True colour
One place, and five different questions
A river running west to east, a town on the south bank, woodland to the north west with a burn through part of it, and farmland either side worked in a mosaic of plots. There is also a discharge into the river, though you would struggle to point at it in this view.
Everything below is this same picture. Nothing is re-photographed and nothing is re-rendered. All that changes from one section to the next is which two of the five bands go into the same piece of arithmetic.
NDVI · near infrared and red
Ask where the living plants are
The index from the last piece, run over this landscape. Standing crop and woodland come out high, the turned plots drop back, and the town and the river fall away to nothing.
Notice the burn. It does not read as bare ground and it does not read as forest either. It sits between them, which is a real signal and not enough on its own to tell you what happened there.
NDWI · green and near infrared
Swap the bands and it finds the water instead
Same subtraction, same division, two different bands. Water is bright in green relative to the near infrared, and everything with a leaf on it is the other way round, so the river separates from the land in a single step.
This is why an index is better thought of as a question than as a formula. Nothing about the arithmetic changed. We asked a different thing.
NDTI · red and green
Ask what is in the water and the discharge appears
Clear water reflects more green than red. Water carrying suspended sediment reflects more red than green, because the particles scatter the longer wavelength back out. Take the difference and the plume separates from the river it is sitting in.
The outfall is now obvious, and so is how far downstream the plume travels before it disperses. This is the arithmetic behind effluent and discharge verification, and it is the same three lines of code as the two indices above it.
NBR · near infrared and shortwave infrared
Ask what has burned and the scar separates from the soil
Char does something distinctive: it darkens the near infrared and brightens the shortwave. That is the reverse of living canopy, so the difference between those two bands puts fire scars at one end of the scale and healthy vegetation at the other.
The burn that was ambiguous under NDVI is unambiguous here. Bare turned earth, which looked similar before, stays near the middle. Two indices disagreeing is not a problem; it is how you tell a cleared field from a burnt one.
Ignore the river. Water reflects almost nothing in either of these two bands, so the fraction is a small number divided by another small number and the result swings to whatever the noise says. Every index has ground it means nothing on, and in practice water is masked out before a burn index is read at all.
NDBI · shortwave infrared and near infrared
Ask where it is built, and get a less clean answer
NDBI is meant to pick out built-up ground, and it does separate the town decisively from vegetation and from water. But look at the bare plots and the burn scar. They score as high as the town or higher, because what this index actually measures is dry surfaces with no leaf water in them, and a ploughed field qualifies.
This is worth being straight about. No single index is a classifier. NDBI narrows the question to a set of candidates and something else has to finish the job, which is usually NDVI to exclude the fields and a look at how the answer changes across seasons. Anyone who shows you a built-up map from one index and one date has skipped that step.
