Forest aboveground biomass (AGB) is an important indicator for evaluating forest growth. Based on the 2D and 3D data generated by digital aerial photography (DAP), totally 41 point clouds height variables and 16 visible light vegetation indices were calculated respectively, and AGB estimation models were developed with single variable set and comprehensive variable set respectively by using six regression algorithms (random forest, RF;bagged tree, BT;support vector regression, SVR；Cubist;categorical boosting，CatBoost；extreme gradient boosting，XGBoost) to explore the contribution of different variables to the AGB estimation model. The results showed that the highest accuracy AGB prediction models for spectral and point cloud datasets were Cubist and XGBoost, with R2 of 0.5309 and 0.6395, respectively, and the highest accuracy model for the combined dataset was XGBoost, with R2 of 0.7601, and the XGBoost model had a higher stability of AGB estimation. The result also showed that the contribution of the six machine learning models mainly depended on the regression method considered, and the number of features chosen and the importance of the features to the model were not consistent across the models. DOM spectral features had a higher importance in the estimation of AGB. Overall, the combination of 2D and 3D data can effectively improve the accuracy of forest AGB estimation, and the RGB images acquired based on UAV tilt photography can realize the fast and nondestructive estimation of forest AGB.