Application of Air Quality Multi-Model Forecast System in Guangzhou: Model Description and Evaluation of PM10 Forecast Performance

The air quality multi-model forecast system was introduced and its 24-h forecast performance for meteorological parameters and PM10 daily mean concentration in Guangzhou during September 2010 was evaluated. The results show that although wind speed is overestimated, the model system can effectively predict variation in the meteorological parameters. All air quality models analyzed are shown to reasonably predict temporal and spatial variations of PM10 daily mean concentration in Guangzhou. In addition, all model forecasts satisfy the performance criteria such that mean fractional bias errors are less than or equal to ±60% and 75%, respectively, and several even reached performance goals of less than or equal to ±30% and 50%, respectively. However, all model forecasts overestimate PM10 daily mean concentration in suburban Guangzhou while underestimating the value in the urban region. An optimal model in this operational air quality forecast is not detected through model intercomparison. Variety in stations and statistical indicators may result in significant differences in forecast performance for the same model. Moreover, model ensemble based on arithmetic average does not reveal optimal forecast performance. Optimization of spatial distribution of the emission and usage of improved model ensemble forecast methods such as weighted average, neural network, and multiple regressions may improve forecast performance of the air quality multi-model system.