Validating UV reactors: UV dose, DIN 19294 and sensor placement

Validating a UV reactor demonstrates which disinfection performance is achieved within a defined operating range. What governs this is not lamp power or a single irradiance reading, but the relationship between flow rate, UV transmittance, sensor reading and biodosimetrically determined effect. The dose received by the least favourably treated parcel of water cannot be measured directly in a flow-through reactor; it can only be inferred.

In the German public drinking water supply, UV disinfection is the only approved physical disinfection process under the German Environment Agency’s (UBA) list of treatment substances and disinfection processes pursuant to § 20 of the German Drinking Water Ordinance (DVGW). How the proof is to be furnished is regulated accordingly closely.

This page compares DIN 19294, ÖNORM M 5873, the UV Disinfection Guidance Manual of the US EPA and NSF/ANSI 55, places fluence and reduction equivalent fluence in context, and describes the measurement chain of collimated beam test, process sensor and traceable reference measurement.

The fundamentals of measuring at 254 nm – spectral matching, measurement geometry and calibration – are described on the page measuring UVC radiation at 254 nm.

Three terms circulate for what looks like one quantity, and the difference between them is the difference between a laboratory result and a plant approval.

• Fluence (H, in mJ/cm² or J/m²) is the radiant energy arriving at a point from all directions. In a stirred, thin, uniformly irradiated sample under a collimated beam it is calculable to within a few percent. It is the quantity in every published dose-response table.
• Dose is used loosely for the same thing. In a flow-through reactor it is not one number but a distribution: residence times differ, the radial irradiance profile differs, and every parcel of water collects its own value.
• Reduction equivalent dose (RED) is what a validated reactor is rated in. It is measured backwards: a challenge organism with a known dose-response curve is passed through the reactor, the log reduction at the outlet is measured, and the RED is the collimated-beam fluence that would have produced the same reduction. It is an effect expressed in units of dose.

The consequence is uncomfortable and rarely stated plainly: the RED depends on which organism was used to measure it. A resistant surrogate survives further into the reactor, samples more of the dose distribution, and yields a higher RED than a sensitive one would – for exactly the same reactor at exactly the same setting. This is the RED bias, and the US EPA's validation factor exists to correct for it. German and Austrian practice sidesteps the problem instead, by fixing one organism and one number for everyone.

The four regimes below all certify that a UV reactor disinfects. They disagree about what has to be proven, against which organism, and how the plant is supposed to know during operation that it still holds.

Dose figures from different sets of rules are comparable only where reference wavelength, test organism, log credit, validation factor and operating model agree. An identical numerical value can serve a different evidentiary function in different approval systems. The design must therefore be referred to the set of rules required at the site of use and to the specific reactor approval.

The second row of the table shows this most clearly. The EPA dose requirements derive from the log credits of the LT2 rule and differ by more than an order of magnitude depending on the target organism: three log Cryptosporidium about 12 mJ/cm², three log Giardia about 11, four log for viruses about 186. The flat value of 400 J/m² under DIN 19294 therefore lies above the protozoa requirement and below the virus requirement – each relating to a different evidentiary purpose.

The two figures answer different questions. The German and Austrian requirement is a conservative flat value for a supply whose raw water is already protected and monitored. The American one is a per-pathogen credit within a multi-barrier chain in which UV forms the barrier against protozoa while virus reduction is covered chemically. Transferring one numerical value to the other system therefore presupposes that target organism, log level and barrier concept are transferred with it.

The figures state the commonly applied requirements. What governs in each case is the currently valid edition of the standard and the approval the reactor in question actually holds.

Every one of these regimes ends up relying on a single sensor in the reactor wall to confirm, second by second, that a dose distribution across the whole cross-section is still what it was during validation. The sensor sees one point. The requirement is about a volume. The bridge between them is built during validation, and it holds only if the sensor stays where it was put then.

Four properties of that position do the work:

• Distance from the lamp. Close to the sleeve, the reading is dominated by one lamp and follows its ageing faithfully, but says little about the far wall. Further out, the reading is a weighted average over several lamps and over the water between them – it carries transmittance information, but responds sluggishly to a single lamp failing.
• Angular response. A sensor with a wide field of view integrates over a large part of the cross-section; a narrow one looks at a tube of water. That is why DIN 19294-3 fixes the aperture angle of the reference radiometer at 160° instead of leaving it to the manufacturer: two sensors with the same calibration and different acceptance angles read differently in the same reactor, and only one of them matches the validation.
• Window fouling. The sensor window fouls at a different rate from the lamp sleeve, because it sits in a different flow regime. When it fouls faster, the plant derates itself and wastes energy. When it fouls more slowly, the plant reports a dose it is no longer delivering – the failure mode that matters.
• Direction of error. The one property worth insisting on: the sensor position must be one where the reading falls at least as fast as the minimum delivered dose when transmittance drops or a lamp ages. A conservative sensor position is one that fails early.

This is why the sensor cannot be treated as a component chosen after the reactor. It is part of what was validated. Replacing it with a different type, or moving the port, voids the relationship the approval rests on. The practical consequence is a pairing: a permanently installed duty sensor in the process, and a traceable reference measurement at defined intervals to tell drift apart from a real change in the water. How sensors differ in spectral response and geometry is set out under selecting UV sensors, and what happens inside one under how UV sensors work.

Before a reactor is validated, the dose-response curve of the organism and the behaviour of the actual water have to be known. That is bench work, and the instrument for it is the collimated beam device: a UV-C source above a stirred sample dish, shielded so that the radiation arriving at the surface is near-parallel and the irradiance across the dish is uniform enough to be treated as one number.

That number is not the raw sensor reading. The established protocol – Bolton and Linden, widely adopted as the reference method – corrects the measured surface irradiance by four factors before it becomes an average fluence rate in the sample:

• Petri factor – the ratio of the average irradiance over the dish area to the irradiance at its centre, mapped point by point.
• Reflection factor – the fraction that passes through the air-water interface, close to 0.975 at 254 nm for normal incidence (around 2.5 % is reflected).
• Divergence factor – the correction for the beam not being perfectly parallel over the depth of the sample.
• Water factor – the attenuation over the sample depth, (1 − 10^−ad) / (ad · ln 10) for decadic absorbance a per cm and depth d in cm. It is the factor that makes a real water matrix behave differently from a clear buffer.

Multiply the four by the measured irradiance, multiply by the exposure time, and the result is a fluence in mJ/cm² that can be compared with the literature – and with the RED a reactor achieves. Leave them out and the bench result is optimistic by anything from ten to forty percent, which is easily enough to make a reactor look adequate when it is not.

Bench-scale work also answers the questions a validation certificate cannot: how this particular raw water behaves, whether the organism of concern on this site matches the surrogate, and how far the dose-response curve tails off at high log reductions.

A collimated beam device for this purpose needs a defined irradiated area, a measured irradiance and dose-controlled exposure, so that an experiment is set as a fluence rather than as a stopwatch reading. One instrument of this design is the BSH-03 CBD.

Three instruments, three roles. Each is a weak link without the other two.

• The dose-response curve. The BSH-03 CBD collimated beam device produces the bench-scale fluence against which any reactor result is later interpreted. Without it, the RED has no reference.
• The process measurement. The UVC-SE is a pressure-proof UV-C sensor for installation directly in the flow, measuring at 254 nm through the reactor wall where the validation put it. It is the instrument that has to hold its reading between service intervals – not the one that has to be most accurate on day one.
• The traceable check. The DVGW reference radiometer is what the duty sensor is compared against – built to DIN 19294-3, with the 160° aperture angle and the pressure-tight measuring window tube the standard prescribes. The rhythm is fixed too: annual calibration, monthly reference measurement. It closes the chain to the national standard and turns sensor drift into a measured number instead of an assumption.

Where the source is not a monochromatic low-pressure lamp – a medium-pressure lamp, say, or a UV-C LED array with a manufacturer-dependent peak wavelength – the broadband sensor's spectral mismatch becomes a question of its own, and a spectral measurement with a SR900 spectroradiometer is needed to quantify it. What the underlying radiometric quantities mean, and why an electrical power rating is not one of them, is set out under radiometric quantities. The dose-response figures behind the log stages, together with a calculator, are on the UV disinfection page.

What is the required UV dose for drinking water?
It depends on the regime. DIN 19294 and ÖNORM M 5873 require a minimum of 400 J/m² (40 mJ/cm²), referenced to Bacillus subtilis spores at 254 nm, as a single figure for all pathogens. The US EPA sets no single figure: the required reduction equivalent dose follows from the log credit claimed and the target organism, from about 12 mJ/cm² for 3-log Cryptosporidium to about 186 mJ/cm² for 4-log virus inactivation. NSF/ANSI 55 requires 40 mJ/cm² for Class A devices and 16 mJ/cm² for Class B.

What does reduction equivalent dose mean?
The RED is the collimated-beam fluence that would produce the same log reduction as the reactor produced in the biodosimetry test. It is not a measured dose but a measured effect converted into dose units, which is why it depends on the challenge organism used. A more resistant surrogate yields a higher RED for the same reactor – the RED bias that the EPA validation factor corrects for.

How do DIN 19294 and the US EPA UVDGM differ?
DIN 19294 certifies a reactor type against one fixed organism and one fixed dose, and the plant then operates within the approved envelope. The UVDGM validates an operating envelope of flow, UV transmittance and sensor reading, and awards log credit per pathogen with an explicit validation factor for uncertainty. The German approach is simpler to apply; the American one is explicit about what is being credited and how uncertain it is.

Why does the position of the UV sensor matter so much?
Because the sensor is the only continuous evidence that the validated dose distribution still exists. Its distance from the lamp, its acceptance angle and the fouling rate of its window all decide whether the reading falls as fast as the actual minimum dose when the water or the lamp changes. A position where the reading falls more slowly reports a dose the reactor is no longer delivering. The requirements for measuring window, aperture angle and calibration are laid down in DIN 19294 – parts 1 and 3 for low-pressure devices, parts 2 and 4 for medium-pressure devices since April 2026, which superseded the DVGW worksheets W 294-2 and W 294-3 entirely. Device requirements are now found only in DIN 19294.

What is a collimated beam device used for?
To determine dose-response curves and to test a specific water matrix at bench scale, under conditions where the fluence is calculable. The measured irradiance is corrected by the Petri, reflection, divergence and water factors to give an average fluence rate in the sample. These bench results are the reference against which a reactor's RED is interpreted.

Can a reactor validated to DIN 19294 be used in the United States?
Not on the strength of that approval alone. The validation protocols, the challenge organisms and the credited log reductions differ, and a US public water system claiming log credit needs a validation carried out under the UVDGM or an equivalent accepted by the state primacy agency. The reactor may well be capable; the evidence has to be produced again in the form the other regime recognises.

Standard and dose figures in these answers refer to the position as of September 2026. Before any design or approval, the currently valid edition of the standard must be consulted.

Between the irradiance a wall sensor reports and the dose the least favourably treated water actually receives lie transmittance, residence time and sensor geometry. Opsytec builds the instruments for all three steps – collimated beam devices for the bench, pressure-proof UV-C sensors for the process, and DVGW reference radiometers for the traceable check – and runs an accredited calibration laboratory behind them. Tell us about your plant or your test task.