关键词: 关键词：能见度预报, 动态逐步回归方法, 分级检验

Abstract: Based on meteorology observation data and ECMWF numerical prediction products in winter 2016, a dynamically multivariate stepwise regression model was established for visibility prediction in Beijing by selecting some possible impact factors (water vapor, dynamic and thermodynamic factors and so on) as independent variables. Visibility was divided into three levels, namely visibility lower than 1 km, from 1 to 10 km (low visibility levels) and higher than or equal to 10 km, and the relevant forecast effect evaluation was conducted from the views of regional average, spatial distribution and cases analysis, respectively. The model had certain stable forecast capability for visibility and its change trend. The threat score (TS) of forecasted visibility greater than and equal to 10 km was largest (64.2%) , followed by the visibility of 1-10 km (53.1%), and the TS of visibility lower than 1 km was smallest (51.3%). The forecasting was better in plain areas than mountain areas for the lower two grades visibility, showing the decline distribution from the northeast to the southwest. On the contrary, the forecast capability of visibility more  than 10 km increased from the northeast to the southwest. Besides, forecasts of three cases also confirmed that this model had a relatively good forecasting ability of persistent fog-haze episodes occurring in Beijing.

Key words: Key words: visibility forecast, dynamically stepwise regression, classification test