Radar and SAR

On a clear day none of this is difficult

A floodplain: a river held between embankments, a village on the south bank, two patches of woodland, and low hills along the northern edge. Everything in the series so far would work on this scene without complaint.

It is worth noticing that a clear day is doing a lot of the work here.

Cloud arrives and the picture is worth nothing

Optical sensors measure reflected sunlight, and cloud sits between the sun and the ground. There is no processing step that recovers what is underneath, because the photons never reached it.

This is not a rare inconvenience. Across the wet tropics, and across South Asia in the monsoon, useful optical passes can be scarce for months at a time. Worse, the periods when cloud is heaviest are exactly the periods when the questions are most urgent: a flood arrives with the weather that hides it.

Radar brings its own light and looks sideways

This is the same ground on the same day, imaged by radar rather than by sunlight. The sensor emits its own pulse at a wavelength centimetres long rather than fractions of a micron, and cloud at that scale is effectively transparent. It also works at night, for the same reason: it never needed the sun.

The picture looks strange because it is not a picture of colour. It is a map of how much of the pulse came back, and that is decided by shape and roughness rather than by pigment.

Calm water is a mirror, not a sponge

The usual shorthand is that water absorbs radar. It does not. Water is a strong reflector, but a calm surface is smooth compared with the wavelength, so it reflects the pulse away in the mirror direction. The receiver sits beside the transmitter, so nothing comes back. The energy left; it was never consumed.

A wall meeting the ground does the opposite. It forms a corner that returns the pulse back along the path it arrived on, which is why the village is the brightest thing in the frame. Vegetation does a third thing: the pulse enters the canopy, bounces around between leaves and branches, and some of it finds its way out again. That is volume scattering, and it is why woodland reads mid grey while grass, which is a thin layer rather than a volume, reads darker.

The figures beside the picture are the mean return for each of those surfaces, computed from the model rather than quoted from a textbook.

Under water, the village gets brighter rather than vanishing

Now the river is over its banks. The flooded ground has become smooth, so it mirrors the pulse away and reads almost black, and the extent of the water falls out of the image without any interpretation at all.

The counterintuitive part is the village. Standing water against a vertical wall makes a better corner reflector than dry ground does, so buildings in the flood light up rather than disappearing. That is how you find inundated settlement in a radar image: not by looking for something missing, but by looking for something that has become brighter than it was.

Put wind on the water and the trick stops working

Everything above rests on the flood surface being smooth. Roughen it and it stops behaving like a mirror and starts scattering the pulse back like any other rough surface. Watch the flood water climb through an order of magnitude, until it is returning as much as the bare ground beside it. At that point an outline drawn on brightness alone has nothing left to separate.

So flood extent from radar is reliable on a calm morning and progressively less so as the wind gets up, which is worth saying before a client finds it out. In practice it is handled by checking wind at the time of the pass, by preferring certain wavelengths and polarisations, and by quoting extent with a confidence rather than as a single outline.