Six columns and one row per year, for one field at a time. Collection date, month, sampling depth, laboratory, the method printed beside the determination, and the result with its unit written out. Nothing converted and nothing filled in from memory: the date, the depth and the laboratory come off the header of the report, the month comes out of that date, and the method comes off the determination line beside the result.
Where any of those five changed from one year to the next, a horizontal line goes across the table, and one sentence under the table says what the line means, which is that from there back you are comparing against a different ruler. Two hours for the first field, twenty minutes for each one after that.
The reports exist, all of them, filed. Somebody finally stacks them and the series will not close. Year one went to one laboratory and year three to another, because the first one was backed up in the wrong week. Two years came back in parts per million and one in pounds per acre, or in kilograms per hectare where the laboratory is metric. One collection happened in May and the next in September, fifteen days after a fertilizer pass, because September was when there was time. One report names a method beside the number and another names none.
The year nobody sampled is simply absent, and the field was split in two halfway through, so half the series belongs to an area that no longer exists. Laid side by side, what appears is a line going up and down with no explanation, and the conversation ends with somebody saying that soil tests vary anyway. What that costs is the only long-run evidence the farm could have about the effect of what it has been doing, turned into noise, after five collections and five laboratory invoices.
What has to hold still for two numbers to be comparable?
Four things, and one technical note puts all four in the same list. The USDA Natural Resources Conservation Service, in Soil Health Technical Note No. 470-16 of May 2025, sets out what to consider when planning soil collection for testing. First the plan: “It is imperative that the chosen soil sampling plan be used consistently over time”. Then depth: “Soil should be sampled at the same depth 0-6 inches (0-15 cm) at each collection”. Then timing: “sampling should be conducted at the same time of year”.
Then the laboratory, with the reason attached: “Laboratories often have different instruments or protocols that will affect the results”, so “The same laboratory should be used over time to ensure results can be compared over sampling years”. That is a United States agency writing for conservation planning in the United States, and none of the four is American in any way that matters: a ruler that moves is a ruler anywhere. Three of them become columns in the table below, depth, month and laboratory. The fourth column that has to hold still is the method, which the note does not list because it arrives printed on the report rather than decided in the sampling plan.
Why is the method part of the number and not a footnote?
Because the method is what the number means. Antonio Mallarino and Angie Rieck-Hinz, writing for Iowa State University Extension and Outreach, put it as a definition: “Both the extractant chemical solution and the procedure used to measure the extracted nutrient define a soil test”, which is why “soil-test laboratories should clearly indicate what method is used for the test results provided”. Two procedures aimed at the same nutrient may not return the same amount, and which pair you are looking at decides whether they do: for phosphorus, the extracted P can be measured by a colorimetric procedure or by inductively coupled plasma, and the second measures more.
Their summary line is written for the person holding the page: “Labs need to clearly specify what tests they are using, and you need to know”. They also record that the field is sometimes missing, because “some laboratories provide test results for the Mehlich-3 without clearly stating which method was used”. So the method column has a third state besides a name, and that state is blank. A blank is information. It is the honest way to record a year whose report did not say, and it is what stops a later reader from assuming the method was the same.
Should you convert the old years into this year’s unit?
No, and the reason is not caution. Between the two potassium tests Iowa State supports, one run on dried samples and one on field-moist samples, the extraction can come out lower, similar or higher depending on the soil and the drainage, and “no simple factor can be used to transform results of one to the other”. The part worth carrying away is one step further in.
Some laboratories analyze phosphorus by one procedure and report the result as if it came from the other, “using some transformation factor each lab had determined by preliminary analyses of many samples using both procedures”. A number on a report can already be a converted number, converted by a factor that belongs to that laboratory and is not printed anywhere. Converting it again at the kitchen table stacks a second conversion on top of one you cannot see. Copy what was printed, write the unit out in full beside it, and let the break line carry the discontinuity instead.
Why does the month get a column of its own?
Because the date of collection moves the number, and the month is the part of the date you can compare. Embrapa Amazonia Oriental, in Documentos 266, tells the sampler to avoid taking samples when lime or fertilizer went on in the days before, and for perennial crops, where that situation is normal, to sample only thirty days after the last fertilizer application. That is Brazilian public research guidance written for Brazilian conditions, and the mechanism it describes travels: fertilizer that has just gone on is still sitting where the probe goes.
Put that next to the NRCS instruction to sample at the same time of year and the column pays for itself. A collection on the third of May and one on the twelfth of May are the same month and the same standing crop position. One in May and one in September are two different soils wearing the same field name, and no amount of care in the laboratory recovers that.
What the six columns are, and what a blank means
One row per year, six columns, and five of them copied rather than judged. Each row is one report read the way the soil test report by field sets out, so if that page already exists for a year, this row is a transcription and takes four minutes.
| Column | What goes in it | Where it comes from |
|---|---|---|
| Collection date | the day printed on the report, as printed | report header |
| Month | that month on its own, so a glance compares them | derived from the date |
| Sampling depth | the depth range with its unit, as printed | report header |
| Laboratory | the full name of the laboratory | report header |
| Method | the extractant and the procedure, or blank | the determination line, not the header |
| Result | the number with its unit written out in full | the determination line |
The method is the one field that lives on the determination line rather than in the header, because it belongs beside the result it defines: the expectation set for laboratories is that they state the method alongside the test results they provide. Blank is a legitimate entry in the method column and in no other. A blank date, a blank depth or a blank laboratory means the report was not read all the way, and those three are always printed somewhere on it.
What do you write in the year nobody sampled?
A row with the year and the words no collection, and nothing else on the line. Skipping the year closes the gap visually and makes four collections in nine years look like four consecutive readings. The years with no collection here are the same years that show up as gaps on the list that carries a declared sampling interval, seen from the other side. A field split in two partway through gets the same treatment: from the split forward it is two series, and the rows before it belong to an area that no longer exists, which the table says in a line rather than hiding.
Metadata written down is what keeps a series alive across the people who work it. BonaRes defines a long-term experiment as “field experiments with a minimum duration of 20 years and a static design”, and its merged collection with EJP SOIL, which took in shorter trials as well, holds “616 LTEs from 30 different countries across Europe with a duration of typically 20 years”, each one carrying its own written set of metadata. Those are European research stations with staff paid to keep records, so what crosses to a working farm is not the apparatus but the requirement under it: a design held still, and written down where the next person finds it.
Will the same laboratory still be the same laboratory in five years?
Keeping the same laboratory is a bet, and the bet has something behind it. The condition comes first, and the laboratory literature states it with a limit built in: most methods of soil analysis are standardized so that results are comparable among laboratories applying the same standard. Most of them, and not all. Then the test of whether that holds over years.
Two batches of agricultural soil went to “27 laboratories in Belgium, France, Germany, and Morocco”, blind, analyzed twice with the same standardized methods at intervals of seventeen and twenty-eight months. The great majority of them came back consistent with themselves across that gap, which is the finding a series can use: the same laboratory, working to the same standard, gave the same answer to the same soil more than a year later.
Two limits come with it, and both are the authors’ own. “the findings were specific to the two types of soil used for these tests”, and they say a wider range of soil types is still needed. And a study like that one needs “a group of laboratories that are strongly committed to the continuous improvement of their quality management systems”, which makes them the laboratories that already hold themselves to a standard, not the average of every laboratory taking samples. What survives both limits is enough to act on: the same laboratory, using the same standard, is the most stable of the four columns, and writing its name down is what makes the bet checkable later.
Who else writes the method beside the number?
The people who write the rules do it, which is the cleanest evidence that the number alone is not a number. Uruguay’s Ministerio de Ganaderia, Agricultura y Pesca records that the action plan for the Rio Santa Lucia basin requires, of every rural holding in that basin, control of nutrient application and fertilizing on the basis of soil analysis so that phosphorus does not go above 31 ppm, and beside that number it writes the method by which the number is measured, Bray1. That is Uruguayan regulation and it applies inside one basin, so the figure is nobody’s limit outside it.
The form is what crosses the border. Whoever has to defend a number in front of somebody else writes the method next to it, because a number that has to survive an argument cannot arrive without its ruler.
What goes at the top of the table, and who dates it?
One line, above the first row: same laboratory, same depth, same month, from this date forward, decided by a named person on a named day. The method is the fourth thing that has to hold still and it is not in that line, because the farm does not choose it: what the farm does about the method is read it off every report and write it in its column, so that a change shows up as a break instead of passing as a trend. Marking the breaks is what the farm does about the years it already has, and that line is what it does about the years it does not have yet.
The technical note gives the reason in one line, saying that “Repeating the sampling plan over time” can reduce the variability you would otherwise spend years arguing about. A rule with nobody’s name on it gets broken the first time the usual laboratory is backed up, and the person who breaks it will be right to, because there was nothing to weigh the delay against. A rule with a name and a date turns that week into a decision somebody makes and records as a break, which is a different event with the same outcome.
Where to start
One field, the oldest one on the list. Two hours for the first, twenty minutes for each field after it.
A series with three breaks marked beats a smooth line
A table with three horizontal lines through it looks worse than a clean column of numbers and tells you more. The clean column is usually clean because nobody recorded the two laboratory changes and the September collection, so the breaks are still in there, doing their work invisibly on whatever conclusion gets drawn. The first honest series most farms can build is short: one field, two years that actually compare, and a third marked off as belonging to a different ruler. Two comparable years are a series.
Five uncomparable reports are five reports. The same discipline of writing the condition beside the measurement runs through a compaction check with a date on it, where the measurement comes from a penetrometer and a spade rather than a laboratory, and through most of what sits under the production axis. All of it belongs to the third and fourth of the four functions of management, which is where you find out whether the ruler moved before you decide anything on the strength of it.