Background The optimization of biomass and carbon storage models is not only a key tool for the transformation of soil and water conservation from empirical governance to scientific decision-making, but also a core technical support for realizing the systematic governance of "mountains, rivers, forests, fields, lakes, grasslands and sands" and coping with climate change. Climate change affects not only temperature, precipitation and solar radiation, but also the carbon cycle of terrestrial ecosystems. Taking Pinus massoniana in China as a case study, the study has established a precise and reliable estimation model system for biomass and carbon storage. By integrating multiple modeling approaches, the study aims to identify the optimal model to facilitate more accurate assessments of bioenergy potential and ecosystem carbon sequestration capacity.

Methods This study selected P. massoniana, a major conifer species in China, as the research object. Based on biomass data from 366 sample trees, parametric models (including the basic model, dummy variable model, and additive model) and non-parametric models (primarily employing artificial intelligence techniques, machine learning and neural network training, BP neural network, GA-BP neural network model, and PSO-BP neural network), were utilized for model construction. The study systematically evaluated the advantages and disadvantages of parametric and non-parametric models to identify the most suitable estimation model.

Results 1) The dominant factors of biomass of P. massoniana were DBH, tree height, stand age and stand density, followed by the maximum, minimum and average rainfall of growing season and slope direction, and the soil had a low influence. 2) On the whole, the model based on PSO-BP had the best performance in biomass estimation. Except for the lower precision of the leaf model, the R² and Adj-R² of the other models were all greater than 0.93, and the RSE and RMSE were both less than 21.66, and there were very few abnormal cases of negative values and extreme high underestimation. In the biomass non-parametric model, the contribution of forest variables was the highest, and the response of interaction factors was obvious. 3) The additive construction of the basic parameter model of P. massoniana not only maintained a good accuracy but also ensured the logical relationship that the total biomass is equal to the sum of each component. Therefore, it is recommended to choose the additive model without introducing dummy variables as the optimal biomass parameter model of the whole plant and each organ of P. massoniana.

Conclusions This study developed a model based on artificial intelligence and multi-model integration, incorporating 23 factors across five categories to simulate biomass and carbon storage for Masson pine in China. The optimized model demonstrated strong performance, with both accuracy and stability surpassing those of traditional models.