The nitrogen figure printed on a bag of fertiliser is a legally declared value. In the European Union, Regulation 2019/1009 sets out the label rules for CE-marked fertilising products, and total nitrogen is one of the declarations that determines the product’s classification and its conformity with the value on the sack. A granule with less nitrogen than the label claims is a non-conforming batch; a granule with more is a giveaway that costs the producer margin.
The reference chemistry for that number - Kjeldahl digestion and Dumas combustion - is a century old, well understood, and slow. It runs in a laboratory on a discrete sample, with a turnaround measured in hours and a sampling frequency measured in tonnes. Continuous granulation lines, by contrast, run steadily, sometimes for weeks, and produce a granule whose surface texture, coating, and mineral filler load vary from hour to hour.
This piece walks through where the reference methods sit, what continuous producers actually want from a measurement, and where the three principal candidates for a faster loop - near-infrared spectroscopy, Raman spectroscopy, and laser-induced breakdown spectroscopy - land in practice.
Why total nitrogen is the parameter that matters
Nitrogen is the nutrient that gives most fertilisers their headline number: the “N” in an NPK grade like 20-10-10. It is delivered in a fertiliser granule in several chemical forms - ammoniacal (as ammonium sulfate or ammonium phosphate), nitric (as ammonium nitrate or calcium nitrate), and amidic (as urea) - and the label declares the sum. The regulator does not care which form is present, only that the total matches the value on the sack within a permitted tolerance.
That has a knock-on effect for measurement. A method that measures only nitrate, or only ammonium, or only the intact urea molecule, does not on its own return the label figure. To close the total-N loop, an inline analyser has to either see all three forms directly or be calibrated to a matrix in which the ratio between forms is stable enough that a single strong signal predicts the whole.
What “reference” means for total nitrogen
Two methods carry the reference label almost universally. In Kjeldahl digestion, the sample is boiled in concentrated sulfuric acid with a catalyst until all organic and reduced nitrogen is converted to ammonium sulfate; the ammonia is then distilled off and titrated. In Dumas combustion, the sample is burned in oxygen at around 900 degrees and the evolved nitrogen gas is measured by thermal-conductivity detection after CO2 and water removal.
The two agree closely on most fertiliser matrices, but they differ on scope. Kjeldahl does not convert nitrate quantitatively unless a pre-reduction step is added. Dumas measures everything that combusts, nitrate included, and is the modern default for automated laboratories. Both are destructive, both require gram-scale samples, and both need a properly trained laboratory. The methods annex retained from the now-repealed Regulation 2003/2003 is still cited by European producers as the practical procedural guide for the nitrogen forms declared under 2019/1009.
Where the lab loop breaks
The reference methods work fine when the goal is release testing of a finished lot or an occasional quality check. They start to break down when the plant wants the number often enough to catch drift on the granulator or the coater in time to correct it. A four-hour lab turnaround on a line that produces sixty tonnes of granule in that same window means several tonnes of off-spec product before the operator sees the deviation.
The vocabulary distinction between inline, online, at-line and offline matters here. What most producers actually want is not truly inline sensing at the granulator throat but at-line measurement at a sampling station a few metres from the line: fast enough to close the feedback loop, forgiving enough to survive coating dust and vibration, and cheap enough to justify per line rather than per plant.
Near-infrared spectroscopy
Near-infrared (NIR) is the most mature candidate. It reads combination and overtone bands of N-H, O-H, and C-H, so it responds strongly to urea, to the water and ammonia bands of ammonium salts, and to any coating hydrocarbons. Diffuse-reflectance NIR heads are compact, robust, and available from several instrument houses; they are widely used offline in fertiliser laboratories as a rapid replacement for Kjeldahl on stable product families.
The pinch point is calibration. NIR bands are broad and overlapping, and the model that predicts total nitrogen has to be trained on samples that cover the full range of matrix variability the plant will actually see: filler load, coating thickness, moisture, temperature. On a line that produces one grade steadily, this is tractable. On a line that switches between grades weekly, calibration transfer becomes the dominant engineering problem, and the tools that solve it - orthogonal signal correction, generalised least squares, direct standardisation - are the same chemometric techniques any spectroscopic PAT project relies on.
Raman spectroscopy
Raman looks at the same molecular vibrations from the other side: it responds to changes in polarisability, which makes the nitrate stretch near 1050 cm-1 sharp, intense, and effectively free of interference in most fertiliser matrices. On a granule whose nitrogen is delivered mainly as nitrate - calcium ammonium nitrate, straight ammonium nitrate, NPK grades built around AN - a single, well-resolved band tracks nitrogen content with a linearity and sensitivity that NIR cannot match.
The trade-off is fluorescence. Coatings, mineral fillers such as dolomite, and organic residues from the granulation process can luminesce under visible-laser excitation and swamp the Raman signal. The pragmatic response is longer-wavelength excitation - 1064 nm rather than 785 nm or 532 nm - which pushes the excitation below the absorption thresholds of most fluorophores at the cost of some Raman intensity. Where 1064 nm is not sufficient, moving to at-line measurement with a rotating sample stage that averages signal from a representative area of granule surface tends to close the gap. The pattern is the same one that shows up in inline Raman probes for fouling-prone reactor media: the physics of the probe interface, not the spectrometer, usually decides whether the method works.
LIBS and other approaches
Laser-induced breakdown spectroscopy is the newest of the three in a fertiliser plant. A short, high-energy laser pulse ablates a microscopic amount of the granule and the emission of the resulting plasma is dispersed and read. LIBS sees elemental nitrogen directly, without regard to chemical form, and can run at a shot rate high enough to survey a moving conveyor.
Its weakness is quantification. The plasma is highly matrix-sensitive, ablation depth varies with granule hardness and coating, and calibration for total nitrogen requires stringent control of standards. LIBS is finding traction as a fast identification and grade-check tool - is this the right product, roughly the right composition - more than as a certified quantitative measurement replacing Dumas.
Other techniques (X-ray fluorescence, microwave, dielectric methods) show up in narrower niches, usually for a specific auxiliary parameter rather than for total nitrogen itself.
The coated-granule problem
Modern fertiliser granules are seldom naked. Coatings - anti-caking agents, controlled-release polymers, hydrophobic waxes - are applied to reduce dust, prevent moisture pickup, and slow nutrient release. Every one of these coatings interposes an optical layer between the analytical probe and the nitrogen-bearing crystal underneath.
That has different consequences for different techniques. NIR reads a millimetre or so into the granule and generally averages through thin coatings, at the cost of some sensitivity. Raman with focused optics can be optimised to sample the coating, the interior, or a weighted average of both, depending on the working distance and the confocal geometry. LIBS ablates the surface, so its first shot reads the coating and later shots read the interior - useful for depth profiling, but complicating a single-number total-N readout.
Where each approach lands
None of the three is a drop-in replacement for Kjeldahl or Dumas as the certified reference. Each is a candidate for a faster process loop that runs alongside the reference laboratory rather than replacing it, and each is picked for reasons a plant engineer can articulate concretely.
For urea-heavy grades and product families where the plant runs the same recipe for long periods, NIR is the incumbent. For nitrate-dominant grades, coated granules with mineral fillers, and situations where fluorescence has already been characterised and pushed down with 1064 nm excitation, Raman with an at-line rotating station is the strongest candidate; several process-Raman vendors, including at least one European specialist that publishes application notes on ammonium-nitrate granules, position their instruments explicitly for this workflow. For rapid grade identification and screening rather than certified quantification, LIBS is worth evaluating.
The consistent lesson from producers who have moved beyond a laboratory-only workflow is that the choice is dictated by matrix and coating, not by a vendor scorecard. A method that works cleanly on one plant’s granule can be silent on another’s. Feasibility on the actual granule, with the actual coating and filler package, remains the only reliable predictor.