基于深度学习算法的杉木人工林单木冠幅回归模型

周钰祺; 孙玉军; 谢运鸿; 梁瑞婷; 颜雯

doi:10.12302/j.issn.1000-2006.202410029

周钰祺, 孙玉军, 谢运鸿, 梁瑞婷, 颜雯

南京林业大学学报（自然科学版） ›› 2026, Vol. 50 ›› Issue (1) : 231-240.

PDF(6852 KB)

国家林草科技领军期刊
中国精品科技期刊
中国高校百佳科技期刊
江苏省新闻出版政府奖期刊奖
RCCSE林学权威期刊（A+）
CSCD核心期刊
Scopus数据库收录期刊
中文核心期刊
SCD核心期刊

作者加群：102861116

微信公众号：南京林业大学学报

高级检索

PDF(6852 KB)

南京林业大学学报（自然科学版） ›› 2026, Vol. 50 ›› Issue (1) : 231-240. DOI: 10.12302/j.issn.1000-2006.202410029

研究论文

基于深度学习算法的杉木人工林单木冠幅回归模型

周钰祺 ¹ ,
孙玉军 ¹^,^* ,
谢运鸿 ¹ ,
梁瑞婷 ¹ ,
颜雯 ¹^,²

作者信息 +

Construction of Cunninghamia lanceolata tree crown width regression model based on deep learning algorithms

ZHOU Yuqi ¹ ,
SUN Yujun ¹^,^* ,
XIE Yunhong ¹ ,
LIANG Ruiting ¹ ,
YAN Wen ¹^,²

Author information +

文章历史 +

摘要

【目的】以福建省将乐国有林场为研究区,构建福建将乐地区杉木人工林深度学习冠幅模型,分析树木大小、林分结构、林分竞争对杉木冠幅预测的贡献。【方法】利用深度神经网络(deep neural network,DNN)构建6个NDD模型,并利用SHAP(shapley additive explanations)解释器解释各变量在模型中的特征重要性。构建一个广义冠幅模型,并与输入相同相关系数优选变量的DNN模型进行比较,验证DNN模型的可靠性。【结果】构建的DNN模型均无过拟合,十折交叉验证的均方根误差(RMSE)和平均误差(MAE)均稳定。由所有13个变量构成的DNN模型(DNN5)的决定系数(R²)达到0.60,表现最佳,但输入6个变量的DNN模型(DNN3和DNN4)的R²分别为0.58和0.57,更符合实际应用要求,DNN模型在处理多变量数据时具有良好的鲁棒性。经过相关系数优选变量构成的DNN6模型的R²(0.54)高于同等变量构成的广义冠幅模型的R²(0.46)。胸径的SHAP贡献值最大,此外林分结构中的林分断面积、林分密度指数以及林分竞争中麦金托什直径均匀度等级和基尼系数在特征排序中占据了重要的位置,引入这两类变量能较大幅度地提高冠幅模型的准确度。【结论】构建的DNN模型能够较好地预测研究区杉木冠幅,深度学习在杉木冠幅预测方面有很大的潜力。

Abstract

【Objective】 This study constructed a deep learning-based crown width model for Cunninghamia lanceolata plantations in the Jiangle region, aiming to analyze the contributions of tree size, site quality, stand structure, and competition to crown width prediction. 【Method】Six deep neural network (DNN) models were developed, and the shapley additive explanations (SHAP) interpreter was employed to analyze the feature importance of each variable. Additionally, a generalized crown width model was built and compared with the DNN model that utilized the same set of optimally selected variables based on correlation coefficients, in order to validate the reliability of the DNN approach. 【Result】The constructed DNN models exhibited no overfitting, with stable root mean square error (RMSE) and mean absolute error (MAE) values in 10-fold cross-validation. The DNN model incorporating all 13 variables (DNN5) achieved the highest performance, with an R² of 0.60. However, models with six input variables (DNN3 and DNN4) yielded R² values of 0.58 and 0.57, respectively, which better meet practical application needs. The DNN6 model, constructed using variables selected based on their correlation coefficients, achieved an R² of 0.54, outperforming the generalized crown width model (R²= 0.46) that used the same variables. The SHAP contribution value of DBH was the highest. Additionally, in the feature ranking, basal area (BA), stand density index (SDI), McIntosh evenness of diameter class (Emi), and Gini coefficient (GC) held important positions. The inclusion of both types of variables significantly improved prediction the accuracy of the crown width model. 【Conclusion】The results demonstrate that the constructed DNN model effectively predicted the crown width of C. lanceolata in the study area, indicating that deep learning holds significant potential for crown width modeling.

导出引用

周钰祺, 孙玉军, 谢运鸿, 等. 基于深度学习算法的杉木人工林单木冠幅回归模型[J]. 南京林业大学学报（自然科学版）. 2026, 50(1): 231-240 https://doi.org/10.12302/j.issn.1000-2006.202410029

ZHOU Yuqi, SUN Yujun, XIE Yunhong, et al. Construction of Cunninghamia lanceolata tree crown width regression model based on deep learning algorithms[J]. Journal of Nanjing Forestry University (Natural Sciences Edition）. 2026, 50(1): 231-240 https://doi.org/10.12302/j.issn.1000-2006.202410029

中图分类号： S757.2

参考文献

列表( 原文顺序 | 文献年度倒序 | 文中引用次数倒序 ) 可视化分析

[1]	周柏屹, 孙麟均, 吴鹏飞, 等. 杉木大径材培育研究进展[J]. 世界林业研究, 2024, 37(1):54-58. ZHOU B Y, SUN L J, WU P F, et al. Research progress in cultivation of large-diameter timber tree of Chinese fir[J]. World Forestry Research, 2024, 37(1):54-58. DOI:10.13348/j.cnki.sjlyyj.2023.0105.y. 本文引用 [1]

[2]	LIU C C, LI Y, YAN P, et al. How to improve the predictions of plant functional traits on ecosystem functioning?[J]. Frontiers in Plant Science, 2021, 12:622260. DOI:10.3389/fpls.2021.622260. 本文引用 [2]

[3]	ATKINS J W, SHIKLOMANOV A, MATHES K C, et al. Effects of forest structural and compositional change on forest microclimates across a gradient of disturbance severity[J]. Agricultural and Forest Meteorology, 2023, 339:109566. DOI:10.1016/j.agrformet.2023.109566. 本文引用 [1]

[4]	AHMED S, HILMERS T, UHL E, et al. Neighborhood competition modulates the link between crown structure and tree ring variability in monospecific and mixed forest stands[J]. Forest Ecology and Management, 2024, 560:121839. DOI:10.1016/j.foreco.2024.121839. 本文引用 [1]

[5]	MARSHALL H D, MURPHY G E, BOSTON K. Evaluation of the economic impacts of length and diameter measurement error on mechanical harvesters and processors operating in pine stands[J]. Canadian Journal of Forest Research, 2006, 36(7):1661-1673. DOI:10.1139/x06-064. 本文引用 [1]

[6]

雷相东, 张则路, 陈晓光. 长白落叶松等几个树种冠幅预测模型的研究[J]. 北京林业大学学报, 2006, 28(6):75-79.

LEI

X D

, ZHANG

Z L

, CHEN

X G

. Crown-width prediction models for several tree species including Larix olgensis in northeastern China[J]. Journal of Beijing Forestry University, 2006, 28(6):75-79. DOI:10.3321/j.issn:1000-1522.2006.06.013.

本文引用 [1]

[7]	XU H, SUN Y J, WANG X J, et al. Linear mixed-effects models to describe individual tree crown width for China-fir in Fujian Province,southeast China[J]. PLoS One, 2015, 10(4):e0122257. DOI:10.1371/journal.pone.0122257. 本文引用 [1]

[8]	LEI Y, FU L, AFFLECK D L R, et al. Additivity of nonlinear tree crown width models: aggregated and disaggregated model structures using nonlinear simultaneous equations[J]. Forest Ecology and Management, 2018, 427:372-382. DOI:10.1016/j.foreco.2018.06.013. 本文引用 [1]

[9]	邱思玉, 孙玉军. 长白落叶松人工林单木冠幅模型[J]. 东北林业大学学报, 2021, 49(2):49-53. QIU S Y, SUN Y J. Individual tree crown width prediction models for Larix olgensis plantation[J]. Journal of Northeast Forestry University, 2021, 49(2):49-53. DOI:10.13759/j.cnki.dlxb.2021.02.009. 本文引用 [3]

[10]

吕乐, 董利虎, 李凤日. 黑龙江省东部地区天然椴树单木冠幅预测模型[J]. 东北林业大学学报, 2019, 47(7):37-42.

LÜ

, DONG

L H

, LI

F R

. Individual tree crown width prediction models for natural Tilia tuan in eastern Heilongjiang Province[J]. Journal of Northeast Forestry University, 2019, 47(7):37-42. DOI:10.13759/j.cnki.dlxb.2019.07.007.

本文引用 [1]

[11]	钟思琪, 宁金魁, 黄锦程, 等. 基于混合效应的杉木人工林冠幅模型[J]. 森林与环境学报, 2024, 44(2):127-135. ZHONG S Q, NING J K, HUANG J C, et al. Crown width model of Chinese fir plantation based on mixed effect[J]. Journal of Forest and Environment, 2024, 44(2):127-135. DOI:10.13324/j.cnki.jfcf.2024.02.003. 本文引用 [1]

[12]	AMANULLAH M A, HABEEB R A A, NASARUDDIN F H, et al. Deep learning and big data technologies for IoT security[J]. Computer Communications, 2020, 151:495-517. DOI:10.1016/j.comcom.2020.01.016. 本文引用 [1]

[13]	LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553):436-444. DOI:10.1038/nature14539. 本文引用 [1]

[14]	AGGARWAL, CHARU C. Neural networks and deep learning[M]. Cham: Springer International Publishing, 2018. DOI:10.1007/978-3-319-94463-0. 本文引用 [1]

[15]	HEATON J. Ian goodfellow,yoshua bengio,and aaron courville:deep learning[J]. Genetic Programming and Evolvable Machines, 2018, 19(1):305-307. DOI:10.1007/s10710-017-9314-z. 本文引用 [1]

[16]	ERCANLI I. Innovative deep learning artificial intelligence applications for predicting relationships between individual tree height and diameter at breast height[J]. Forest Ecosystems, 2020, 7(1):12. DOI:10.1186/s40663-020-00226-3. 本文引用 [2]

[17]	梁瑞婷, 孙玉军, 李芸. 深度学习和传统方法模拟杉木树高-胸径模型比较[J]. 林业科学研究, 2021, 34(6):65-72. LIANG R T, SUN Y J, LI Y. Comparison of deep learning and traditional models to simulate the height-DBH relationship of Chinese fir[J]. Forest Research, 2021, 34(6):65-72. DOI:10.13275/j.cnki.lykxyj.2021.06.008. 本文引用 [2]

[18]	QIN Y, WU B, LEI X, et al. Prediction of tree crown width in natural mixed forests using deep learning algorithm[J]. Forest Ecosystems, 2023, 10:100109. DOI:10.1016/j.fecs.2023.100109. 本文引用 [3]

[19]	符利勇, 孙华. 基于混合效应模型的杉木单木冠幅预测模型[J]. 林业科学, 2013, 49(8):65-74. FU L Y, SUN H. Individual crown diameter prediction for Cunninghamia lanceolata forests based on mixed effects models[J]. Scientia Silvae Sinicae, 2013, 49(8):65-74. DOI:10.11707/j.1001-7488.20130810. 本文引用 [2]

[20]	ALI A. Forest stand structure and functioning:current knowledge and future challenges[J]. Ecological Indicators, 2019, 98:665-677. DOI:10.1016/j.ecolind.2018.11.017. 本文引用 [1]

[21]	VOSPERNIK S. Basal area increment models accounting for climate and mixture for Austrian tree species[J]. Forest Ecology and Management, 2021, 480:118725. DOI:10.1016/j.foreco.2020.118725. 本文引用 [1]

[22]	OWEN H J F, FLYNN W R M, LINES E R. Competitive drivers of interspecific deviations of crown morphology from theoretical predictions measured with Terrestrial Laser Scanning[J]. Journal of Ecology, 2021, 109(7):2612-2628. DOI:10.1111/1365-2745.13670. 本文引用 [1]

[23]	LI Q C, LIU Z L, JIN G Z. Impacts of stand density on tree crown structure and biomass:a global meta-analysis[J]. Agricultural and Forest Meteorology, 2022, 326:109181. DOI:10.1016/j.agrformet.2022.109181. 本文引用 [1]

[24]	MOOLAYIL J. Learn keras for deep neural networks: a fast-track approach to modern deep learning with python[M]. Berkeley CA: Apress, 2019. 本文引用 [1]

[25]	RUDER S. An overview of gradient descent optimization algorithms[EB/OL]. arXiv, 2016:1609.04747. https://arxiv.org/abs/1609.04747. https://arxiv.org/abs/1609.04747 本文引用 [1]

[26]	SRIVASTAVA N, HINTON G, KRIZHEVSKY A, et al. Dropout:a simple way to prevent neural networks from overfitting[J]. Journal of Machine Learning Research, 2014, 15(1):1929-1958. DOI:10.5555/2627435.2670313. 本文引用 [1]

[27]	HOSAIN M T, JIM J R, MRIDHA M F, et al. Explainable AIapproaches in deep learning:advancements,applications and challenges[J]. Computers and Electrical Engineering, 2024, 117:109246. DOI:10.1016/j.compeleceng.2024.109246. 本文引用 [2]

[28]	ANTWARG L, MILLER R M, SHAPIRA B, et al. Explaining anomalies detected by autoencoders using Shapley Additive Explanations[J]. Expert Systems with Applications, 2021, 186:115736. DOI:10.1016/j.eswa.2021.115736. 本文引用 [1]

[29]	JADERBERG M, DALIBARD V, OSINDERO S, et al. Population based training of neural networks[EB/OL]. arXiv, 2017:1711.09846. https://arxiv.org/abs/1711.09846. https://arxiv.org/abs/1711.09846 本文引用 [1]

[30]	ALI A., YAN E R, CHEN H Y H, et al. Stand structural diversity rather than species diversity enhances aboveground carbon storage in secondary subtropical forests in Eastern China[J]. Biogeosciences, 2016, 13(16):4627-4635. DOI:10.5194/bg-13-4627-2016. 本文引用 [1]

[31]	SCHRÖTER M, HÄRDTLE W, VON OHEIMB G. Crown plasticity and neighborhood interactions of European beech (Fagus sylvatica L.) in an old-growth forest[J]. European Journal of Forest Research, 2012, 131(3):787-798. DOI:10.1007/s10342-011-0552-y. 本文引用 [1]

[32]	BAYAT M, BETTINGER P, HEIDARI S, et al. Estimation of tree heights in an uneven-aged,mixed forest in northern Iran using artificial intelligence and empirical models[J]. Forests, 2020, 11(3):324. DOI:10.3390/f11030324. 本文引用 [1]