Previously, the empirical models are used for seasonal pollen dispersion forecasts. Such empirical models predict volume of pollen statistically, however, they cannot reflect wind field, which has great impact on the pollen dispersion, of each year. For example, in 2009 season (February – April), westerly wind dominated around the Kanto plain in Japan, but in 2010 season easterly wind dominated on the contrary. In this study, with using WRF and the numerical pollen dispersion model based on 3-months seasonal ensemble numerical model issued by Japan Meteorological Agency (JMA), we calculated seasonal pollen concentration of Japanese cedar and cypress (we use From 31 members of JMA 3-months ensemble model, we chose 1 member which has the strongest correlation with normals of temperature, humidity, wind speed and wind direction. We run our models, with using the member as initial and boundary condition of WRF, until 2160 hours ahead. In our model, we use meteorological parameters not only for pollen advection and diffusion but for estimation of flowering process and of pollen emissivity. As a result, we dramatically improved long term pollen forecast which reflects actual meteorological (mainly wind) situation, and the model represented feature of long distance transformation of the pollen in each year.