Understanding hyperspectral imaging: a farmer's guide
No jargon. A plain explanation of what hyperspectral satellites actually do, why they're more powerful than anything before, and what it means for your farm this season.
Start here: what light tells us
Every material - soil, rock, water, leaf tissue - absorbs and reflects light differently across different wavelengths. Your eyes detect three broad ranges: red, green, and blue. That is enough to see colour. It is not enough to see chemistry.
Different molecules absorb light at very specific wavelengths. Chlorophyll absorbs heavily around 430nm and 670nm. Water absorbs strongly around 970nm and 1190nm. Nitrogen-containing proteins absorb at distinct points in the near-infrared. These are like fingerprints - if you can read the full spectrum, you can read what is in the plant.
What makes an image "hyperspectral"
A standard colour camera captures 3 channels - red, green, blue. A multispectral camera (what most agricultural drones and satellites use) might capture 4 to 12 bands, including some near-infrared. This is where NDVI comes from.
A hyperspectral camera captures 100–400 contiguous narrow bands, often covering wavelengths from 400nm out to 1000nm or beyond. Instead of a colour photograph, you get a data cube - every pixel in the image has a full spectrum attached to it. For a 5-metre resolution hyperspectral satellite image of a 100-hectare field, that is hundreds of millions of individual spectral readings, collected in a single pass.
How nitrogen shows up in the spectrum
Nitrogen in wheat is held mainly in chlorophyll, structural proteins, and enzymes. Each of these has characteristic absorption features in the hyperspectral range. Critically, as nitrogen content in the leaf falls - as the crop starts to become deficient - the entire spectral profile shifts in detectable ways that are invisible in the standard red and near-infrared channels NDVI uses.
This is not a new discovery - plant scientists have known about nitrogen spectral features for decades. What is new is the ability to read those features from a commercial satellite at useful resolution and revisit frequency, and to have a training dataset large enough to turn those spectral signals into reliable, field-specific nitrogen predictions.
From spectrum to recommendation
Messium's process in brief: a hyperspectral satellite passes over your field, capturing 160 bands of spectral data at 5m resolution. The image is processed, calibrated, and fed into a machine learning model trained on 30,000 paired ground-truth lab samples. The model returns an estimated nitrogen concentration for every 5m pixel. That data feeds into a crop and soil model that incorporates weather, soil type, management history, and your yield and protein target. The output is a specific nitrogen recommendation - kg/ha, by zone, with a deficiency date and a confidence interval.
The farmer receives a report and an application map. The map can be loaded directly into variable-rate application equipment - no manual translation, no manual zoning. The recommendation fires at the right time because the data is refreshed every week, not every season.
What this means in practice
For a 500-hectare arable farm running winter wheat, Messium delivers something that was previously impossible: a weekly snapshot of actual nitrogen status across every field, at 5m resolution, without any field walking or tissue testing. The cost of that information - the actual satellite pass, data processing, modelling, and recommendation - is a small fraction of the fertiliser saving it enables.
The result is less guesswork in both directions: no applying fertiliser the crop cannot use, no waiting until deficiency is visible before acting. Deficiency is now detectable from space before it is visible on the ground.