[Purposes] The study aims to establish vegetation height model by aerial photogrammetry. [Methods] Based on digital orthophoto map and digital surface model, spectral and geometric feature factors were extracted for vegetation height modeling. The correlation between vegetation height and feature factor was analyzed by correlation index, the feature factors were selected. The gradient boosting decision tree algorithm was adopted to establish the vegetation height prediction model, and the accuracy of the model was improved through parameter optimization. [Findings] The model accuracy is about 2.000 m under the default parameter. By optimizing the parameters, the model accuracy reaches 0.900 m. Furthermore, the model accuracy enhance to 0.840 m, resulted from excluding some feature factors. By compared with the support vector machine algorithm, the accuracy of the vegetation height model has been increased from 0.893 m to 0.758 m, and the running time has been reduced from 70 minutes to 10 minutes. [Conclusions] The accuracy of the vegetation height model can reach to sub meter, when the error of the original modeling data is neglected, and the accuracy of the model remains stable in many experiments.