AI Soil Maps: From Grid Sampling to Management Zones

Most fields in the UK have been sampled the same way for thirty years: a W across the field, a bag of cores, one lime recommendation and one P&K figure for the whole block. It is cheap and it is wrong in a predictable way. A 20 ha field with a pH range of 5.9 to 7.2 gets told it averages 6.5, and the acid corner never gets what it needs while the alkaline end gets lime it didn’t want.

This post is about the three ways of getting past that, what each costs per hectare, how far you can trust each one, and how to turn the result into a variable-rate (VR) lime and P&K plan that your spreader can actually run. The short version: for most arable and grassland farms, the best value is a hybrid. Scan or model first, then spend your lab money on a small number of targeted samples. Pure grid sampling is the gold standard that few can afford to repeat, and pure AI prediction is a cheap guess until you’ve checked it.

The three routes, with real costs

Prices below are typical UK 2025/26 ranges from agronomy firms and contractors. Get quotes, because they vary by region and field size.

The important point is in that last column. Nothing here is a free lunch. EM scans tell you where the soil changes. Grid sampling tells you what is in the soil at specific points. AI maps tell you where it probably changes, using cheaper data. Use each for what it is good at.

Grid sampling: reliable, but a snapshot

Grid sampling is still the reference. If your agronomist or a firm like Hutchinsons, Agrii or Soil Association-linked labs such as NRM or Eurofins gives you a 2 ha grid, you get a measured value for each cell. Under RB209 (the Nutrient Management Guide), those values translate straight into P and K indices and lime requirements.

Where it falls down:

• Interpolation. A 2 ha grid on a 20 ha field gives you about 10 points. Between them the software guesses, usually by kriging or inverse distance weighting. If your soil changes over 50 metres (old field boundaries, a gravel ridge, a filled pond) a 140 m grid spacing will miss it.
• Cost of repeating. Lime needs rechecking every 4 to 5 years. P&K every 4 years or so. At £18/ha, a 200 ha farm spends £3,600 each cycle.
• Sampling error. Your results depend on the person with the auger. Consistent depth (15 cm arable, 7.5 cm grass) matters more than most people admit.

A fair rule: grid sampling pays on fields over 10 ha with obvious variability, or where you’re spending heavily on lime and fertiliser.

EM scanning: finding the boundaries

An EM scanner (a Veris 3100 pulled behind a quad or tractor, or a Geonics EM38-MK2 sled) records apparent electrical conductivity (ECa) every few metres. Clay and wet soil conduct well. Sand and gravel do not. The map is dense: thousands of readings per field.

What it is for: drawing the zone boundaries. What it is not for: telling you the pH.

A worked example. A 38 ha arable block in Lincolnshire, silty clay loam over a sandy ridge. A contractor scans it at £11/ha: £418. The map shows three ECa bands (low, mid, high). You then take 4 samples per zone, so 12 samples, at £22 each for pH, P, K and Mg: £264. Total: £682, or about £18/ha.

That is the same money as a 2 ha grid, but you get boundaries drawn from thousands of readings, and you can reuse the scan for seed rate, nitrogen and drainage decisions, because texture doesn’t change. That reuse is where EM wins. You pay once and the map lasts essentially for the life of the field.

Where it fails: where pH variation is caused by management history (a field that was limed unevenly in the 1990s) and not by texture. The scan can’t see that. Sample anyway.

AI-predicted soil maps: cheap, useful, needs checking

This is the new option. Tools such as Farmotion-style platforms, Omnia (Hutchinsons), Contour (Agrii), SoilOptix, Cropwise (Syngenta) and Fieldview-type systems combine satellite imagery history (Sentinel-2 bare soil and NDVI across several seasons), yield maps, terrain (LiDAR elevation, slope, wetness index) and national soil data such as the Cranfield LandIS/Soilscapes to predict zones. Some then ask for a small number of samples to calibrate.

Note that SoilOptix is a different animal: it uses a gamma-ray sensor towed over the field, and reports close agreement with lab results for several properties. Treat it as a scanning service, priced around £15 to £20/ha. The satellite-only products are the genuinely AI-driven, cheap ones.

How reliable are they? Here is the honest answer. Published trials and agronomy company data show satellite-derived zone maps matching EM-derived zones reasonably well on fields with clear texture contrasts. They do much worse on fields with subtle variation, or on fields where yield history is patchy (set-aside years, a different crop each year with no yield map). A fair working figure is that the zones are right for texture-driven variation perhaps three times out of four, and unreliable for nutrient status unless you calibrate.

So the practical way to use an AI map is as a sampling guide:

1. Generate the zone map (£0 to £6/ha).
2. Take 3 to 4 samples per zone, placed in the zone’s typical areas, not its edges.
3. Compare the lab results between zones. If pH and P index differ meaningfully between zones, the map is earning its keep. If they don’t, drop back to one rate.

That last check matters. A zone map that doesn’t separate nutrient values is just a pretty picture.

Turning any of them into a variable-rate plan

This is where money is made or wasted. Here is a worked lime example.

Field: 24 ha, winter wheat then spring barley, target pH 6.5. Three zones from an EM scan, with calibration samples.

Total lime: (7 × 4.0) + (11 × 1.5) + (6 × 0) = 28 + 16.5 + 0 = 44.5 t.

A flat-rate approach, using the field average pH of about 6.3, would put on roughly 2 t/ha across 24 ha: 48 t. So you use slightly less, but more importantly it goes where it works. Zone A at 2 t/ha would still sit below target. Zone C would receive 12 t of lime it didn’t need.

At £22/t delivered and spread (ground limestone, check your local price), flat rate costs £1,056 and VR costs £979. The saving is only about £77. That is typical for lime: the cash saving is small, the agronomic gain is in fixing the acid patch, where yield losses of 0.5 to 1.0 t/ha wheat are plausible.

Now P&K, where the saving is bigger. Using RB209 indices for a cereals crop:

Maintenance figures here are illustrative for a mid-yield wheat crop; take the actual numbers from RB209 for your yield and index.

Flat rate at 60 kg P₂O₅ and 60 kg K₂O over 24 ha: 1,440 kg of each. VR: P₂O₅ is (7 × 85) + (11 × 60) + 0 = 595 + 660 = 1,255 kg. K₂O is (7 × 80) + (11 × 60) + 0 = 560 + 660 = 1,220 kg. You save about 185 kg P₂O₅ and 220 kg K₂O. At roughly £1.10/kg P₂O₅ and £0.75/kg K₂O (check current prices, they move), that is about £204 + £165 = £369 in one year, while the build-up on Zone A is actually done.

Over a 200 ha farm with similar variability, that is in the order of £3,000 a year, which covers an EM scan or an AI map plus sampling in the first year.

Getting the plan into the machine

Whatever method you use, the output has to become a prescription file. Practically:

• Format. Most spreaders (Amazone ZA-TS, Kuhn Axis, Bogballe, Sulky) accept ISOXML task data or shapefiles. Check which with your dealer before you buy a map.
• Software. Omnia, Gatekeeper, Muddy Boots-style tools, Farmplan and Fieldview can all export prescriptions. Ask whether export is included or a paid add-on.
• Lime spreading. Many contractors have VR-capable spreaders and weighbridges. Give them the file in advance and ask for an as-applied map afterwards. If they can’t supply one, you cannot check what was put where.
• Keep the as-applied map. Next sampling round, compare it with the new results. That is how you learn whether your zones were real.

If you’re not sure where to start with the wider toolkit, the satellite, drone and variable-rate agronomy guide covers imagery and spreader set-up in more detail.

What I would do on a typical farm

On a 150 to 300 ha mixed arable unit, I would do this:

1. Pay for one EM scan on the fields with the most visible variability. It is a one-off and serves you for a decade.
2. For the other fields, try an AI zone map from your agronomy provider if it’s included free. Treat it as a hypothesis.
3. Sample 3 to 4 points per zone, on a consistent date and depth, and only pay for full analysis (pH, P, K, Mg) once per cycle.
4. Write the lime plan first (it gives the clearest return), then P&K.
5. Run one field of VR next to a flat-rate strip for a year, if your spreader can do it. A yield map across the boundary is worth more than any sales brochure.

On permanent grass for dairy, beef and sheep, the same logic applies with a smaller budget: sample at 7.5 cm, split fields by obvious features (the wet bottom, the old arable headland), and put lime on the acid zones. A field at pH 5.5 will not respond properly to slurry or fertiliser, whatever you apply.

Start with one field and one quote. Ask your agronomist for the price per hectare of a 2 ha grid and of a zone-plus-calibration option, then put the two numbers side by side.