Height-diameter Model of Cunninghamia lanceolata Based on Deep Neural Network

doi:10.13466/j.cnki.lczyyj.2024.01.011

Abstract

Abstract:

By using the deep Neural Network (DNN) model to establish the tree height-diameter model of Cunninghamia lanceolata, we are seeking a more efficient method for predicting the tree height of Cunninghamia lanceolata.The diameter at breast height and tree height data of Cunninghamia lanceolata in 49 plots of state-owned forest farm in Qingzhen City,Guizhou Province were studied,and divided into different proportions of training set and test set data.This consists of training set data(20%,30%,40%,50%,60%,70%,80%,respectively)and test set data(80%,70%,60%,50%,40%,30%,20%,respectively).DNN was used to build a tree height-diameter model,and the model was compared with 11 traditional basic models.Select the model with the best predictive performance by comparing the results of R²,RMSE and MAE.Adding the ratio of diameter at breast height to average diameter at breast height(DDH)of dominant trees based on the optimal model to improve the prediction accuracy.When the training set proportion of the DNN model is 20%,the prediction accuracy of the tree height-diameter model can reach more than 0.89 after adding DDH factor.DNN was used to build a height-diameter model to predict the height of Cunninghamia lanceolata,and adding DDH factor could improve the prediction accuracy of Cunninghamia lanceolata height with a smaller dataset.

Key words: Cunninghamia lanceolata, DNN, DDH, height-diameter model

CLC Number:

S758

WANG Guilin, TAN Wei, CHEN Botao. Height-diameter Model of Cunninghamia lanceolata Based on Deep Neural Network[J]. Forest and Grassland Resources Research, 2024, (1): 82-87.

Figures/Tables 6

Tab.1

Fig.1

Tab.2

Tab.3

Fig.2

Fig.3

References 23

[1]	郭卫红, 郑庆荣, 胡砚秋, 等. 山西五台山主要针叶树种树高-胸径曲线模型研究[J]. 湖南林业科技, 2022, 49(6):72-77.
[2]	王景弟, 杨蕊, 田育新. 枫香和杉木树高-胸径模型的拟合与评价[J]. 湖南林业科技, 2021, 48(4):64-67.
[3]	刘金义. 辽东山区人工红松枝下高模型的研建[J]. 林业科技通讯, 2022(8):95-98.
[4]	李晓晶. 杂种落叶松树高-胸径模型的研究[J]. 林业科技情报, 2020, 52(4):28-30.
[5]	郭嘉, 孙帅超, 田相林, 等. 引入优势木树高建立的秦岭林区松栎林树高-胸径模型[J]. 东北林业大学学报, 2019, 47(11):66-72.
[6]	Cheng Deng, Zhang Yiyi, Lu Jiangyan, et al. Thinning effects on the tree height-diameter allometry of Masson Pine(Pinus massoniana Lamb.)[J]. Multidisciplinary Digital Publishing Institute, 2019, 10(12):1129.
[7]	Mensah S, Pienaar O L, Kunneke A. Height-Diameter allometry in South Africa's indigenous high forests:Assessing generic models performance and function forms[J]. For.Ecol.Manag. 2018, 410:1-11.
[8]	Khamyong N, Wangpakapattanawong P, Chairuangsri S, et al. Tree species composition and height-diameter allometry of three forest types in Northern Thailand[J]. CMU J.Nat. Sci.2018, 17(4):289-306.
[9]	董云飞, 孙玉军, 王轶夫, 等. 基于BP神经网络的杉木标准树高曲线[J]. 东北林业大学学报, 2014, 42(7):154-156.
[10]	卯光宪, 谭伟, 柴宗政, 等. 基于BP神经网络的马尾松人工林胸径-树高模型预测[J]. 浙江农林大学学报, 2020, 37(4):752-760.
[11]	Chen Xinxin, Jiang Kang, Zhu Yunshi, et al. Individual tree crown segmentation directly from UAV-Borne LiDAR data using the point net of deep learning[J]. Forests 2021, 12(2):131. doi: 10.3390/f12020131
[12]	Ercanli I. Innovative DNN artificial intelligence applications for predicting relationships between individual tree height and diameter at breast height[J]. Forest Ecosystems, 2020, 7(1):7-12. doi: 10.1186/s40663-020-0216-9
[13]	Maler S, Miglietta F, Gobakken T, et al. Prediction of stem diameter and biomass at individual tree crown level with advanced machine learning techniques[J]. iForest-Biogeosciences and Forestry, 2019(3).
[14]	Loubota P G J, Bocko Y E, Mavoungou A Y, et al. Height-diameter allometry in African monodominant forest close to mixed forest[J]. Journal of Tropical Ecology, 2021(2):37.
[15]	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.
[16]	Dantas D, Calegario N, Fausto W A J, et al. Multilevel nonlinear mixed-effects model and machine learning for predicting the volume of eucalyptus spp.trees[J]. Cerne, 2020, 26(1),48-57. doi: 10.1590/01047760202026012668
[17]	Aguilar F J, Nemmaoui A, Aguilar M A, et al. Building tree allometry relationships based on TLS point clouds and machine learning regression[J]. Applied Sciences, 2021, 11(21):10139. doi: 10.3390/app112110139
[18]	Casas G G, Gonzales D G E, Villanueva J R B, et al. Configuration of the deep neural network hyperparameters for the hypsometric modeling of the Guazuma crinita Mart in the Peruvian Amazon[J]. Forests 2022, 13(5):697. doi: 10.3390/f13050697
[19]	Wang Jiamin, Chen Xinxin, Cao Lin, et al. Individual rubber tree segmentation based on ground-based LiDAR data and faster R-CNN of deep learning[J]. Forests, 2019, 10(9):793. doi: 10.3390/f10090793
[20]	贺梦莹. 长白落叶松-水曲柳混交林冠幅与冠长预测模型的研究[D]. 哈尔滨: 东北林业大学, 2020.
[21]	曹晓梅, 苗铮, 郝元朔, 等. 基于树种分类的帽儿山阔叶混交林树高-胸径模型[J/OL]. 应用生态学报:1-17(2024-01-12)[2024-02-02].http://doi.org/10.13287/j.1001-9332.202402.016.
[22]	Ercanli I. Artificial intelligence with DNN algorithms to model relationships between total tree height and diameter at breast height[J]. Forest Systems, 2020, 29(2):103.
[23]	Corte A, Souza D V, Rex F E, et al. Forest inventory with high-density UAV-Lidar:Machine learning approaches for predicting individual tree attributes[J]. Computers and Electronics in Agriculture, 2020, 179:105815. doi: 10.1016/j.compag.2020.105815

变量	训练集(n=3 795)
变量	平均值	最小值	最大值	标准差	变异系数
胸径/cm	17.68	5.1	49.4	6.91	0.39
树高/m	12.81	2.7	22.2	3.38	0.26

二八比例(训练集占比为20%、测试集占比80%)
方法	Power模型 (M1)	Michaelis-Menten 模型(M2)	Curtis模型 (M3)	深度神经网络模型(DNN)
方法	Power模型 (M1)	Michaelis-Menten 模型(M2)	Curtis模型 (M3)	Relu	Adam	L2	Dropout
决定系数R²	0.634	0.661	0.63	0.661	0.669	0.671^*	0.661
平均绝对误差MAE	1.716	1.634	1.726	1.622	1.606	1.599^*	1.63
均方根误差RMSE	2.137	2.053	2.149	2.053	2.029	2.024^*	2.052
三七比例(训练集占比为30%、测试集占比70%)
方法	Power模型 (M1)	Michaelis-Menten 模型(M2)	Curtis模型 (M3)	深度神经网络模型(DNN)
方法	Power模型 (M1)	Michaelis-Menten 模型(M2)	Curtis模型 (M3)	Relu	Adam	L2	Dropout
决定系数R²	0.638	0.659	0.634	0.584	0.584	0.583	0.587
平均绝对误差MAE	1.766	1.688	1.774	1.716	1.715	1.716	1.711
均方根误差RMSE	2.191	2.107	2.201	2.116	2.117	2.120	2.110
四六比例(训练集占比为40%、测试集占比60%)
方法	Power模型 (M1)	Michaelis-Menten 模型(M2)	Curtis模型 (M3)	深度神经网络模型(DNN)
方法	Power模型 (M1)	Michaelis-Menten 模型(M2)	Curtis模型 (M3)	Relu	Adam	L2	Dropout
决定系数R²	0.634	0.662^*	0.631	0.654	0.656	0.657	0.654
平均绝对误差MAE	1.646	1.571^*	1.654	1.568	1.563	1.562	1.573
均方根误差RMSE	2.055	1.971^*	2.065	1.976	1.971	1.968	1.976