New methods and algorithms of dynamic-stochastic prediction of temperature and wind fields are considered as applied to assessment of mesoscale atmospheric pollution. Based on data of many-year field observations, the proposed algorithm is studied for its use in spatial prediction of mean temperature and zonal and meridional components of the wind velocity.