On the basis of multi-spectral multi-angular radiative polarization measurements, a prediction method for atmospheric polarization patterns is investigated. Firstly, the principle of an Aviation Multi-angle Polarimetric Radiometer (AMPR) and the regression algorithm of a Support Vector Machine (SVM) are introduced. Then, according to Vector Radiative Transfer Equation (VRTE), it points out that the atmospheric polarization pattern is primarily dependent on view geometry and surface features when atmosphere condition is invariant. Meanwhile, it introduces the relationship between view geometry and platform attitude and the expressing form of surface features. Finally, how to use the regression algorithm of SVM to predict the detected polarization degree of the APMR and to validate the application process is introduced in the consideration of the surface features and platform attitudes. Furthermore, the predicted degrees of polarization is also compared to that of real measurements. The results indicate that the error of polarization degree predicted is less than 1%. It concludes that the serious factor to affect the model accuracy is not the change of attitude but the variation of underlying surface property caused by the changed attitudes.