← Back to Articles

Why NDVI has always been a guess - and what to do instead

For 30 years, NDVI has been agriculture's go-to satellite metric. It measures greenness. But nitrogen is not greenness. Here's what that distinction costs farmers - and what hyperspectral imaging changes.

What NDVI actually measures

The Normalised Difference Vegetation Index compares how much red light a crop absorbs against how much near-infrared light it reflects. That ratio correlates with how much green biomass there is - a thick canopy is different from bare soil. It is useful. It is not nothing. But it tells you the crop is green. It does not tell you whether that crop is well-fed or starving.

Here is the problem: a wheat crop at 2.5% nitrogen in the leaves and one at 3.5% nitrogen can look identical in green colour - both will produce a high NDVI reading, in the same field, on the same week. The NDVI image cannot distinguish them. But they are not the same crop. One will go deficient in 10 days. The other has a surplus the plant cannot absorb. Acting on NDVI alone means treating both the same.

The wavelengths that reveal nitrogen

Nitrogen in the crop - in chlorophyll, in proteins and enzymes - absorbs and reflects light in very specific ways across dozens of wavelengths beyond what NDVI uses. The 430–450nm band (violet), 660–680nm (red-edge), 705–730nm (red-edge shoulder), and several shortwave infrared bands all carry nitrogen-specific signals. None of them are captured by NDVI's two-band approach.

Hyperspectral sensors capture 160 or more contiguous bands across the full visible-to-shortwave-infrared range. Instead of two data points, you have a full spectral fingerprint. Machine learning models trained on thousands of ground-truth lab samples can then read that fingerprint and return a precise nitrogen concentration - the same number a laboratory would give you, without the test tube.

Why the industry never moved

Hyperspectral satellites existed on paper for decades, but the commercial satellite constellation - high-resolution, frequent-revisit, affordable - did not exist until very recently. The sensors, data pipelines and compute cost to process 160-band imagery at 5m resolution across millions of hectares were simply prohibitive before 2022.

The training data problem was equally serious. A hyperspectral model is only as good as its ground truth. Without tens of thousands of paired samples - physical tissue tests taken at the exact same time as a satellite pass, in real fields, across multiple seasons and geographies - you cannot train a model that holds up in commercial conditions.

Both barriers are now gone. New commercial hyperspectral satellites pass over UK and European fields at 5m resolution with frequent revisits. And Messium has spent three years physically collecting over 30,000 paired samples - making it the first company to have a training dataset large enough to use these satellites reliably for commercial nitrogen measurement.

What changes for the farmer

The practical shift is from a coloured map to a number you can act on. NDVI gives you a relative greenness score - useful for comparing one corner of a field to another, not for knowing whether that corner needs 40 kg N or 80. A hyperspectral nitrogen reading gives you the concentration in the leaf tissue, in percentage units, comparable to a lab test. That number flows directly into a recommendation: how much, when, and where on the field.