空气污染潜势-统计结合预报模型的建立及应用

建立了一个空气污染潜势预报和统计预报相结合的模型,该模型以特征气象因子和大气扩散清除因子为基础,并考虑不同因子的权重,定义空气污染潜势指数APPI.所考虑的因子包括:地面风速、混合层高度、混合层内平均风速、风向日变化、稳定度级数、垂直扩散系数、SO2干沉降速率、NO2干沉降速率、PM10干沉降速率、降水时长、地面天气形势.进一步利用统计方法建立空气污染指数API与APPI之间的关系.利用南京地区2009~2010年气象资料计算APPI,通过3项式拟合得到API与APPI的统计方程.结果表明,拟合得到的API与实际API相关系数为0.67,具有显著的相关性,且等级准确率为76.7%.进一步利用2011年1~12月中尺度气象模式WRF预报的气象场开展实况预报.研究表明,24h预报、48h预报、回顾预报的逐月等级正确率分别为44.4%~87.5%,46.4%~100%和63.0%~80.0%,年均等级正确率为60.6%,62.4%.和73.1%.若定义预报API与实际API相差±20以内为正确,则24h预报、48h预报、回顾预报的正确率分别为58.1%, 59.4%和63.8%.在IBM x3500并行集群服务器上计算,48h预报需要机时3h.可见,该模型具有较好的预报性能, 相对数值模型计算效率很高.

An air pollution potential forecast model combined with statistical method and its application

An air quality potential forecast model combined with statistical method was established based on typical meteorological factors and atmospheric diffusion and scavenging factors.Each factor had a weight according to their contributions to the air pollution potential.An air pollution potential index was defined.These factors included surface wind speed,mixing height,average wind speed under mixing height,diurnal variation of wind direction,stability,vertical diffusion coefficient,dry deposition velocity of sulfur dioxide,nitrogen dioxide and particulate matter with particle size below 10 microns,precipitation duration and ground synoptic situation.Equation of the relationship between API and APPI was constructed using statistical methods.Air pollution potential indexes were calculated using meteorological data of Nanjing during the year 2009~2010 and the statistical equation was established as trinomial form.The correlation coefficient of the fitted API and the actual API was 0.67 with high significance.The accuracy rate for air quality level prediction was 76.7%.Meteorological data from mesoscale forecast model WRF during the whole twelve months in 2011 were contemporarily calculated to support air quality forecast.The monthly accuracy rates of 24-hour,48-hour and review forecast were 44.4%~87.5%,46.4%~100% and 63.0%~80.0%,respectively.The annual average accuracy rates were 60.6%,62.4% and 73.1%,respectively.The accuracy rates considering the difference of predicted and observed API within ±20 were 58.1%,59.4% and 63.8%,respectively.On the IBM x3500 server cluster,3 hours were required for 48-hour forecast.The results showed that this model had good performance and high efficiency compared to numerical prediction.