Abstract: This study utilizes Sentinel–2A remote sensing data to optimize the penalty parameter(C) and the kernel function parameter(γ) of the Support Vector Regression(SVR) model using the Particle Swarm Optimization(PSO) algorithm, aiming to enhance the accuracy of Aboveground Biomass(AGB) inversion. During the variable selection process, a hierarchical variable selection method is employed, where variables are ranked by the absolute value of Pearson's correlation coefficient. Different variable combinations are used to construct both SVR and PSO–SVR models to investigate the impact of feature selection on model performance. The results show that, after evaluating the model's generalization ability using 5-fold cross-validation, the PSO–SVR model with the top 10% of selected hierarchical variables performs the best, achieving an R² of 0.989 and RMSE of 4.623 t/hm², significantly outperforming the traditional SVR model(R² = 0.813, RMSE = 18.697 t/hm²). As the number of variables increases, the model's accuracy decreases. When all variables are included in the modeling process, the R² of PSO–SVR drops to 0.311, and the RMSE increases to 35.831 t/hm², indicating that the introduction of redundant variables weakens the model’s predictive ability. In summary, the effectiveness of PSO optimization for SVR parameters is validated, and the combination of reasonable variable selection with optimization algorithms significantly improves AGB estimation accuracy.