Machine learning-based prediction of stand biomass using multi-source environmental data in the Hulunbuir Mixed Forests, Inner Mongolia

• Machine learning models integrating stand attributes, species diversity, and environmental factors boost biomass estimation in mixed forests, with Random Forest achieving top accuracy (R² = 0.839). • Basal area and dominant diameter emerge as dominant predictors, explaining over 60% of biomass variation across models in diverse forest ecosystems. • Simpson diversity index and climate variables like annual heat-moisture index and precipitation significantly enhance predictions, revealing model-specific influences. Accurate estimation of forest aboveground biomass (AGB) is vital for understanding ecosystem productivity and carbon dynamics, especially in complex mixed forests. This study analyzed data from 99 natural mixed forest plots in Hulunbuir, Inner Mongolia, using machine learning models (random forest, support vector machine, and boosted regression tree) to predict AGB based on forest structure, species diversity, and environmental factors. Models explained 50%∼84% of AGB variation, with basal area and dominant diameter as key predictors. Species diversity is crucial for the accurate estimation of AGB. Climatic factors play a significant role in both Random Forest (RF) and boosted Regression Tree (BRT) models, while soil properties, particularly pH, were important in the support vector machine model. The study found that stand structure and the Simpson diversity index (SIM) are the primary determinants of biomass accumulation, indicating that forest management should focus on optimizing stand density, preserving structural diversity, and strengthening climate adaptability.