A published noise impact assessment for a battery energy storage project at Glen Innes in northern New South Wales contains a sentence that should be read carefully by anyone procuring this equipment. Describing the inverter, the assessment states that there is no noise data for the exact unit, and that the data used was based on a similarly sized previously measured inverter type.

That is not a criticism of the consultant. It is a disclosure, made properly, of a limitation the project could not resolve. The equipment was selected, the data did not exist, and the model was built on the nearest available substitute. The assessment then applied the resulting predictions against a night-time project trigger level of 35dB(A).

The question this raises is not whether that assessment was reasonable. It is why acoustic source data is so often the last item to arrive in a procurement process where it constrains almost everything else.

The source figure is doing four jobs, and a single number can only do one

A headline A-weighted sound power level is the figure that appears first in most supplier submissions. In an environmental noise assessment it is asked to support four separate tasks.

It has to propagate. Attenuation over distance is frequency dependent, because atmospheric absorption, ground effect and barrier screening all behave differently at 63 Hz than at 2 kHz. A model that starts from a single A-weighted figure has to assume a spectrum shape, and that assumption is often the largest source of error in the result.

It has to support a tonality decision. Every objective tonality test used in Australian regulation operates on one-third octave band data. An overall figure cannot demonstrate the presence or absence of a tone, and cannot show whether a modifying factor correction applies.

It has to inform mitigation. A barrier is a reasonable response to broadband fan noise and a poor response to a 100 Hz transformer tone. Choosing between them requires knowing where the energy sits.

It has to be capable of being verified. If the figure is a contractual commitment, the project needs to know what was measured, how, and under what conditions, so that a post commissioning measurement can be compared with it meaningfully.

Only the first of those can be attempted from an overall dB(A) value, and only badly.

Ask for the operating state, not just the level

Battery energy storage is not a constant source. Its acoustic output tracks its thermal load, and its thermal load tracks what the asset is being asked to do.

Cooling equipment stages up as cell temperature rises, which happens with high rate charging and discharging and with high ambient temperature. A site operating at low rate on a mild night and the same site absorbing a fast charge on a summer afternoon are different acoustic sources. Power conversion equipment behaves differently across its load range. Auxiliary equipment continues to run when the asset is neither charging nor discharging, which matters because a night-time assessment is often the governing case and a battery is rarely silent at idle.

A useful data request therefore names the states rather than asking for a level. Idle or standby with auxiliary load only. Charging and discharging at the rated continuous rate. Operation at the design maximum ambient temperature, since that is usually when cooling plant is loudest. And where the asset will provide frequency support or other fast response services, the operating pattern associated with those services, because it may involve frequent cycling rather than sustained output.

For most projects the governing case is high ambient temperature at night, and that is the state the data has to cover.

Ask for spectra, and be specific about resolution

Octave band sound power data is sufficient for outdoor propagation modelling. One-third octave data is what tonality assessment requires.

That distinction matters because Australian jurisdictions assess tonality on one-third octave bands, and they do not agree with one another. Work presented to the Australian Acoustical Society in 2015 comparing tonal noise regulations across Australia found that in New South Wales a tone is deemed objectionable where a one-third octave band exceeds both adjacent bands by 5dB above 400 Hz, 8dB between 160 and 400 Hz, and 15dB below 160 Hz, while Western Australia deems tonality present where a band exceeds the arithmetic average of the two adjacent bands by more than 3dB. The same comparison found more than 12dB of difference at low frequency between the point at which a tone would attract a 5dB penalty in those two jurisdictions. Several of these frameworks have been revised since that comparison was published, so the operative test for the applicable jurisdiction should be confirmed rather than assumed.

A supplier who provides octave band data has given the project enough to model. A supplier who provides one-third octave data has given the project enough to know whether a 5dB penalty is coming. On a site with a 35dB(A) night-time trigger, that penalty is the difference between a design that works and one that does not.

Ask which standard produced the figure, and which grade it is

Sound power levels are determined under a family of standards with materially different accuracy.

ISO 3744 describes methods of accuracy grade 2, engineering grade, for an essentially free field over a reflecting plane. ISO 3746 is the survey method, grade 3, applied where the requirements of ISO 3744 cannot be met. ISO 3745 is the precision method requiring anechoic or hemi-anechoic conditions.

The reproducibility values published in ISO 3744 are worth carrying into a specification, because they show where the uncertainty sits. The values below are those published in the 2010 edition, which was withdrawn on 11 December 2025 and replaced by ISO 3744:2025. The 2025 edition leaves the standard deviation of reproducibility for the main procedure unchanged, so the figures still stand, but confirm them against the current edition before they are written into a specification.

The uncertainty is twice as large in the 100 Hz to 160 Hz bands as it is in the A-weighted figure. Those are precisely the bands where transformer harmonics and low-speed fan content sit, and precisely the bands that propagate furthest and screen worst. A project that treats an octave band figure at 125 Hz with the same confidence as an overall dB(A) figure has misread the data it was given.

Transformers add a further complication. IEC 60076-10-1 notes that spatially averaged sound pressure levels are typically 2dB to 5dB higher than sound intensity measurements, because of near field effects. Two transformer figures determined by different methods are not directly comparable, and the method should be stated alongside the level.

Ask what was tested, and in what configuration

A sound power level belongs to a defined object in a defined arrangement. Several questions follow.

Was the figure measured on a single container or on an array, and if a single container, does it include the associated cooling equipment or exclude it. Was the transformer inside or outside the tested boundary. What was the fan speed or duty at the time of measurement, and was it the maximum. Was the unit in the configuration and orientation proposed for this site, including any factory fitted attenuation. And whether the figure is a measured result, a declared value with a stated uncertainty, or a guarantee.

Where the answer is that the data comes from a similar unit rather than the proposed one, as at Glen Innes, that should be recorded as an assumption in the design basis rather than absorbed silently into the model.

Do not ask for everything on every project

Proportionality matters. A small battery well removed from any sensitive receiver, assessed against a comfortable margin, does not need third octave spectra across five operating states.

The trigger for a detailed data request is the combination of a tight criterion, a close or acoustically sensitive receiver, a low background level, night-time operation and equipment types with known tonal character. Where those conditions are absent, an octave band sound power level at the governing operating state and a statement of the test standard will usually be enough.

Where they are present, an overall dB(A) figure is not a starting point. It is an unresolved risk carried into procurement.

Where better data changes the design rather than the report

The reason to obtain this information before procurement rather than after is that it changes what gets bought and where it goes.

Spectrum determines the treatment type. Broadband content from cooling plant responds well to barrier screening and to absorptive treatment, which is where barrier and screen walls built from Sonic System acoustic modular panels apply. The panels publish NRC up to 1.10 for the V100 and Rw up to 45dB for the V100SP, assessed to AS 1191-2002 and AS ISO 354. Low frequency tonal content from transformers and power conversion equipment responds differently, and screening alone will not resolve it.

Where equipment needs to breathe, the treatment has to pass air. That is the domain of Sonic Series acoustic louvres and Sonic acoustic attenuators, and the free area and pressure drop implications need to be reconciled with the cooling design rather than applied to it afterwards.

Layout responds to data as well. Knowing which unit types carry tonal content allows those units to be placed where the site itself provides screening, rather than distributing everything evenly and treating the result.

The specification is written before the model is run

The order that produces good outcomes on battery energy storage projects is not complicated. Establish the receiver criteria and the governing scenario. Write the acoustic data requirement into the request for tender so that suppliers price it. Assess the submissions on what the data shows rather than on the headline figure. Then model.

The order that produces late redesigns is the reverse. Select equipment on commercial terms, obtain whatever acoustic data the supplier happens to hold, substitute a similar unit where none exists, model the result, and discover the margin at the point where the layout is fixed.

A single dB(A) figure is a commercial statement. Octave or third octave sound power data, at named operating states, produced under a named standard, in a stated configuration, is a specification. The difference costs almost nothing at tender and a great deal afterwards.