Air quality model performance: A comparative analysis of 15 model evaluation studies

This paper presents a comparative analysis of the results of 15 air quality model performance evaluation studies involving 35 rural, urban, complex-terrain and regional models. Common themes are identified, and their implications for the improvement of model performance and usage discussed. Among study findings are the following:

• 1.(a) flat-terrain rural models nearly always predict the second highest concentrations within a factor of two (often less), with no evidence of systematic over- or underprediction, but some models may underpredict the highest concentrations by more than a factor of two at distances far downwind (20 km);
• 2.(b) none of the models accurately predicts spatial patterns of ? 24 h-averaged concentrations, and while increasing the averaging time from 24 h to annual improves performance for some models, it does not appear to do so for flat-terrain rural models; and
• 3.(c) while some models appear to predict more frequent concentrations (e.g. 5th- or 10th high, 95th percentile) better than they do the highest concentrations (e.g. 2nd high), no overall, systematic tendency is evident. Although definitive proof is beyond the scope of this paper, the hypothesis is offered that the observed performance of flat-terrain rural models may be attributable, in part, to slower-than-actual cross-sectional plume dispersion under neutral and stable atmospheric conditions, particularly in the vertical direction, but possibly also in the horizontal.