Performance evaluation of approaches to predict sub-hourly peak odour concentrations

Abstract Atmospheric dispersion models are widely used to predict the potential for annoyance of odour emissions. It is well established that the predictive accuracy of these models is higher for the mean concentration. In reality, the concentration of a substance advected by a turbulent boundary layer flow often shows large fluctuations around its mean value. In the case of odorous gases, the mean concentration field only may thus hide information on the real impact on receptors, since such a process exhibits nonlinearity. Short-term odour concentrations have to be accounted for, thereby requiring additional knowledge on concentration fluctuations. Here, the main goal is to evaluate three selected approaches to predict sub-hourly odour peak concentrations. These approaches are the constant factor of 4, an empirical-based peak-to-mean procedure and the concentration-variance computation. A full-scale field dispersion experiment (Uttenweiler), for which concentration fluctuation measurements have been conducted, was designated for the investigation. In this respect, additional objectives of this work are to investigate the mean flow conditions encountered in the different dispersion trials, to examine the measured tracer fast-response concentration time series, and to characterise the performance of a Lagrangian particle dispersion model (LASAT) for the mean concentration. Several statistical indicators are used to compare predictions against observations. First, there is evidence, gained from evaluating the ultrasonic anemometer data, to show that not only neutral but also stable atmospheric conditions occurred during the trials. Second, the dispersion model showed an overall satisfactory performance under the given range of study conditions. A general bias towards underestimation was detected, with the dispersion model performing better for distances further away from the emission source. Third, for the concentration fluctuations, results are presented and discussed in terms of the fluctuation intensity, intermittency, peak-to-mean factors and probability density functions. Finally, regarding the three evaluated approaches, it is found that as the inherent complexity of the approaches grows, more accurate predictions of peak-to-mean factors Ψ90 are obtained. Namely, the concentration-variance computation approach performed best, with bias towards overestimating Ψ90. While the constant factor of 4 overestimated all Ψ90 observations, the empirical-based peak-to-mean procedure underestimated Ψ90 to a great extent. The results also confirm and reveal the advantages and shortcomings of each evaluated approach. The findings of this work have potential implications for future research and policymaking in this topic area.