The Maersk Mc-Kinney Moller Institute, Faculty of Engineering, University of Southern Denmark, Odense, Denmark.
Department of Environmental Science, Aarhus University, Roskilde, Denmark.
Sci Total Environ. 2017 Dec 15;605-606:702-712. doi: 10.1016/j.scitotenv.2017.06.220. Epub 2017 Jul 1.
The assessment of air pollution exposures in epidemiological studies does not always account for spatio-temporal variability of pollutants concentrations. In the case of odor studies, a common approach is to use yearly averaged odorant exposure estimates with low spatial resolution, which may not capture the spatio-temporal variability of emissions and therefore distort the epidemiological results. This study explores the use of different exposure assessment methods for time-variant ammonia exposures with high spatial resolution, in rural communities exposed to odors from agricultural and livestock farming activities. Exposure estimations were based on monthly ammonia concentrations from emission-dispersion models. Seven time-dependent residential NH exposures variables were investigated: 1) Annual mean of NH exposures; 2) Maximum annual NH exposure; 3) Area under the exposure curve; 4) Peak area; 5) Peak-to-mean ratio; 6) Area above the baseline (annual mean of NH exposures); and 7) Maximum positive slope of the exposure curve. We developed binomial and multinomial logistic regression models for frequency of odor perception and odor annoyance responses based on each temporal exposure variable. Odor responses estimates, goodness of fit and predictive abilities derived from each model were compared. All time-dependent NH exposure variables, except peak-to-mean ratio, were positively associated with odor perception and odor annoyance, although the results differ considerably in terms of magnitude and precision. The best goodness of fit of the predictive binomial models was obtained when using maximum monthly NH exposure as exposure assessment variable, both for odor perception and annoyance. The best predictive performance for odor perception was found when annual mean was used as exposure variable (accuracy=71.82%, Cohen's Kappa=0.298) whereas odor annoyance was better predicted when using peak area (accuracy=68.07%, Cohen's Kappa=0.290). Our study highlights the importance of taking temporal variability into account when investigating odor-related responses in non-urban residential areas.
在流行病学研究中，空气污染暴露评估并不总是考虑污染物浓度的时空变异性。在气味研究中，一种常见的方法是使用每年平均的气味暴露估计值和低空间分辨率，这可能无法捕捉排放的时空变化，从而扭曲流行病学结果。本研究探讨了在农村社区暴露于农业和畜牧业活动气味的情况下，使用高空间分辨率的时变氨暴露评估方法。暴露估计基于排放-扩散模型的每月氨浓度。研究了七种与时间相关的住宅 NH 暴露变量：1）NH 暴露的年平均值；2）最大年 NH 暴露；3）暴露曲线下面积；4）峰面积；5）峰-均比；6）基线以上面积（NH 暴露的年平均值）；7）暴露曲线的最大正斜率。我们根据每个时间暴露变量开发了二项式和多项逻辑回归模型，用于感知气味和气味烦恼的频率。比较了每个模型的气味反应估计值、拟合优度和预测能力。除了峰-均比外，所有与时间相关的 NH 暴露变量都与气味感知和气味烦恼呈正相关，尽管结果在幅度和精度上有很大差异。当使用最大月 NH 暴露作为暴露评估变量时，预测二项式模型的最佳拟合优度，无论是用于气味感知还是烦恼，都能得到。当使用年平均值作为暴露变量时，对气味感知的预测性能最佳（准确率=71.82%，Cohen's Kappa=0.298），而当使用峰面积时，对气味烦恼的预测效果更好（准确率=68.07%，Cohen's Kappa=0.290）。本研究强调了在非城市住宅区调查与气味相关的反应时，考虑时间变异性的重要性。
