Fertiliser and phosphoric acid production sit in an awkward middle ground for process spectroscopy. The analytes are simple ions and small oxo-anions - P2O5, sulfate, nitrate, urea, ammonium - present at percent levels rather than trace. But the matrices are hostile: strongly coloured sulphuric-phosphoric mother liquors, wet slurries carrying gypsum, dust-laden granulator off-gas, and cooling loops that carry corrosive vapours. That mix defeats a lot of the received wisdom that process spectroscopy is a solved problem.
Two techniques dominate the conversation for inline measurement here: near-infrared (NIR) and Raman. Both are non-contact optical methods that deliver a spectrum in seconds and lend themselves to chemometric quantification of several species at once. That is where the similarity ends. Below is how to think about the choice, analyte by analyte, before you write the URS.
What each technique actually sees
NIR reads overtones and combination bands of X-H bonds - O-H, N-H, C-H. The bands are broad and heavily overlapped, so quantification always leans on multivariate calibration built to ASTM E1655. NIR is at its best where the analyte carries a distinctive X-H fingerprint (water, urea, ammonium, alcohols, aliphatic organics) and the matrix is either transparent or a well-defined solid. It is at its worst when the analytes are inorganic oxo-anions with no X-H bonds, when the sample is very dark, or when a metre-scale water background dominates the near-infrared window.
Raman reads fundamental molecular vibrations. Bands are narrow and structurally specific, which lets you resolve phosphate (P-O around 890 and 970 cm-1), sulphate (S-O around 980 cm-1), nitrate (N-O around 1050 cm-1) and ammonium (N-H around 3050 cm-1) as separable peaks in a single spectrum. Water is a weak Raman scatterer, so aqueous or slurry chemistry is not automatically ruled out. The two things Raman handles badly are fluorescence from coloured or organic-loaded liquors, and particulates that scatter the excitation laser before it reaches the analyte volume. Both are common in fertiliser plants.
For a general treatment of the technique-selection question, see our earlier Raman vs NIR decision framework; the framework here specialises it for phosphate and nitrogen chemistry.
Fertiliser granulator, cooler and coating loops
The measurement most granulator operators want is total nitrogen, urea or ammonium content, and moisture - measured on the moving granule bed on a belt or in a chute. NIR wins this outright. The N-H, O-H and C-H combination bands of urea and residual water are strong and separable, the geometry is diffuse reflectance across a wide field of view, and the granule bed averages out particle-size effects if the measurement window is large enough. Process NIR has been used in this configuration for two decades and is well documented in the Handbook of Near-Infrared Analysis.
Raman does not play here for three reasons. The measurement volume is millimetre-scale rather than centimetre-scale, so the shot-noise on a moving bed is much worse. The granule surface often carries organic anti-caking coatings that fluoresce. And nitrate and ammonium in solid granules give respectable Raman signal, but urea’s dominant Raman band overlaps with them in a way that penalises calibration transfer between formulations.
The exception is when the target is a specific micronutrient or contaminant band that NIR simply cannot resolve - for example, chloride content limits under Regulation (EU) 2019/1009 for certain product function categories. There, a targeted Raman measurement at a diverted sample point can add value alongside the primary NIR gauge, but as an at-line lab tool, not as the granulator loop’s primary inline analyzer. Total-nitrogen measurement on granules is covered in more depth in Total nitrogen in fertiliser granules.
Phosphoric acid wet-process reactors
The wet process for phosphoric acid attacks phosphate rock with sulphuric acid, yielding a hot slurry of phosphoric acid, sulphate, calcium sulphate crystals and unreacted rock. Operators care about P2O5 and free H2SO4 concentration, and about the P2O5-to-H2SO4 ratio, because both drive yield and product spec.
NIR is a poor fit here. Phosphate and sulphate have no X-H bonds, so what NIR actually sees is dominated by water - a useful moisture signal, but not the analytes of primary interest. The dark-brown slurry compounds the problem: attenuation over the NIR path length gets steep, calibration windows narrow, and probe fouling on wetted optics is chronic.
Raman is far better suited on paper, and does work in the clarified liquor. Phosphoric acid’s 890 cm-1 P-OH stretch and sulphate’s 980 cm-1 S-O symmetric stretch are separable in the same spectrum, the water background is minimal, and 785 nm excitation with narrow-linewidth lasers keeps fluorescence tractable for most rock feedstocks. The complication is the crystalline solid phase. An immersion probe in the raw slurry receives spectra dominated by elastic scattering off calcium sulphate particles and interference fringes from the sapphire window - the analyte signal is buried. The practical implementation route that works is either an at-line configuration with a filtered clarified sample, or an inline installation on a slipstream where the solids are removed by an in-loop filter or cross-flow module before the optical cell. This trades some cycle time for a spectrum that is actually quantitative.
Vendor choice matters more than usual here because probe design determines whether the analyzer survives contact with hot sulphuric-phosphoric slurry at all. See our Inline Raman buyer’s guide 2026 for the peer set that publishes chemical-service ratings on wetted parts, and Inline Raman probes: fouling and reactor media for the fouling side.
A short decision framework for these streams
The five questions worth asking before you specify:
- Do the analytes have X-H bonds? If yes (urea, ammonium, water), NIR is the default. If no (phosphate, sulphate, nitrate as separated species), Raman is the default.
- How dark is the liquid, or how coated are the solids? Deep colour and organic coatings favour NIR for solids and disqualify Raman for direct contact with the mother liquor unless fluorescence rejection is engineered in.
- What is the particulate loading? Above roughly 2-5 percent solids by mass in a Raman-optical path, plan on filtration or a clarified slipstream. NIR tolerates more particulates but pays in path-length variability.
- What is the measurement window? Centimetre-scale diffuse reflectance across a granule bed is NIR territory. Point measurement in a defined liquid volume is Raman territory.
- What is the regulatory reference method? If the plant reports against a wet-chemistry reference (titration for free acid, Kjeldahl for total nitrogen), the chosen inline method has to be calibrated against it and revalidated when the reference is updated. That is a workload question, not a technique question, but it should sit in the URS.
For most fertiliser plants the answer is a mixed installation: NIR on the solids trains and at moisture control points, Raman on the liquid trains where phosphate and sulphate quantification matter, at-line lab spectroscopy anchoring both. Choosing one technique to cover the whole plant usually costs more in calibration effort than running two techniques where each is at its natural best.