基于气溶胶三维变分同化天津PM2.5数值预报研究

基于气溶胶三维变分同化技术，建立天津空气质量数值模式气溶胶同化模块，通过天津地区两次重污染过程同化模拟敏感性试验，分析了观测资料范围对同化结果的影响，并结合一个月的滚动预报试验，分析了气溶胶同化对天津地区PM_2.5数值预报效果的影响，以期为提升天津空气质量预报能力提供支撑.结果表明：气溶胶同化各控制变量背景误差水平相关系数的衰减尺度约50km，垂直方向上400m高度与模式底层的相关系数衰减至0.6左右；观测资料范围对同化结果影响显著，仅采用天津地区观测数据进行同化，对重污染过程期间天津地区PM_2.5浓度模拟的影响时效约12h，采用模拟区域内所有观测数据进行同化影响时效可持续24h以上，且模拟效果更优；采用三维变分同化技术，实现地面PM_2.5观测资料同化，天津地区PM_2.5数值预报效果显著提升，预报值和实况值之间的相关系数由0.74增加到0.87，均方根误差由32.3μg/m³减小为22.4μg/m³，平均相对误差由39.9%减小为27.1%；同化对模式初始时刻的改进效果最明显，随时间同化效果衰减，14h内改进效果最佳，对24h PM_2.5浓度预报也有明显改进.

Impact of aerosol data assimilation with 3-DVAR method on PM2.5 forecast over Tianjin

Based on a 3-DVAR assimilation method of aerosol, this paper developed aerosol assimilation module applied to air quality model in Tianjin, analyzed the influence of observation area on PM_2.5 forecasts through data assimilation and prediction experiments during two heavy pollution episodes in Tianjin, and then analyzed the impact of aerosol data assimilation on PM_2.5 forecast in Tianjin by conducting experiments with data assimilation over a month, in order to provide support for improving the capability of air quality forecast in Tianjin. The results showed that the attenuation scale of horizontal correlation for background errors was about 50km, and the correlation coefficient between the lowest level of model and the height of 400m decreased to about 0.6. The observation area had a significant impact on the assimilation results. During heavy pollution episodes, the improvement of PM_2.5 forecast continued about 12 hours when only observations in Tianjin were used for data assimilation, while the improvement continued above 24 hours when all observations within the model domain were used. 3-DVAR assimilation of surface PM_2.5 observations significantly improved PM_2.5 numerical forecast over Tianjin. The correlation coefficient between observation and forecast increased from 0.74 to 0.87, the root mean square error decreased from 32.3μg/m³ to 22.4μg/m³, and the mean relative error decreased from 39.9% to 27.1%. The improvement was the most significant at the initial time, and decreased with forecast time increasing. PM_2.5 concentration forecasts within 24 hours were improved obviously which were much better within 14hours.