机载LiDAR数据的林分胸高断面积反演研究

doi:10.13466/j.cnki.lyzygl.2021.05.009

摘要/Abstract

摘要：

机载LiDAR数据目前在森林生物量、树高以及郁闭度估测方面得到广泛研究,但估测林木胸高断面积的研究较少。以高峰林场为研究区,借助机载LiDAR数据并结合地面实测的105块样地反演林分胸高断面积。首先,选取4块坡度以及林分郁闭度有差异的样地,运用渐进不规则三角网(PTIN)、渐进形态学滤波算法(PMF)、布料模拟滤波算法(CSF)和基于插值的滤波方法(IBF)分别对点云数据进行滤波,以便选取最适的滤波方法完成样地滤波;其次,通过随机森林(RF)和迭代的决策树(GBRT)算法分别对林分胸高断面积进行估测;最后,选用精度较好的模型完成林分胸高断面积反演和制图。结果表明:布料模拟滤波算法(CSF)在坡度为25~33°、郁闭度为0.5~0.7的样地环境时,LiDAR数据滤波效果较好,这和研究区的环境基本一致,故选用CSF算法对研究区进行滤波;研究所构建的胸高断面积反演模型中,RF模型泛化能力优于GBRT模型,RF模型的R²为0.77,RMSE为3.99m²/hm²,rRMSE为17.76%;独立样本检验的R²为0.66,RMSE为3.27m²/hm²,rRMSE为14.73%,故采用RF模型完成研究区林分胸高断面积反演。

关键词: 胸高断面积, 机载LiDAR数据, 滤波, 特征选择, 机器学习

Abstract:

Airborne LiDAR data has been widely studied in the estimation of forest biomass,tree height and canopy density,but the research on the estimation of basal area is rare.Taking the Gaofeng Forest Farm as the research area,the forest basal area was inversed by using airborne LiDAR data and combining with 105 sample plots measured on the ground.Firstly,in order to find the optimal filtering method to complete the LiDAR data filtering,four algorithms,namely,progressive triangulated irregular network(PTIN),the progressive morphological filter (PMF),the cloth simulation filter (CSF)and the Interpolation-based filtering (IBF)were used in the LiDAR data for 4 sample plots with different slopes and different forest canopy densities respectively,and then,random forest (RF)and iterative decision tree (GBRT)algorithms were used to estimate the basal area of the forest respectively.Lastly,the model with good precision was selected to complete the inversion and mapping of stand basal area.The results showed that cloth simulation filter (CSF)had good filtering effect on the LiDAR data of the sample plots with slope of 25~33° and canopy density of 0.5~0.7,which was basically consistent with the forest condition of the study area,therefore,CSF algorithm was selected to filter the LiDAR data in this study;in the stand basal area inversion model,the generalization ability of RF model was superior to that of GBRT model,with the R² of 0.77,RMSE of 3.99m²/hm² and rRMSE of 17.76%,respectively;while for the independent sample test for the RF model,the correspondent R ²,RMSE and rRMSE was 0.66,3.27m²/hm²,and 14.73% respectively.So RF model was used in the inversion of basal area of forest stand in the study area.

Key words: basal area, airborne LiDAR data, filtering, feature selection, machine learning

中图分类号:

S771.8

龙飞, 岳彩荣, 金京, 李春干, 罗洪斌, 徐婉婷. 机载LiDAR数据的林分胸高断面积反演研究[J]. 林业资源管理, 2021,(5): 62-69.

LONG Fei, YUE Cairong, JIN Jing, LI Chungan, LUO Hongbin, XU Wanting. Study on the Inversion of Basal Area from Airborne LiDAR Data[J]. FOREST RESOURCES WANAGEMENT, 2021,(5): 62-69.

图/表 9

图1

表1

表2

表3

表4

表5

图2

表6

图3

参考文献 30

[1]	Taubert F, Fischer R, Knapp N, et al. Deriving tree size distributions of tropical forests from iidar[J]. Remote Sensing, 2021, 13(1):131. doi: 10.3390/rs13010131
[2]	郭庆华, 苏艳军, 胡天宇, 等. 激光雷达森林生态应用——理论、方法及实例[M]. 北京: 高等教育出版社, 2018: 12.
[3]	李海奎, 雷渊才. 中国森林植被生物量和碳储量评估[M]. 北京: 中国林业出版社, 2010: 28.
[4]	陈迪. 基于遥感技术的甘南州陆地生态系统NEP研究[D]. 甘肃:兰州大学, 2016.
[5]	刘亚男. 基于多源遥感数据的森林地上生物量及净初级生产力估算研究[J]. 测绘学报, 2020, 49(12):1641.
[6]	Paolo M, Vibrans A C, Mcroberts R E, et al. Methods for variable selection in LiDAR assisted forest inventories[J]. Forestry, 2017, 90:112-124. doi: 10.1093/forestry/cpw041
[7]	郭庆华, 刘瑾, 陶胜利, 等. 激光雷达在森林生态系统监测模拟中的应用现状与展望[J]. 科学通报, 2014, 59:459-478.
[8]	罗洪斌, 岳彩荣, 张国飞, 等. 机载激光雷达在不同区域尺度森林叶面积指数反演中的应用[J]. 西部林业科学, 2021, 50(4):33-40.
[9]	王蕊, 邢艳秋, 尤号田, 等. 基于星载LiDAR波形数据的森林胸高断面积估测研究[J]. 西北林学院学报, 2014. 30(5):156-162.
[10]	Strimbu V F, Ene L T, Gobakken T, et al. Post-stratified change estimation for large-area forest biomass using repeated ALS strip sampling[J]. Canadian Journal of Forest Research, 2017, 47(6):839-847. doi: 10.1139/cjfr-2017-0031
[11]	White J C, Wulder M A, Varhola, Andrés, et al. A best practices guide for generating forest inventory attributes from airborne laser scanning data using an area-based approach[J]. Forestry Chronicle, 2013, 89(6):722-723. doi: 10.5558/tfc2013-132
[12]	庞勇, 李增元, 陈尔学, 等. 激光雷达技术及其在林业上的应用[J]. 林业科学, 2005(3):129-136.
[13]	Naesset E. Determination of mean tree height of forest stands using airborne laser scanner data[J]. ISPRS journal of photogrammetry and remote sensing, 1997, 52(2):49-56. doi: 10.1016/S0924-2716(97)83000-6
[14]	Nelson R, Krabill W, Tonelli J. Estimating forest biomass and volume using airbone laser data[J]. Remote Sensing of Environment, 1988, 24(2):247-267. doi: 10.1016/0034-4257(88)90028-4
[15]	Naesset E. Estimating timber volume of forest stands using airborne laser scanner data[J]. Remote Sensing of Environment, 1997, 61(2):246-253. doi: 10.1016/S0034-4257(97)00041-2
[16]	Silva C A, Carine K, Hudak A T, et al. Modeling and mapping basal area of Pinus taeda L.plantation using airborne LiDAR data[J]. Anais da Academia Brasileira de Ciências, 2017, 89(3):1895-1905. doi: 10.1590/0001-3765201720160324
[17]	Ferreiro G, Aranda D, Miranda A. Estimation of stand variables in Pinus radiata D.Don plantations using different LiDAR pulse densities[J]. Forestry, 2012, 85(2):281-292. doi: 10.1093/forestry/cps002
[18]	Dash J P, Marshall H M, Brian R. Methods for estimating multivariate stand yields and errors using k-NN and aerial laser scanning[J]. Forestry, 2015(2):237-247.
[19]	Lefsky M A, Cohen W B, Acker S A, et al. Lidar Remote Sensing of the Canopy Structure and Biophysical Properties of Douglas-Fir Western Hemlock Forests[J]. Remote Sensing of Environment, 1999, 70(3):339-361. doi: 10.1016/S0034-4257(99)00052-8
[20]	赵勋, 岳彩荣, 李春干, 等. 基于机载LiDAR数据估测林分平均高[J]. 林业科学研究, 2020, 33(4):59-66.
[21]	Zhao Xiaoqian, Guo Qinghua, Su Yanjun, et al. Improved progressive TIN densification filtering algorithm for airborne LiDAR data in forested areas[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2016, 117:79-91. doi: 10.1016/j.isprsjprs.2016.03.016
[22]	Zhang Keqi, Chen Shucheng, Whitman D, et al. A progressive morphological filter for removing nonground measurements from airborne LIDAR data[J]. IEEE transactions on geoscience and remote sensing, 2003, 41(4):872-882. doi: 10.1109/TGRS.2003.810682
[23]	Evans J S, Hudak A T. A multiscale curvature algorithm for classifying discrete return LiDAR in forested environments[J]. IEEE Transactions on Geoscience and Remote Sensing, 2007, 45(4):1029-1038. doi: 10.1109/TGRS.2006.890412
[24]	赵勋, 岳彩荣, 李春干, 等. 基于机载LiDAR点云数据森林郁闭度估测[J]. 遥感技术与应用, 2020, 35(5):1136-1145.
[25]	曹林, 代劲松, 徐建新, 等. 基于机载小光斑LiDAR技术的亚热带森林参数信息优化提取[J]. 北京林业大学学报, 2014, 36(5):13-21.
[26]	张加龙, 胥辉, 陆驰. 应用Landsat8 OLI和GBRT对高山松地上生物量的估测[J]. 东北林业大学学报, 2018, 46(8):25-30.
[27]	吴立志, 陈振南, 张鹏. 基于随机森林算法的城市火灾风险评估研究[J/OL]. 灾害学:1-10 [2021-09-25]. http://kns.cnki.net/kcms/detail/61.1097.P.20210729.1558.014.html .
[28]	Friedman J H. Stochastic gradient boosting[J]. Computational Statistics & Data Analysis, 2002, 38(4):367-378. doi: 10.1016/S0167-9473(01)00065-2
[29]	高婷, 李卫忠, 赵鹏祥, 等. 基于ArboLiDAR的大野口林区森林参数估测[J]. 西北林学院学报, 2017, 32(4):172-177.
[30]	郝红科. 基于机载激光雷达的森林参数反演研究[D]. 西北农林科技大学, 2019.

森林参数	针叶林			阔叶林
森林参数	范围	平均值	标准差	范围	平均值	标准差
林分平均高/m	6.2~ 18.22	13.35	2.97	6.52~ 23.56	13.74	4.30
蓄积量/(m³/hm²)	57.93~387.2	17.59	63.76	21.28~317.13	139.23	76.61
胸高断面积/(m²/hm²)	15.20~ 50.21	25.60	6.70	3.92~ 36.69	19.25	7.46

		滤波结果		误差计算公式
		地面点	非地面点	TⅠ=b/(a+b)	Po=(a+d)/e
参考	地面点	a	b	TⅡ=c/(c+d)	Pe=((a+b)(a+c)+(c+d)(b+d))/e²
参考	非地面点	c	d	TE=(b+c)/e	Kp=(Po-Pe)/(1-pe)

点云变量符号	变量描述
Hmin/Hmax/ Hmean/ Hmad	点云最低高度/最高高度/平均高度/中位数
Hccr	冠层起伏率;Hccr=(mean-min)/(max-min)。mean,min,max分别为每一统计单元内所有点的平均高度、最小高度、最大高度
Hvar/Hstdv/Hskew/Hkurt/Hcv	点云高度方差/标准差/偏态/峭度/变动系数
H10,H20,H25, H30,H40,H50, H60,H70,H75, H80,H95,H99	点云高度百分位数,样地内所有归一化的激光雷达点云按高度进行排序,然后计算每一统计单元内X%的点所在的高度
Hiq	高度百分位数四分位数间距:Hiq=H75-H25
D[0-9]	密度变量(10个):将点云数据从低到高分成10个相同高度的切片,每层回波数的比例就是相应的密度变量
CC	郁闭度^[24]

样地号	坡度/ (°)	郁闭度	点云个数/ point	点云密度/ (point/m²)
1	35.69	0.49	2312	0.93
2	26.11	0.58	4109	2.65
3	32.56	0.52	4590	2.24
4	25.69	0.65	2828	1.18

样地	滤波方法	TⅠ	TⅡ	TE	Kp/%
1	PMF	0.20	0.04	0.19	74.70
	CSF	0.09	0.08	0.09	72.40
	PTIN	0.11	0.08	0.01	69.00
	IBF	0.13	0.04	0.05	78.70
2	PMF	0.08	0.23	0.20	51.49
	CSF	0.04	0.19	0.16	59.12
	PTIN	0.02	0.21	0.17	57.78
	IBF	0.12	0.16	0.15	58.90
3	PMF	0.20	0.17	0.18	53.73
	CSF	0.20	0.16	0.17	54.90
	PTIN	0.17	0.11	0.12	64.90
	IBF	0.22	0.10	0.12	63.76
4	PMF	0.09	0.21	0.13	71.17
	CSF	0.13	0.13	0.13	71.60
	PTIN	0.05	0.26	0.13	71.00
	IBF	0.11	0.17	0.13	70.66