Estimation on Forest Above-Ground Biomass Based on Simulated Large-Footprint LiDAR and Multi-Layer Perceptron

doi:10.13466/j.cnki.lyzygl.2021.01.008

Abstract

Abstract:

Forests are important global terrestrial ecosystems.Sample survey is a commonly used method by countries to assess their forest resources and biomass.With the development of LiDAR technology,spaceborne large-footprint ladar become an option to estimate forest above-ground biomass(AGB) in large areas.In order to develop the method to estimate forest AGB with large-footprint LiDAR,the study proposes an AGB estimation model based on simulated large-footprint LiDAR and multi-layer perceptron.Based on 13 groups of LiDAR waveform parameters,the multi-layer perceptron achieves higher accuracy than multiple linear regression to estimate AGB.Compared with the field measured AGB,the deviation range of the estimated AGB from the multiple linear regression is between -34.96 to 23.28 t/hm² and the estimated deviation of the multi-layer perceptron is between -19.09 to 20.19 t/hm².Therefore,multi-layer perceptron is better than multiple linear regression in estimating forest AGB.

Key words: above-ground biomass, LiDAR, simulated waveform, multi-layer perceptron

CLC Number:

S771.8

XV Changjian, LIU Yingchun, ZUO Lijun, LI Jiangeng, ZHANG Ting, HAN Lumeng, FANG Yu, ZHANG Yin, WANG Tian. Estimation on Forest Above-Ground Biomass Based on Simulated Large-Footprint LiDAR and Multi-Layer Perceptron[J]. FOREST RESOURCES WANAGEMENT, 2021, (1): 50-60.

Figures/Tables 16

Tab.1

Fig.1

Fig.2

Tab.2

Allometric equation of biomass of main tree species

树种	异速生长方程	参考文献
落叶松(L.gmelinii)	$W = 0.271(D 2 H) 0.682$	[14]
油松(P.tabulaeformis)	$W = 0.160 D 2.100$	[15]
杨树(Populus L.)	$W = 0.078 D 2.558$	[16]
白桦(B.platyphylla Suk.)	$W = 0.164 D 2.400$	[17]
山杨(P.davidiana)	$W = 0.078 D 2.558$	[17]
蒙古栎(Q.mongolica)	$W = 0.131 D 2.437$	[17]
樟子松(P.sylvestnis var.)	$W = 0.153(D 2 H) 0.742$	[18]

Tab.2

Fig.3

Tab.3

Definition of waveform parameters

参数	定义
$Q x$	由回波波形确定的第 $x$ 百分位数的高度( $x$ =10,20,25,30,40,50,60,70,75,80,90,95)
$Avg$	平均高,即波形起始位置到波形结束位置之间的高度加权均值
$Max$	最大高,即波形起始位置与波形结束位置之间的高度
$CV$	波形变异系数
$SD$	波形标准差

Tab.3

Fig.4

Fig.5

Fig.6

Fig.7

Fig.8

Tab.4

Univariate linear regression model parameters and fit results

输入变量	$β$	$ε$	R²	Adj.R²	MAE/(t/hm²)	RMSE/(t/hm²)	RMSEr/%
Q₁₀	[1.263]	45.067	0.01	0.01	26.79	38.99	15.27
Q₂₅	[2.297]	36.302	0.05	0.05	25.83	38.26	14.98
Q₅₀	[3.419]	23.167	0.14	0.14	23.31	36.40	14.25
Q₇₅	[4.601]	11.668	0.26	0.26	21.81	33.67	13.19
Q₉₅	[4.131]	3.678	0.40	0.40	20.29	27.88	10.92
Max	[3.735]	2.625	0.49	0.49	21.13	30.35	11.88
Avg	[2.057]	48.043	0.01	0.01	26.74	38.92	15.24
CV	[-9.357]	66.130	0.07	0.07	26.97	37.86	11.83
SD	[76.263]	34.017	0.01	0.01	27.15	38.96	15.26

Tab.4

Tab.5

Parameters of MLR model

编号	波形参数组合	$β$	$ε$
1	Avg,CV,SD	[0.423,-10.109,-35.482]	74.666
2	Q₂₅,Q₅₀,Q₇₅	[-3.503,-7.643,12.551]	14.323
3	Q₁₀,Q₅₀,Q₉₅	[-2.014,-4.683,7.631]	11.499
4	Q₁₀,Q₇₅,Q₉₅	[-2.336,-6.626,9.844]	11.534
5	Q₁₀,Q₅₀,Max	[-0.993,-1.858,4.682]	10.311
6	Q₁₀,Q₂₅,Q₅₀	[-2.534,-7.793,11.455]	31.133
7	Q₅₀,Q₇₅,Q₉₅	[-3.218,-5.908,10.787]	10.976
8	Q₅₀,Q₇₅,Max	[-4.641,2.380,4.316]	9.926
9	Q₇₅,Q₉₅,Max	[-7.749,7.178,2.679]	9.198
10	Q₁₀,Q₂₅,Q₅₀,Q₇₅,Q₉₅	[-1.897,-0.935,0.223,-7.099,10.619]	11.493
11	Q₁₀,Q₂₅,Q₅₀,Q₇₅,Q₉₅,Max	[-0.847,-1.295,-0.015,-5.287,6.608,2.218]	10.363
12	Q₁₀,Q₂₅,Q₅₀,Q₇₅,Q₉₅,Max,Avg,CV,SD	[-2.105,-2.792,-0.809,4.231,8.881,1.419,0.833,13.184,138.172]	-46.816
13	ALL^*	[-1.131,-4.441,1.826,-0.799,-0.863,1.066,-1.685,-2.498,4.522,-9.013,6.122,5.783,0.895,13.100,1.543,137.889]	-46.717

Tab.5

Tab.6

Fig.9

Fig.10

References 24

[1]	Fang Jingyun, Yu Guirui, Liu Lingli, et al. Climate change,human impacts,and carbon sequestration in China[J]. Proceedings of the National Academy of Sciences, 2018,115(16):4015-4020.
[2]	Nascimento H E M, Laurance W F. Total above-ground biomass in central Amazonian rainforests:a landscape-scale study[J]. Forest ecology and management, 2002,168(1-3):311-321.
[3]	Nie Sheng, Wang Cheng, Zeng Hongcheng, et al. Above-ground biomass estimation using airborne discrete-return and full-waveform LiDAR data in a coniferous forest[J]. Ecological Indicators, 2017,78:221-228.
[4]	刘茜, 杨乐, 柳钦火, 等. 森林地上生物量遥感反演方法综述[J]. 遥感学报, 2015,19(1):62-74.
[5]	刘清旺, 李增元, 陈尔学, 等. 机载LIDAR点云数据估测单株木生物量[J]. 高技术通讯, 2010,7:765-770.
[6]	洪奕丰, 张守攻, 陈伟, 等. 基于机载激光雷达的落叶松组分生物量反演[J]. 林业科学研究, 2019(5):12.
[7]	Margolis H A, Nelson R F, Montesano P M, et al. Combining satellite lidar,airborne lidar,and ground plots to estimate the amount and distribution of above-ground biomass in the boreal forest of North America[J]. Canadian Journal of Forest Research, 2015,45(7):838-855.
[8]	Muss J D, Mladenoff D J, Townsend P A. A pseudo-waveform technique to assess forest structure using discrete lidar data[J]. Remote Sensing of Environment, 2011,115(3):824-835.
[9]	吴娇娇, 欧光龙, 舒清态. 基于BP神经网络模型思茅松天然林生物量遥感估测[J]. 中南林业科技大学学报, 2017,37(7):30-35.
[10]	孙翊, 姜树海, 陈至灵. 人工神经网络在林业上的应用研究进展[J]. 世界林业研究, 2019,32(3):7-12.
[11]	徐辉, 潘萍, 宁金魁, 等. 多元线性回归与神经网络模型在森林地上生物量遥感估测中的应用[J]. 东北林业大学学报, 2018,46(1):63-67.
[12]	徐奇刚, 雷相东, 国红, 等. 基于多层感知机的长白落叶松人工林林分生物量模型[J]. 北京林业大学学报, 2019,41(5):97-107.
[13]	Le Cun Y, Bengio Y, Hinton G. Deep learning[J]. nature, 2015,521(7553):436-444. doi: 10.1038/nature14539 pmid: 26017442
[14]	罗云建, 张小全, 王效科, 等. 华北落叶松人工林生物量及其分配模式[J]. 北京林业大学学报, 2009,6(1):13-18.
[15]	白静. 油松人工林生长特征及其与林分结构关系研究[D]. 呼和浩特:内蒙古农业大学, 2008.
[16]	吴俊民, 苏贵林, 刘成志, 等. 黑龙江省西部杨树人工林生物量的估算方法[J]. 林业科技, 2000(3):14-16.
[17]	Wang Chuankuan. Biomass allometric equations for 10 co-occurring tree species in Chinese temperate forests[J]. Forest Ecology and Management, 2006,222(1-3):9-16.
[18]	贾炜玮, 于爱民. 樟子松人工林单木生物量模型研究[J]. 林业科技情报, 2008,40(2):1-2
[19]	Hancock S, Armston J, Hofton M, et al. The GEDI simulator:a large-footprint waveform Lidar simulator for calibration and validation of spaceborne missions[J]. Earth and Space Science, 2019,6(2):294-310. doi: 10.1029/2018EA000506 pmid: 31008149
[20]	Chen J M, Pavlic G, Brown L, et al. Derivation and validation of Canada-wide coarse-resolution leaf area index maps using high-resolution satellite imagery and ground measurements[J]. Remote Sensing of Environment, 2002,80(1):165-184. doi: 10.1016/S0034-4257(01)00300-5
[21]	Richardson J J, Moskal L M, Kim S H. Modeling approaches to estimate effective leaf area index from aerial discrete-return LIDAR[J]. Agricultural and Forest Meteorology, 2009,149(6-7):1152-1160. doi: 10.1016/j.agrformet.2009.02.007
[22]	Hyer E J, Goetz S J. Comparison and sensitivity analysis of instruments and radiometric methods for LAI estimation:assessments from a boreal forest site[J]. Agricultural and Forest Meteorology, 2004,122(3-4):157-174. doi: 10.1016/j.agrformet.2003.09.013
[23]	张志, 田昕, 陈尔学, 等. 森林地上生物量估测方法研究综述[J]. 北京林业大学学报, 2011,33(5):144-150.
[24]	Goodfellow I, Bengio Y, Courville A. Deep learning[M]. Cambridge,MA:MIT press, 2016.

参数名称	参数数值
激光发射频率/(kHz)	400
飞行速度/(km/h)	20
平均绝对航高/m	1650
高程/m	1500
相对航高/m	150
航线间距/m	260
旁向重叠度/%	50

编号	参数组合	R² (MLR\|MLP)	Adj.R² (MLR\|MLP)	MAE(t/hm²) (MLR\|MLP)	RMSE(t/hm²) (MLR\|MLP)	RMSE_r/% (MLR\|MLP)
1	Avg,CV,SD	0.07\|0.23	0.07\|0.22	26.75\|23.39	37.81\|36.62	14.80\|14.34
2	Q₂₅,Q₅₀,Q₇₅	0.43\|0.44	0.43\|0.44	19.57\|16.58	29.60\|31.18	11.59\|12.18
3	Q₁₀,Q₅₀,Q₉₅	0.53\|0.66	0.53\|0.66	19.05\|13.44	26.75\|24.31	10.47\|9.52
4	Q₁₀,Q₇₅,Q₉₅	0.55\|0.71	0.54\|0.71	18.92\|12.73	26.43\|22.31	10.35\|8.74
5	Q₁₀,Q₅₀,Max	0.54\|0.77	0.54\|0.77	19.46\|11.12	26.70\|19.97	10.46\|7.81
6	Q₁₀,Q₂₅,Q₅₀	0.28\|0.42	0.28\|0.41	20.76\|17.83	33.33\|31.90	13.05\|12.49
7	Q₅₀,Q₇₅,Q₉₅	0.54\|0.69	0.53\|0.68	19.34\|12.24	26.73\|23.35	10.47\|9.14
8	Q₅₀,Q₇₅,Max	0.54\|0.78	0.54\|0.77	19.64\|11.04	26.66\|19.69	10.44\|7.71
9	Q₇₅,Q₉₅,Max	0.55\|0.76	0.55\|0.76	19.40\|11.79	26.27\|20.30	10.29\|7.95
10	Q₁₀,Q₂₅,Q₅₀,Q₇₅,Q₉₅	0.55\|0.75	0.54\|0.74	18.96\|11.97	26.42\|20.90	10.35\|8.18
11	Q₁₀,Q₂₅,Q₅₀,Q₇₅,Q₉₅,Max	0.56\|0.76	0.56\|0.75	19.12\|11.15	26.04\|20.62	10.20\|8.08
12	Q₁₀,Q₂₅,Q₅₀,Q₇₅,Q₉₅,Max,Avg,CV,SD	0.60\|0.77	0.60\|0.76	17.44\|10.81	24.74\|19.97	9.69\|7.82
13	ALL	0.60\|0.80	0.60\|0.79	17.45\|10.48	24.66\|18.52	9.65\|7.25