3.3  Factors Affecting Risk in Environmental Noise Predictions

The preceding section sets out the challenges facing any attempt to objectively rate an environmental noise field. In this section attention is now directed to the challenges and risks specifically relating to objective ratings derived from predictions.

The origins of this risk can be described by two broad categories described as follows:

• Assessment Conditions: The inherent variability of environmental noise levels presents the challenge of determining which source and propagation conditions should be used for the assessment. This challenge applies equally to measurement and prediction based studies. However, in the case of environmental noise predictions there will often be less information to quantify the full range of variability that may be expected in practice. Also, there is often a restricted range of conditions for which the available prediction methodologies can offer meaningful representations, e.g., some engineering methods only enable the calculation of noise levels which occur under downwind conditions, whilst in some cases, upwind conditions may be equally or more important.
• Prediction Quality: The accuracy with which the prediction model determines the noise level that would occur in practice for the exact physical conditions that the model is attempting to represent. These inaccuracies relate to the limitations of the input data and to the ability of the chosen sound propagation algorithm to represent actual transmission conditions.

In both of these two broad categories, the limits of application and accuracy of practical prediction techniques represent important considerations. These limits are discussed below in the context of the input data used to construct the models, available calculation models, proprietary software packages that implement these calculations, and the practical constraints typically encountered in dealing with these challenges.

3.3.1  Input Data

The key input information to most noise prediction studies is a representation of the sound emission level and character of the sources to be investigated. Where available, such information might comprise a test emission level deduced from measurements made in relatively controlled environmental and operational conditions. In other instances, emission characteristics may be deduced from empirical relationships according to the type of equipment under consideration and some aspect of its performance rating. In cases where no such information is available, an estimate may be acquired from field measurements of an installed item of plant. In all cases, the data will be a relatively simple representation of the total emission and character of a very complex sound-compgenerating mechanism. The total emission of a machine will comprise many contributions from individual components which have their own sound emission characteristics. This complexity introduces some important factors to bear in mind when considering the representation provided by sound emission data:

• Source directivity: Many types of noise sources have directional characteristics such that the noise level observed at a constant distance from the machine will vary according to the orientation of the machine. These characteristics are often not evident in sound emission data, and it is usually very difficult to establish the extent to which the emission in a particular direction may vary from the average quoted value.
• Source geometry: The distribution of the individual component noise sources associated with a complex machine will affect the pattern of the noise field that emanates from it. This is relevant to the source directivity as described above, but also has an important relationship to how the source’s sound field will interact with the surrounding environment. For example, a large piece of equipment may be almost entirely visually obstructed from a receiver location of interest by a solid screen; the question as to whether sound levels at the receiver location will be significantly reduced by the obstruction will be highly dependent on whether or not the exposed portion of the machine is a significant source of the machine’s total sound emission. As with directivity, the distribution of sound sources about a complex machine will often be very difficult to establish from the available emission data.
• Source input versus actual emission level and character: the actual emission level and character of the item of plant being modelled may vary from the input representation for a wide variety of reasons including manufacturing variations between the tested and installed item of plant, sensitivity of the installed plant item to mounting conditions and variations in the operational duty of the installed machine.
• Representation of the physical environment: An accurate representation of the physical dimensions of all acoustically significant features of a potentially large assessment area can be very problematic. The additional challenge is then the assignment of acoustical properties to surfaces that absorb, impede and reflect sound to varying degrees. Accounting for the complexities of the terrain and built environment generally requires simplifying assumptions to be made in order that they can be included in a noise model. Subtle factors and variations not represented in the estimation can lead to differences between predicted and actual noise levels, particularly where the influence of certain environmental features have interdependencies (e.g. the effect of screening and wind conditions cannot be considered in isolation from each other).

The complexity further extends to the atmosphere through which the sound must propagate. This is discussed in further detail in the following section.

3.3.2  Algorithms for Outdoor Sound Propagation

The ability of mathematical algorithms to accurately represent sound propagation has been the focus of considerable research, particularly given the role of noise prediction as an integral assessment tool in the fulfilment of the European Noise Directive (i.e. EU Directive 2002/49/EC, which requires member states to produce noise maps and action plans for urban areas and major transport infrastructures, including roads, railways and airports). Predictive algorithms vary widely in sophistication from commonly employed engineering methods (empirically based) through to more complex scientific methods that are mostly employed for specialist research applications. Engineering methods offer the benefit of robust and practical computation, but are generally limited to the prediction of longer term average A-weighted noise levels, and exhibit increasing uncertainty when attempting to evaluate noise fields with complex sources and/or propagation conditions. At the opposite end of the spectrum, scientific methods can provide significantly greater accuracy for complex situations, but generally only for very specific and limited scenarios, and are computationally intensive to an extent that limits their viability as a practical assessment tool.

The complexity of atmospheric conditions and the impracticability of measuring all the relevant environmental parameters throughout the sound propagation path, require that several assumptions and simplifications in the models are adopted. This set of assumptions and approximations has led to the existence of a variety of methods to mathematically represent sound propagation. Generally, all these methods can be classified as one of the following three categories:

• • Practical engineering methods
  • Approximate semi-analytical methods
  • Numerical methods

3.3.3  Proprietary Software

Given the practical complexity of applying predictive algorithms to the computation of a large number of sources over extensive areas, proprietary software packages are most commonly employed to generate predictive noise models. These packages provide an efficient and organised approach to the collection of input data, and then provide the option to execute various predictive algorithms to generate calculated noise levels over a wide range of receiver locations. The development of these software packages involves the translation of standard predictive algorithms (such as ISO 9613) into computational code. In order to achieve practical computation times, these software packages will often implement efficiency algorithms that enable users to strike a balance between likely computational accuracy and calculation time. Whilst such aspects of proprietary packages are clearly beneficial for practical assessment purposes, there is limited information available to the user to understand the extent to which such efficiency techniques may compromise the predicted value. Experience of several commonly implemented packages has indicated that variations in the approach to efficiency can potentially amount to significant variations in calculated outputs. At present, there are no international or British Standards that provide a user with any certification of a proprietary package's accuracy in applying a given predictive algorithm, and the credibility of a particular package will therefore often derive from brand recognition. This contrasts with objective studies based on measurements that require that any sound level meter used for such an exercise to be calibrated and demonstrated to achieve agreement with set reference conditions when tested in a laboratory scenario.

Furthermore, a common feature of most software packages is the assignment of standard atmospheric conditions which are then applied to the calculation of propagation from all sources. Wind direction related sound propagation effects is an important example of a factor for which such an assumption can be a source of error in the calculation. For example, industrial sites may be sufficiently large that some sources at the site are upwind of a calculation position at the same time as other sources at the site are downwind of the calculation, with potentially significant implications for the total received noise level in practice.

3.3.4  Other factors related to the ways in which models are used in practice

There are many other factors which influence the accuracy and usefulness of models in practice, including the following:

• Absence of nationally standardised requirements for the verification and quality assurance of commercial software which implement engineering methods, leading to a very wide range of performance and suitability for purpose.
• In the absence of clearly defined assessment requirements, the conditions that should be included in prediction models are often selected in a somewhat arbitrary manner.
• There is no defined system for generating traceable accounts of a model output's development/construction.
• There is varied industry understanding of modelling limitations.
• There is an absence of guidance on how to deal with the limitations and uncertainties of predictions.
• Noise modelling studies are generally constrained to fall short of the ideal, due to budget restrictions.
• Many environmental noise assessment projects are won through a competitive tendering process, so bidders are often obliged to limit their scope to one which allows for less rigorous investigations to be undertaken than the bidder would recommend under less competitive circumstances, and those commissioning prediction studies understandably seek the lowest cost options without necessarily understanding the potential trade-off between decreasing study cost and increasing outcome uncertainty. For example, the increased risk of consequential loss associated with increased uncertainty is not always appreciated.
• There may also be limited access to information: due either to limited resources to collect information, or as often occurs, due to the confidentiality surrounding certain information depending on the relationship between the practitioner, the party commissioning the study, and the noise producer.