can try your own further down, in “Try the rule”.
Apply the rule and clean up.
A pixel is a candidate when it got at least 3 dB darker and is now at or below −14 dB. Permanent water is excluded. Groups smaller than 4 pixels (400 m²) are removed, which filters out 3,323 isolated speckle pixels.
IN SIMPLE WORDSA square is only flagged if it got clearly darker and is now dark enough to look like calm water. Lakes and rivers that were already there are ignored. Tiny lonely dots are deleted because they are usually just radar noise.
Got ≥ 3 dB darker+Now ≤ −14 dB+Not permanent water+Group of 4+ pixels
Set aside uncertain areas.
Pixels that got at least 3.5 dB brighter are held as uncertain surface change: 11.56 km². They are not counted as flood.
IN SIMPLE WORDSPlaces that got much brighter are not water. They may be rough, disturbed ground. We do not know exactly what happened there, so we keep them separate instead of guessing.
Got ≥ 3.5 dB brighter→Uncertain surface change→Not counted as flood
Step 1 of 5
The three thresholds
A threshold is just a line. Cross it and the rule says “yes”. The needles below move when you use the sliders in “Try the rule”.
Decrease
−3 dB or lower
Right now: –
Got at least 3 dB darker between the two dates, roughly half the signal.
In simple words
This asks: did this spot get clearly darker? A tiny dip could be noise, so it must drop by at least 3 dB, about half the signal, to count.
Post-water
−14 dB or lower
Right now: –
After the event, dark enough to look like calm water.
In simple words
This asks: is it dark enough now to look like calm water? A spot can get darker and still be too bright to be water, so both rules must pass together.
Anomaly
+3.5 dB or higher
Right now: –
Got at least 3.5 dB brighter. Treated as rough surface change, not water.
In simple words
This asks: did the spot get much brighter instead? Water does not make radar images brighter, so these spots are set aside as “uncertain” and are not counted as flood.
Test the rule on one pixel
Set how bright one 10 m square was before and after the flood. The page applies the same rule AAB uses and tells you what happens.
Your pixel
Change = after − before = −10 dB
What the rule says
- –
- –
- –
- –
A flagged pixel still has to survive the cleanup step below: it must belong to a group of at least 4 touching pixels.
Why tiny dots are removed
Radar images are grainy. Lonely flagged pixels are usually noise, so small groups are deleted. Drag the slider to change the minimum group size.
Practice map (made-up pixels)
KeptRemovedNot flagged
4 pixels × 100 m² = 400 m²
0groups kept
0groups removed
0pixels kept
0pixels removed
A “group” is a set of flagged pixels that touch each other. The slider starts at the value AAB really uses.
Real result: 55 candidate patches, 278 pixels, 0.0278 km². 3,323 isolated speckle pixels were filtered out.
Pair validity
Why the two scenes must be taken from the same viewpoint. Switch between a valid pair and a mixed pair.
The checks
Same track and frame (Track 84, Frame 19)
PASS
Same viewing direction (Ascending)
PASS
Orbital compatibility check
PASS
A pair from different directions would fail this check and be rejected.
Processing record
The exact settings behind the result, for anyone who wants to check the work.
Cleaning
Groups under 4 pixels (400 m²) removed; 3,323 speckle pixels filtered.
Why no machine learning
Supervised machine learning was bypassed because there is no independent ground truth to train or test it. The result comes from a transparent rule anyone can check.
Data status
NISAR GCOV, HH, 10 m pixels, provisional product maturity, v1.0 (16-stage NISAR GCOV pipeline).
Technical details
Show technical details
Schema validation: PASSED. Rejected complex term: HHHV. Pixel area: 100 m².
Words in plain English
Tap any term for a short, simple explanation.
In short
AAB takes two radar images from the same viewpoint, one before and one after the flood. Where the ground turned dark enough to look like calm water, it marks a possible flood. Where it turned much brighter, it marks uncertain change instead. Nothing here is confirmed without a field check.