Objective The accurate estimation of biomass and carbon storage is essential for assessing forest carbon sinks. Constructing an optimized model based on environmental factors can provide general technical support. Chinese fir, due to its typical coupling with the site environment, serves as an exemplary case. Existing models often encounter challenges such as limited accuracy due to site-specific variations and inadequate multi-factor coupling modeling and verification. This study aims to construct a universal model, taking Chinese fir as a case study to perform modeling verification, thereby providing a reference for developing models for similar tree species.

Method In this paper, measured biomass data of Chinese fir across the country were collected through five categories of 23 multimodal data (topography, climate, soil, location and forest variables). Parametric models (basic model, passive variable model, and additive biomass parameter model) and non-parametric models (GA-BP, PSO-BP and BP models) were constructed.The SHAP interpreter was employed to analyze the contribution and interaction effects of each factor. Dummy variables were introduced to evaluate the performance of different models in terms of accuracy enhancement and usability.

Result The key findings were as follows: (1) The primary factors influencing Chinese fir biomass included four forest variables: diameter at breast height, tree height, forest age and stand density. Secondary factors were longitude and the maximum temperature during the growing season. Soil variables exhibit a relatively minor influence. Physical quantities had the strongest correlation and the highest contribution with variables such as diameter at breast height, tree height, forest age and stand density. The influence of longitude and the maximum temperature of the growing season were secondary, while the contribution of soil variables was relatively low. (2) Overall, except for leaf biomass, the PSO-BP model demonstrated a higher degree of accuracy improvement in biomass estimation. The SHAP interpreter results indicated that in non-parametric biomass models, the contribution of forest tree variables was the highest, and the response of the interaction factor was obvious. (3) The comparison results of verification indicators revealed that the introduction of dummy variables significantly improved the accuracy of the Chinese fir biomass model.

Conclusion Both parametric and non-parametric models established in this study can provide more accurate biomass simulation for Chinese fir trees. When combined with carbon content parameters, these models can accurately calculated the biomass and carbon storage of various components of Chinese fir trees (whole plant, above ground, below ground, trunk, crown, branches, leaves). The model and environmental variables in this paper have been embedded in an online cloud platform, and can be used online to calculate biomass.