The methods of image fusion and transform, including PCA, wavelet based fusion, MNF and RBV transform etc. were executed to generate 37 feature bands on the basis of the ASTER original 9 bands. Coupling with observations of 48 poplar sample plots, traditional univariate regression models and regression tree models for average height, age and stem volume of poplar were cstablished respectively. After comparing the performance of models fit ring and grouud-truthing, it was found that regression tree models were superior to traditional univariate models in mapping spatial distribution of poplar stands’ parameters. Consequently, taking regression tree models to retrieve and map biophysical variables at a regional scale based on remote sensed data was more viable and reliable.