Soil test extractants

Why does a chemical bath no root would ever perform predict what a crop needs?

Send a bag of soil to a laboratory and something strange happens to it. For phosphorus, the technician may shake it for five minutes in half-molar sodium bicarbonate held at pH 8.5 — the Olsen method — or in a dilute mix of hydrochloric and ammonium fluoride, which is Bray P1, or in Mehlich-3, a cocktail of acetic acid, nitric acid, ammonium fluoride, ammonium nitrate and a chelating agent. A number comes back, and a fertiliser recommendation is written from it.

No root does any of this. Roots exude organic acids and protons, host mycorrhizal partners, forage over months, and work at concentrations and timescales nothing like a five-minute shake in a strong reagent. So the obvious objection is a good one: how can a procedure that resembles nothing in the soil predict what a plant, which does something entirely different, is going to find there?

The instinct is to answer "because the extractant simulates the root". Hold that thought up to the light and it fails immediately — Bray and Olsen use opposite chemistries, one acid and one alkaline, and both work. They cannot both be simulating the same root. So simulation is not the mechanism, and something else has to be.

Try a different question. What would it even mean to measure "available phosphorus"? Phosphorus in soil sits in a continuum: dissolved in solution, adsorbed on iron and aluminium oxides or on clay edges, precipitated with calcium, bound in organic matter, locked into minerals. There is no line in that continuum marked "available", because availability depends on the crop, the season, the moisture, the rooting depth and the biology. The quantity being asked for is not a well-defined physical property of the soil at all.

That reframes everything. If the target quantity does not exist as a thing to be measured, then the extractant cannot be failing to measure it. What can the extractant be doing instead?

Here is the move. Consider a laboratory that runs an entirely arbitrary procedure — shake the soil in cold tea, say — and records a number. Now take that same soil from two hundred field plots covering a wide range of fertility, grow a crop on all of them, and record how much yield each plot gained from added phosphorus. If the tea number turns out to sort the plots reliably — low numbers responding strongly to fertiliser, high numbers not responding at all — then the tea number is useful, whatever it is chemically. It has become an index.

This is precisely how soil tests were built, and the two steps have names. Correlation asks whether the extractant number tracks what the plant actually takes up. Calibration asks the harder question: at what value of the index does a crop stop responding to added nutrient? That value is the critical level, and it is read off a response curve gathered from many field trials, not derived from chemistry.

Follow the consequences, because they are sharp. If the number's meaning comes from the trials, the number is only valid inside the conditions those trials covered — the same crop, the same soil group, the same climate. This is why Bray P1 is unreliable on calcareous soils: the acid is neutralised by free carbonate before it has done its work, so the number collapses for reasons that have nothing to do with the plant. Olsen's bicarbonate was developed for exactly those soils. Neither method is more "correct"; they have different domains of calibration.

It also explains why a soil test result is meaningless without its method attached. Fifteen milligrams per kilogram of Olsen P and fifteen of Mehlich-3 P are different statements about different soils, and a critical level from one cannot be applied to the other. And it explains the sensitivity to procedure — shaking time, soil-to-solution ratio, temperature. Change the recipe and you have not made a slightly better measurement; you have made a different index, whose calibration no longer applies. That is why methods are standardised so obsessively for what looks like a crude test.

Three refinements, and one misconception worth correcting directly.

The misconception is treating the number as a stock. "My soil has 18 ppm of phosphorus" invites the picture of a tank with a measured contents. It is better read as a position on a scale — a rank against the trials that defined the scale. Ask not "how much is there?" but "how likely is this crop to respond to more?"

First refinement: an index can be right about ranking and wrong about quantity, and for the purpose it serves that is fine. Fertiliser decisions are ordinal decisions — apply, apply less, do not bother — and an index calibrated to that question does not need to be a physical measurement to answer it well.

Second: this is why extrapolation is the real hazard. A new crop, a very different soil, an unusual pH, a management system unlike the trial plots — in each case the chemistry still runs perfectly and returns a confident number whose interpretation has quietly become unsupported. The failure is silent, because nothing in the analysis reports that you have left the calibrated range.

Third: as varieties, yields and rooting habits change, the response curves shift beneath a fixed chemical method, which is why calibrations must periodically be redone.

• How correlation and calibration trials are actually designed, and why they take years.
• Why the Bray and Olsen methods diverge on calcareous soils, in chemical detail.
• Whether isotopic dilution or diffusive gradients in thin films come closer to a mechanistic measure of availability.

Every term the collection defines is gathered in the glossary.