上半年工作总李结和下半年工作计划.ppt

工作报告 Section 12Air Quality Forecasting Tools Background Forecasting tools provide information to help guide the forecasting process. Forecasters use a variety of data products, information, tools, and experience to predict air quality. Forecasting tools are built upon an understanding of the processes that control air quality. Forecasting tools: Subjective Objective More forecasting tools = better results. Background Persistence Climatology Criteria Statistical Classification and Regression Tree (CART) Regression Neural networks Numerical modeling Phenomenological and experience Predictor variables Selecting Predictor Variables (1 of 3) Many methods require predictor variables. Meteorological Air Quality Before selecting particular variables it is important to understand the phenomena that affect pollutant concentrations in your region. The variables selected should capture the important phenomena that affect pollutant concentrations in the region. Selecting Predictor Variables (2 of 3) Select observed and forecasted variables. Predictor variables can consist of observed variables (e.g., yesterday’s ozone or PM2.5 concentration) and forecasted variables (e.g.,?tomorrow’s maximum temperature). Make sure that predictor variables are easily obtainable from reliable source(s) and can be forecast. Consider uncertainty in measurements, particularly measurements of PM. Selecting Predictor Variables (3 of 3) Begin with as many as 50 to 100 predictor variables. Use statistical analysis techniques to identify the most important variables. Cluster analysis is used to partition data into similar and dissimilar subsets. Unique (i.e., dissimilar) variables should be used to avoid redundancy. Correlation analysis is used to evaluate the relationship between the predictand (i.e.,?pollutant levels) and various predictor variables. Step-wise regression is an automatic procedure that allows the statistical software (SAS,?Statgraphics, Systat, etc.) to select the most importa