NDVI — Normalized Difference Vegetation Index

Green leaves do two things to light that almost nothing else does: they absorb red strongly, because chlorophyll uses it for photosynthesis, and they scatter near-infrared strongly, because the spongy mesophyll inside the leaf reflects it rather than absorbing it. Between about 700 and 750 nm the reflectance of a healthy leaf jumps from a few per cent to forty or fifty — the "red edge", the sharpest spectral feature in the natural world. NDVI is the simplest possible way of measuring how big that jump is: $(\rho_{NIR} - \rho_{red})/(\rho_{NIR} + \rho_{red})$.

Rouse, Haas, Schell and Deering defined it in NASA SP-351 (1974), page 309, working with ERTS-1 imagery of the Great Plains, and its form is doing something clever. Dividing the difference by the sum normalises out overall brightness, so a canopy in full sun and the same canopy under a passing cloud return roughly the same value. It also bounds the result between −1 and +1 regardless of the sensor's gain. That is why NDVI survived while dozens of other indices did not, and why it has an unbroken satellite record running back half a century.

The rough reading: negative values are water, snow or cloud, where red exceeds near-infrared; 0 to 0.2 is bare soil, rock, pavement or dead residue; 0.2 to 0.5 is sparse to moderate canopy; 0.5 to 0.75 is a healthy developed canopy; above 0.75 is dense vegetation. Those bands are useful and they are not measurements — they shift with sensor, with sun angle, and with the specific band passes used.

The first honest limitation is saturation. Above a leaf area index of about 3 — roughly three square metres of leaf per square metre of ground, which a healthy crop reaches well before peak season — the near-infrared has already been scattered about as much as the canopy can scatter it, and adding more leaves changes the index barely at all. A field at LAI 3 and the same field at LAI 6 can read within a few hundredths of each other. NDVI therefore loses its ability to discriminate exactly when a grower most wants it, which is why it is a poor mid-season yield predictor and why indices using the red edge or the green band, which saturate later, have displaced it for that job.

The second is soil. When the canopy is thin, most of what the sensor sees is ground, and a wet dark soil and a dry bright soil under identical plants give noticeably different NDVI. An early-season NDVI map is frequently a soil moisture map with a crop signal on top. That specific failure is what SAVI was built to fix, and the two indices belong side by side.

The third matters most for drone work, and it is the one that puts wrong numbers into agronomy reports. NDVI FROM AN UNCALIBRATED CONSUMER CAMERA'S RAW DIGITAL NUMBERS IS NOT COMPARABLE BETWEEN TWO FLIGHTS OF THE SAME FIELD. The index is defined on REFLECTANCE — the fraction of incident light returned by the surface — and a raw digital number is not reflectance. It carries the solar elevation at the moment of exposure, the atmospheric state, the exposure and gain the camera chose, the white balance, the gamma curve baked into the JPEG, and the spectral response of whatever filter happened to be screwed on the front. Fly the same unchanged field at 10 a.m. and again at 2 p.m. and the DN-based values will differ by more than any real treatment effect. Fly it in June and again in July and the sun has moved twenty degrees. Reports that put those two numbers on the same axis and describe the difference as crop response are reporting the weather and the camera's autoexposure.

What makes flights comparable is specific and it is not optional: a multispectral sensor with per-band radiometric calibration and discrete bandpass filters, a calibrated reflectance panel imaged on the ground immediately before and after each flight, and a downwelling light sensor logging incident irradiance throughout so that changing cloud can be corrected. With all three you have reflectance, and a time series means something. Without them, treat the map as a RELATIVE picture of one flight — genuinely useful, because "which corner of this field is worst today" is a real and answerable question — and never as a number to trend. The distinction is not pedantry; it is the difference between an agronomic decision and an artefact of the light.

One further caution for converted cameras specifically. A "NDVI camera" made by removing the infrared-blocking filter from a consumer camera and adding a coloured one produces bands that are broad, overlapping and nothing like the narrow near-infrared and red bands the index was defined on. The number it computes is a vegetation index of a sort; it is not NDVI, and its values do not correspond to published NDVI thresholds. Say so when you deliver it.