Noise measurement uncertainty and how to report it

Noise measurement uncertainty is the part of an acoustic report most often reduced to a sentence saying the instrument was calibrated. That is not an uncertainty statement. A properly constructed budget for an environmental measurement will usually combine instrument tolerance, calibration drift, position, weather, source operating variation and the correction for residual sound, and the combined figure is frequently large enough to matter to the conclusion.

It matters because the decisions these reports feed are threshold decisions. A rating level 2 dB above a background sits in one category of the assessment and 2 dB below sits in another, and if the measurement carries an expanded uncertainty of several decibels then the report needs to say what happens across that range rather than presenting one number.

Where uncertainty comes from in a noise survey

Group the contributors, because that is how the budget gets built.

• The instrument: class, frequency response tolerance, linearity, and the microphone and windshield response.
• Calibration: the reference sound source tolerance and any drift between the pre and post checks.
• Position: height, distance from the facade or from reflecting surfaces, and how well the measured position stands in for the receptor.
• Meteorology: wind speed and direction, temperature gradient and their effect on propagation over the measurement distance.
• The source: operating mode, load, duty cycle and whether the measured condition is the worst case reasonably expected.
• Analysis: the correction applied to remove residual sound from the measured ambient, and any averaging over time or across positions.

Each of these can be estimated. Combine them in quadrature as independent contributors, then apply a coverage factor to give an expanded uncertainty at a stated confidence level, conventionally around 95 per cent with a factor of two. That method is standard metrology and is set out in the acoustic standards themselves.

Instrument class, calibration and drift

Class 1 instruments have tighter tolerances than Class 2 across the frequency range, and for a measurement that will support a planning condition the difference is worth the hire cost. Where a Class 2 instrument has been used, the report should say so and carry the larger tolerance into the budget rather than quoting the same uncertainty as a Class 1 survey.

Field calibration before and after the survey is the minimum. The pre and post readings should agree closely, and a drift beyond a fraction of a decibel usually invalidates the measurement. Report both figures. A report that says the instrument was calibrated, without giving the readings, has not given the reviewer anything to check.

Two further items belong in the record. The date of the last laboratory verification of the instrument and the calibrator, and the windshield used, because measurements without one in any appreciable wind are not reliable.

Spatial and temporal variation

This is usually the largest contributor and the least often quantified.

Spatially, a position 1.5 metres from a facade reads differently from a free-field position, and a measurement at ground level may not represent a second floor bedroom window. Where the assessment position is not the receptor position, the transfer between them carries its own uncertainty and should be stated as such.

Temporally, background sound levels vary night to night with weather, traffic, insects, birds and the neighbours. A survey over three nights gives a spread, and the spread is data. Choosing a representative background from that spread is a judgement with several decibels riding on it, which is why survey duration carries so much weight, as our note on how long a background noise survey should run explains.

The residual correction deserves separate attention. Where the measured ambient exceeds the residual by a comfortable margin, subtracting to isolate the specific sound is reliable. Where the margin is small, the correction becomes very uncertain, and below a few decibels it cannot properly be made at all. A report that quietly corrects across a 2 dB difference has produced a specific sound level that means little.

Uncertainty in a predicted level

Prediction has its own budget, and the input uncertainties usually dominate the model.

The propagation standard commonly used for industrial and commercial sources states an estimated accuracy of the order of 3 dB for the distance and height ranges it covers, and that is before the uncertainty in the source data. A manufacturer's sound power level carries a tolerance. A directivity assumption may be optimistic. Ground absorption, screening from a barrier or a building, and meteorological conditions all carry judgement.

So a modelled level presented to one decimal place is a presentational error. Report the prediction with its uncertainty and identify which input the result is most sensitive to, because that is the input the reviewer should test. The same discipline applies in dispersion modelling for an Air Quality Assessment, where verification against monitored data is the accepted way of bounding the model.

Sensitivity testing is more persuasive than an error bar. Run the model with the plant at its upper tolerance, with the barrier 200 millimetres lower than drawn and with a downwind condition, and report whether the conclusion holds.

When uncertainty changes the conclusion

Uncertainty only matters where it crosses a decision boundary. Three situations arise.

The conclusion holds either way: the predicted rating level sits well below the criterion, and even at the unfavourable end of the range the answer does not change. Say so in one sentence and move on.

The conclusion is marginal: the range straddles the criterion. This is where the report has to be explicit, and where the honest recommendation is usually additional mitigation with margin, or a condition written against a measured limit so that compliance is tested after installation.

The conclusion depends on an untested assumption: the range is narrow but the central estimate rests on a manufacturer's figure nobody has verified. The remedy is a commissioning test, not a wider error bar.

Illustrative example: a plant enclosure predicted to give a rating level of 41 dB against a representative background of 40 dB. The prediction carried a combined uncertainty of around 3 dB, so the outcome ranged from clearly below background to 4 dB above it. The assessment recommended additional attenuation to bring the central estimate to 37 dB and a condition specifying a measured limit at the facade. The figures are illustrative and used to show how the range drives the recommendation.

Presenting a range rather than a single figure

Three presentational habits make an uncertain result usable.

State the central estimate, the combined standard uncertainty, the coverage factor and the resulting range, in that order, in a table. One row per assessment position.

Keep the uncertainty symmetrical unless there is a reason for it not to be. Applicants sometimes subtract the uncertainty from their own predicted level while ignoring it when it would work against them, and a reviewer should reject that immediately.

Separate measurement uncertainty from scenario uncertainty. The question of how accurately you measured the plant is different from the question of whether the plant will run at night, and mixing the two produces a range that cannot be interpreted.

What a reviewer should challenge

Work through this list on any Noise Impact Assessment that supports a condition. Councils normally expect a suitably qualified acoustician and professional review is strongly advised, because most of these points turn on judgement rather than arithmetic.

1. Are the pre and post calibration readings given, and do they agree.
2. Is the instrument class stated, and does the uncertainty budget reflect it.
3. Are measurement positions described by height and distance, with photographs.
4. Is the wind speed recorded for each measurement period, and was a windshield used.
5. What was the margin between ambient and residual, and was a correction applied within a reliable range.
6. How was the representative background chosen from the measured spread, and what would a different choice do to the conclusion.
7. For predictions, what is the source data source, what tolerance does it carry, and which input is the result most sensitive to.
8. Is the uncertainty applied symmetrically.
9. Does the recommended mitigation carry margin over the criterion.

And ask the simplest question last. If the measurement were repeated next week by someone else, would the report's conclusion be the same. A report that cannot answer that has not dealt with uncertainty, whatever it says about calibration.

This is general guidance, not legal advice, and the uncertainty in any particular assessment depends on the instrument, the site and the source, so check the position with your local planning authority and a qualified acoustician before relying on it. PlanScope produces AI-assisted drafts structured to UK planning policy, leaving measured data and uncertainty analysis to your acoustician. Enter your postcode to see which reports your site needs.

PlanScope provides AI-assisted draft reports for planning applications. Reports should be reviewed before submission. We do not guarantee planning approval.