Understanding the contribution of structural diversity to stand biomass for carbon management of mixed forests using machine learning algorithms

Abstract

The structural properties of mixed stands and their effects on forest carbon sink function have attracted the attention of forest managers. Understanding the comprehensive effects of stand factors and structure on forest biomass is critical for better carbon management. However, data and information on biomass variability and its relationships to stand structural features are still insufficient. The purpose of this study was to develop models linking stand-level biomass with stand factors and structure, and to quantify the effects of each variable on stand biomass in natural mixed forests, especially stand structure. Four machine learning (ML) algorithms named Support Vector Machine (SVM), Artificial Neural Network (ANN), Random Forest (RF) and Boosted Regression Trees (BRT) were adapted. The results showed that SVM and ANN outperformed RF and BRT algorithms in stand biomass estimation. ANN with tree size diversity as the input had the highest accuracy (R2=0.9255±0.0421) among the models. Furthermore, structural diversity was a reliable predictor of mixed stand biomass estimation which is superior to the stand average height traditionally used. The positive correlation between stand biomass and structural diversity suggested that the complex stand structure promoted the accumulation of stand biomass. Thus, our study offered a ML protocol for predicting stand biomass of natural coniferous-broadleaved mixed forests, and suggested that using comprehensive management measures such as properly promoting tree differentiation can help forest managers enhance ecosystem carbon.