NDVI Explained: How Satellites See Your Crops

Every few days a satellite passes about 786 km above your farm and measures how much light bounces off your fields. Most of the vegetation maps in UK farm software start from that measurement: the green-to-red field maps in xarvio FIELD MANAGER, Data Farming, Hutchinsons Omnia and Climate FieldView, and the “biomass” layer in John Deere Operations Center. Those maps are useful. They also make it easy to send a sprayer or spreader after a patch of cloud shadow. This piece covers what NDVI satellite imagery measures, why UK weather limits it more than the sales demos suggest, and how to check whether a red patch is a real crop problem before you spend money on it.

What the number actually is

A healthy green leaf absorbs most red light, because chlorophyll uses it for photosynthesis. It reflects a lot of near-infrared (NIR) light, which you can’t see, because the cell structure inside the leaf scatters it back out. Bare soil reflects both at roughly similar levels. NDVI (Normalised Difference Vegetation Index) puts those two facts into one sum:

NDVI = (NIR − Red) / (NIR + Red)

Here are some example figures, using reflectance values typical of what Sentinel-2 records:

NDVI is therefore a measure of how much green leaf the satellite can see per square metre. It does not measure nitrogen, yield, disease or plant health directly. Plenty of crop problems shrink green leaf area, so they show up in the number. But a thick patch of blackgrass also shows up as healthy green leaf, and a clean, well-fed crop that was drilled late looks “poor” in March.

Most of the free and low-cost tools use Sentinel-2, the European Space Agency satellites under the Copernicus programme. Their red and NIR bands have 10 m pixels, so each pixel covers 100 m² and a hectare is 100 pixels. That’s fine for zoning a 15 ha field. It’s too coarse to see tramlines, a single blocked coulter or a 2 m strip by a hedge.

Saturation: why your best fields go flat

Once a cereal canopy reaches a green area index of roughly 3 (usually somewhere between GS31 and GS37 in a decent wheat crop), NDVI stops rising much. The leaves at the top already reflect nearly all the NIR they’re going to, and the red has nowhere left to drop. Readings bunch up between 0.82 and 0.90 across the whole field.

So the “uniform green” field you see in May may not be uniform at all. The index has simply run out of range. If you’re making a final nitrogen or PGR decision in late April or May, ask your software for a red-edge index instead. Data Farming, Omnia and xarvio all offer NDRE or a similar red-edge layer. Red-edge keeps separating dense canopies for longer. NDVI is at its most useful from establishment through to early stem extension, and again at senescence, when differences in ripening show up clearly.

The UK cloud problem, in numbers

Sentinel-2 passes over a given UK field every two to three days, because the satellite tracks overlap at our latitude. On paper that gives you ten or more images a month. In practice it doesn’t.

The satellite’s sensor is optical, so it can’t see through cloud. Across England and Wales, a typical March or April gives you somewhere between one and four usable passes per field, and a wet February in the west can give you none. Lincolnshire usually does better than Devon, and Herefordshire dairy ground can go weeks without a clean view.

The bigger risk is the partly cloudy image, because it gets through. The processed Sentinel-2 product (Level-2A) carries a scene classification layer that flags each pixel. If you open a tile in the free Copernicus Browser, the classes look like this:

SCL value  Class
3          Cloud shadows
4          Vegetation
5          Not vegetated
6          Water
7          Unclassified
8          Cloud medium probability
9          Cloud high probability
10         Thin cirrus

Commercial platforms use these flags, or their own versions of them, to throw out bad pixels. The masks are good, not perfect, and three things get through regularly:

• Thin haze and cirrus that the mask rates as clear. Haze scatters extra light into the red band, so NDVI across the whole field drops by 0.05 to 0.15. Because the drop is often uneven, you can end up with a “gradient” running across a field that was evenly drilled and evenly fed.
• Cloud shadow just outside the masked cloud. Shadow reduces NIR more than red, so the shadowed area reads low. It is offset from the cloud by a distance that depends on sun angle and cloud height. On a February afternoon, with the sun low, the offset can be several hundred metres, so the shadow lands in the next field.
• Mixed-edge pixels. A 10 m pixel that is half wheat and half hedge, track or water trough reads as something in between. Headlands and field edges nearly always look worse than they really are.

Stretched colour scales make small differences look big

Before you look at a single map, understand how it was coloured. Most farm software stretches the colour ramp to each field’s own range. That way you always see differences, even when they’re tiny.

In a field where NDVI runs from 0.66 to 0.71, a stretched map puts deep red at 0.66 and deep green at 0.71. That spread of 0.05 is within the noise of a hazy image, but it looks like a field split into good and bad halves. Put the same field on a fixed 0 to 1 scale and it’s one even shade of green.

Every tool mentioned here lets you see the actual values, either by hovering over the map or by exporting zone statistics. Check the real numbers before you believe the colours.

Worked example: the red corner in Long Field

Here’s a typical case. It’s a 22 ha field of winter wheat on clay loam, drilled 4 October. On 12 March the farm manager opens Data Farming and sees a red block of about 3 ha in the north-east corner. The plan was a uniform 70 kg N/ha first split. The question is whether to push an extra 40 kg N/ha onto that corner.

Before ordering anything, the manager goes through the following checks.

1. Read the values, not the colours. The zone statistics show:

Zone        Area (ha)   Mean NDVI   Pixels
NE corner   3.1         0.58        310
Rest        18.9        0.67        1890
Field mean  22.0        0.66        2200

A gap of 0.09 is big enough to matter at this growth stage, so the check goes on.

2. Look at the image date and cloud cover. The 12 March image is flagged as 18% cloud across the tile. The tile is 110 km square, so that figure says almost nothing about this one field. The manager opens the same date in Copernicus Browser using the true-colour view. A bank of cumulus sits about 400 m south-west of the field. With a low March sun in the south, the shadow of that cloud falls to the north-east. That’s a suspect.

3. Compare the date before. The previous clear image, from 24 February, shows the north-east corner at 0.49 and the rest at 0.51. That’s a gap of 0.02, which means nothing. A real establishment or waterlogging problem doesn’t appear from nowhere in 16 days of cold weather.

4. Check the date after, if there is one. On 17 March there’s a clear pass, and the corner reads 0.65 against 0.68 for the rest of the field.

5. Walk it anyway. Plant counts in the corner come out at 240 plants/m², compared with 260 near the gate. Tiller numbers are similar.

The verdict is a cloud shadow artefact plus a slightly lighter patch of soil. No action is needed.

Here’s what acting on the map would have cost. An extra 40 kg N/ha over 3.1 ha is 124 kg N, which is about 360 kg of ammonium nitrate. At roughly £330/t that’s under £120 in fertiliser. That’s a small sum, which is why farmers tend to just do it. The real cost is what comes next. The same shadow-affected image gets used to build a variable-rate PGR or fungicide plan, or to start an argument with your agronomist, and next season’s zones are drawn from a date that was never valid. One bad image fed into a zone map can carry its error into three years of decisions.

When the red patch is real

Some patterns hold up across dates, and those are worth acting on:

• It persists across two or more clear images at least a week apart. Artefacts tied to one image don’t.
• Its shape follows something on the ground: an old field boundary, a drain run, a headland, a soil change visible on the 1940s aerial photos, a compaction line from a wet harvest.
• It matches your yield map. If the last two harvests show that zone 1 to 1.5 t/ha down, the satellite is confirming what you already know.
• The anomaly is the right size for the problem. Slug grazing, waterlogging and patchy establishment show up as low NDVI in autumn and winter. Blackgrass shows up as unexpectedly high NDVI in spring, which surprises people.

For livestock farms the same rules apply to grass. Low NDVI paddocks on a rotational grazing plan may simply be the ones grazed last week, so check your grazing records before you blame poor soil.

Picking a tool without overpaying

You don’t need to pay for anything to start looking. Copernicus Browser is free and gives you raw Sentinel-2 images, a true-colour view and NDVI. Use it to cross-check any map your farm software shows you. Data Farming offers free Sentinel-2 NDVI for UK fields and charges for variable-rate scripts. xarvio FIELD MANAGER has a free tier that shows biomass maps. Omnia normally comes as part of a Hutchinsons agronomy relationship rather than as a standalone purchase.

Paying for higher resolution only makes sense in particular cases. Planet’s 3 m daily imagery, which some platforms resell, roughly triples your chance of getting a clear view in a cloudy fortnight, and that matters most in the west and north. A drone survey at 5 cm per pixel gets under the cloud entirely and can show tramlines and individual weed patches, but someone has to fly it. Where satellite, drone and variable-rate application fit together, and what each layer costs per hectare, is set out in our satellite, drone and variable-rate agronomy guide.

The cheapest habit is free. Before any satellite map changes a plan, open the image date in true colour and look for cloud or shadow, then check the dates before and after. It takes five minutes, and you’ll catch most of the artefacts that would otherwise get sent to the spreader.