This study investigates the improvement of forest above-ground biomass (AGB) estimation using ensemble machine learning methods. The authors compared the performance of four well-known machine learning models—CatBoost, LightGBM, random forest (RF), and XGBoost—and analyzed the impact of forest types, independent variables, and spatial autocorrelation on AGB estimation accuracy. The study used eight scenarios based on two study regions, two variable types, and two validation strategies. The results showed that no single model outperformed the others in all scenarios. RF demonstrated superior performance in scenarios 5, 6, and 7, while CatBoost showed the best performance in the remaining scenarios. A hybrid model combining the strengths of these individual models was proposed and consistently performed best across all scenarios, despite some uncertainties. The ensemble strategy significantly improved estimation accuracy and stability, effectively addressing the challenge of model selection in forest AGB forecasting. The study also explored the impact of spatial autocorrelation on AGB models and found that spatial cross-validation was more effective than random cross-validation in correcting model bias. The findings provide valuable insights into improving forest AGB estimation accuracy and model selection for sustainable forest management and carbon cycle tracking.This study investigates the improvement of forest above-ground biomass (AGB) estimation using ensemble machine learning methods. The authors compared the performance of four well-known machine learning models—CatBoost, LightGBM, random forest (RF), and XGBoost—and analyzed the impact of forest types, independent variables, and spatial autocorrelation on AGB estimation accuracy. The study used eight scenarios based on two study regions, two variable types, and two validation strategies. The results showed that no single model outperformed the others in all scenarios. RF demonstrated superior performance in scenarios 5, 6, and 7, while CatBoost showed the best performance in the remaining scenarios. A hybrid model combining the strengths of these individual models was proposed and consistently performed best across all scenarios, despite some uncertainties. The ensemble strategy significantly improved estimation accuracy and stability, effectively addressing the challenge of model selection in forest AGB forecasting. The study also explored the impact of spatial autocorrelation on AGB models and found that spatial cross-validation was more effective than random cross-validation in correcting model bias. The findings provide valuable insights into improving forest AGB estimation accuracy and model selection for sustainable forest management and carbon cycle tracking.