关键词: 杉木, 大径材出材量, 贝叶斯网络, 近自然改造

Abstract:

Objective: With Bayesian network model, a machine learning method based on probabilistic inference, this study aims to analyze the effects of factors such as growth traits, soil nutrients, and understory vegetation diversity on the yield of large-diameter timber of Chinese fir under close-to-nature transformation from Chinese fir to Phoebe bournei, so as to provide theoretical support for the optimal management of Chinese fir stand and the cultivation of large-diameter timber. Method: The Chinese fir plantations planted in 2004 in Xishan State-owned Forest Farm in Linwu County, Hunan Province were targeted, and in 2015, the plantations were thined and, then interplanted with P. bournei. Key variables, including retained density of Chinese fir, DBH, dominant height, crown width, soil nutrients (total nitrogen and total phosphorus), and understory vegetation diversity, were selected. By integrating empirical data with expert knowledge, a mechanism model for the influence of Chinese fir large-diameter timber yield was constructed based on the Bayesian network model, and Expectation–Maximization (EM) algorithm was used to learn model, revealing the effects and interactions of different factors on the large-diameter timber yield. Result: The yield of large-diameter timber of Chinese fir was comprehensively affected by factors such as retained density of Chinese fir, crown width, DBH, dominant height, soil nutrients and understory vegetation diversity. The growth of DBH and the expansion of crown width were the key factors affecting the yield of large-diameter timber of Chinese fir (43.0%), and their influence on the yield was greater than that of the dominant height (2.07%). Suitable retained density of Chinese fir was able to promote the growth of DBH and crown width, so as to improve the yield of large diameter timber. Total phosphorus, as an important nutrient element in soil, had a positive effect on the growth of Chinese fir (1.40%), while the diversity of understory vegetation had little effect on the yield of large-diameter timber, which mainly affected the growth of Chinese fir through indirect ways. The Bayesian network model showed high prediction accuracy (88.9%, AUC=0.916 7) and good interpretability in capturing the complex relationship between multiple factors and predicting the large-diameter timber yield of Chinese fir. Conclusion: Based on the Bayesian network model, this study reveals the influence mechanism of large-diameter timber yield of Chinese fir under close-to-nature transformation, and proposes that Chinese fir plantations management should focus on the growth of DBH and crown width, optimize stand density and soil phosphorus supply, so as to promote the sustainable improvement of large-diameter timber yield. As a machine learning approach, the Bayesian Network model shows high prediction accuracy and interpretability in revealing the complex relationships among multiple factors such as Chinese fir growth conditions, soil nutrients, and understory vegetation diversity, etc. This study provides a scientific basis for the efficient management of Chinese fir plantations and improvement of large-diameter timber yield, and an efficient and interpretable tool for forest management decision-making.

Key words: Chinese fir, large-diameter timber yield, Bayesian network, close-to-nature silviculture