Abstract:The visibility features are analyzed using automatic hourly visibility observations at Tianjin Port from 2009 to 2013. Based on the NCEP reanalysis data and Tianjin Port observations in the autumn and winter from 2006 to 2012, higher impact meteorological factors are given wich influence low visibility at port area through the correlation analysis. Three statistical models to different visibility samples have been established with artificial neurological network method, excluding dust and rain weather. Moreover, the statistical models are linked to output products in WRF and are applied in operational forecasting on visibility at Tianjin Port in autumn and winter through progressive screening method according to the three statistical models. So visibility forecast products of 72 h period are provided every day. The test results show that, for the hourly visibility forecast, the forecasting techniques score (TS) of BPTFP (back propagation three filter product) for less than 10 km visibility is from 10.5% to 35.4%, higher than the WRF. The TS is a comparable level for visibility less than 0.5 km, but WRF model forecast is a little better when it forecasts precipitation. For forecasting 0.5-1 km fog, the TS of WRF is less than 1%, and the BPTFP is from 14% to 21%. The average tests of daily minimum visibility show that the TS of BPTFP is 75%, which is 24% higher than WRF for the fog process of less than 1 km. Moreover, for the fog process from 1 to 10 km, the TS of BPTFP is 60% higher than WRF.