在 WRF 模式输出的基础上，结合卡尔曼滤波(KALMAN)、偏最小二乘回归(PLS)和辨识理论实时迭代统计方法(RTIM)组成3种模式输出统计预报(MOS)，并运用这3种MOS模式分别建立多模式平均、递归正权综合和多元线性回归集成模式，对上海市2014年12月2日~31日(PM2.5轻度污染天气过程)以及2015年12月15日~2016年1月13日(PM2.5重污染天气过程)进行了1d、2d和3d的PM2.5日均浓度试预报.结果表明：相比于3种单一的MOS模式，集成模式通过获取更为准确的信息而减少了系统误差，这不仅可以提升对污染天气过程的预报能力，且有可能降低污染过程中决策失败的风险.通过对 PM2.5轻度污染天气和重污染天气过程预报的比较分析，多元线性回归集成模式整体预报显示出更高的精度和稳定性.统计释用方法综合集成模式对于 PM2.5预报显示出良好的性能，可为业务化预报模型的选择提供可借鉴的参考.%The comprehensive integration approaches for statistical interpretation were employed to predict PM2.5 concentration in Shanghai. Three kinds of model output statistics (MOS) models were built up by combining WRF model output with kalman filtering (KALMAN), partial least squares regression (PLS) and real-time iterative model (RTIM) respectively. And then three types of integrated model, based on MOS model above Multi-model average integration, recursive positive weight synthesis and multivariate linear regression integration, were separately apply to three days prediction for daily average concentration of PM 2.5 from Dec. 2 to 31, 2014 (light pollution process) and Dec. 15, 2015 to Jan. 13, 2016 (heavy pollution process). The prediction ability of pollution weather process for these integration models was improved by providing reasonable information and reducing the systematic errors in comparison with a single MOS models, which reduced the risk of decision-making in the process of pollution. The multivariate linear regression integration model presented higher precision and stability by comparativeanalysis of light and heavy pollution prediction processes. In all, the comprehensive integration approaches for statistical interpretation have great potential to be applied to regional air pollution prediction in operational model.
展开▼
