Research on Estimation of Coniferous Forest Volume in Longnan County Based on Landsat 8 and PALSAR-2 Images

doi:10.13466/j.cnki.lyzygl.2021.01.010

Abstract

Abstract:

Forest stand volume estimation is an important research field of forestry remote sensing.Factors such as cloud and fog weather and spectral saturation have restricted the accuracy of optical remote sensing image estimation.Synthetic Aperture Radar(SAR) images have the characteristics of strong penetrability and are less likely to be affected by cloud and fog,which make up for the deficiencies of optical remote sensing.This study uses the coniferous forest in Longnan county,Jiangxi Province as the study area,combines Landsat 8 and PALSAR-2 dual-polarization SAR image data to extract a total of 245 remote sensing factors such as spectral information,vegetation index,texture information and backscattering coefficient based on remote sensing data preprocessing.Based on the Pearson correlation coefficient method and the multiple stepwise regression method,65 remote sensing factors are selected for the estimation of stand stock.Taking forest stand canopy closure as a stratification factor,the study adopts five models of linear,KNN,support vector machine(SVM),multi-perceptron(MLP) and random forest(RF) to estimate the forest stand volume,and tests the accuracy of the estimated results.The experimental results show that:1) Compared with using the spectrum and texture information of Landsat8 alone,the backscatter information of PALSAR-2 based on the canopy closure classification and fusion significantly improves the inversion accuracy of accumulation 2) In low canopy closure forest stand,the linear model has the highest accuracy(rRMSE=21.16%),in medium canopy closed forest stand,the MLP model has the best estimation effect(rRMSE=30.61%),in high canopy closure forest stand,the MLP model has the best estimation effect(rRMSE=27.53%).Based on the backscattering coefficient of PALSAR-2,the canopy closure stratification can effectively improve the inversion accuracy of the medium and high accumulation areas.

Key words: canopy density classification, PALSAR-2, stand volume, multi-perceptron model, coniferous forest

CLC Number:

S758
S771.8

LUO Kaijian, XV Xiaodong, LONG Jiangping, XV Congrong, LIN Hui, HE Xiaofeng. Research on Estimation of Coniferous Forest Volume in Longnan County Based on Landsat 8 and PALSAR-2 Images[J]. FOREST RESOURCES WANAGEMENT, 2021, (1): 69-76.

Figures/Tables 10

Fig.1

Fig.2

Fig.3

Fig.4

Tab.1

Tab.2

Tab.3

Fig.5

Fig.6

Fig.7

References 24

[1]	胡卫东. “3S”技术在森林资源二类调查中的应用[D]. 湖南:中南林业科技大学, 2018.
[2]	贾嘉辉. 高分遥感森林蓄积量估测算法研究[D]. 西安:西安科技大学, 2017.
[3]	张安定, 吴孟泉, 王大鹏, 等. 遥感技术基础与应用[M]. 北京: 科学出版社, 2013.
[4]	Van Den Eeckhaut M, Kerle N, Poesen J, et al. Object-oriented identification of forested landslides with derivatives of single pulse LiDAR data[J]. Geomorphology, 2012,173:30-42.
[5]	Curran P J, Dungan J L, Gholz H L. Seasonal LAI in slash pine estimated with Landsat TM[J]. Remote Sensing of Environment, 1992,39(1):3-13.
[6]	刘兆华, 林辉, 龙江平, 等. 基于高分二号的旺业甸林场蓄积量估测模型研究[J]. 中南林业科技大学学报, 2020,40(3):79-84.
[7]	Delbart N, Melon P, Florsch G, et al. Forest biomass retrieval using L-band polarimetric measurements [C]//IEEE International Geoscience and Remote Sensing Symposium.IEEE, 2002,3:1789-1791.
[8]	Askne J, Santoro M. Multitemporal repeat pass SAR interferometry of boreal forests[J]. IEEE Transactions on Geoscience and Remote Sensing, 2005,43(6):1219-1228.
[9]	李静. 基于光学和SAR遥感的福建省森林生物量估算[D]. 福建:福州大学, 2017.
[10]	Zhang Haibo, Zhu Jianjun, Wang Changcheng, et al. Forest growing stock volume estimation in subtropical mountain areas using PALSAR-2 L-Band PolSAR data[J]. Forests, 2019,10(3):276.
[11]	李佳. 基于GF-2影像杉木材积反演研究[D]. 湖南:中南林业科技大学, 2019.
[12]	王月婷, 张晓丽, 杨慧乔, 等. 基于Landsat 8卫星光谱与纹理信息的森林蓄积量估算[J]. 浙江农林大学学报, 2015,32(3):384-391.
[13]	刘俊. 基于ALOS遥感影像纹理信息的怀柔区针、阔叶林蓄积量反演模型研究[D]. 北京:北京林业大学, 2014.
[14]	孙雪莲. 基于Landsat8-OLI的香格里拉高山松林生物量遥感估测模型研究[D]. 昆明:西南林业大学, 2016.
[15]	Li Xinyu, Liu Zhaohua, Lin Hui, et al. Estimating the growing stem volume of Chinese Pine and Larch Plantations based on fused optical data using an improved variable screening method and stacking algorithm[J]. Remote Sensing, 2020,12(5):871.
[16]	徐婷, 曹林, 佘光辉. 基于Landsat 8 OLI的特征变量优化提取及森林生物量反演[J]. 遥感技术与应用, 2015,30(2):226-234.
[17]	姚明臣. 机器学习和神经网络学习中的若干问题研究[D]. 大连:大连理工大学, 2016.
[18]	琚存勇, 蔡体久. 用泛化改进的 BP 神经网络估测森林蓄积量[J].林业科学,2006(12).
[19]	李德仁, 王长委, 胡月明, 等. 遥感技术估算森林生物量的研究进展[J]. 武汉大学学报:信息科学版, 2012,37(6):631-635.
[20]	刘伯涛, 李崇贵, 郭瑞霞, 等. 运用GF-1影像光谱和纹理信息构建森林蓄积量估测模型[J]. 东北林业大学学报, 2020,48(1):9-12.
[21]	徐新良, 曹明奎. 森林生物量遥感估算与应用分析[J]. 地球信息科学学报, 2012,8(4):122-128.
[22]	宋亚斌, 邢元军, 江腾宇, 等. 基于距离相关系数和KNN回归模型的森林蓄积量估测研究[J]. 中南林业科技大学学报, 2020,40(4):22-27.
[23]	杜培军, 谭琨, 夏俊士. 高光谱遥感影像分类与支持向量机应用研究[M]. 北京: 科学出版社.
[24]	蒋馥根, 孙华, 林辉, 等. 旺业甸林场人工林生物量遥感反演研究[J]. 中南林业科技大学学报, 2019,39(10):88-94.

变量名	变量形式
单波段(7个)	Band1			Band2		Band3		Band4		Band5	Band6		Band7
植被指数 (32个)	ARVI			B		W		G		EVI	II		NDVI
	PVI			SARV		MVI5		MVI7		ND43	ND67		ND563
	VI3			B_ALBEDO		G_ALBEDO		R_ALBEDO		N_ALBEDO	R1		R3
	R2			DVI(24)		DVI(34)		DVI(54)		RVI(24)	RVI(34)		RVI(54)
	ALBEDO			VIS234		VIS2345		R4
图像变换特征 (26个)	KT1			KT2		KT3		PC1_A		PC1_B	PC1_P		PC2_A
	PC2_B			PC2_P		PC3_A		PC3_B		PC3_P	MNF_B1		MNF_B2
	MNF_B3			MNF_B4		MNF_B5		MNF_B6		MNF_B7	PCA_B1		PCA_B2
	PCA_B3			PCA_B4		PCA_B5		PCA_B6		PCA_B7
纹理特征(168个)	窗口3×3,5×5,7×7
平均值	MEAN1			MEAN2		MEAN3		MEAN4		MEAN5	MEAN6		MEAN7
方差	VAR1			VAR2		VAR3		VAR14		VAR5	VAR6		VAR7
协同性	HOM1			HOM2		HOM3		HOM4		HOM5	HOM6		HOM7
对比度	CON1			CON2		CON3		CON4		CON5	CON6		CON7
相异性	DIS1			DIS2		DIS3		DIS4		DIS5	DIS6		DIS7
熵	ENT1			ENT2		ENT3		ENT4		ENT5	ENT6		ENT7
变量名		变量形式
二阶矩		SM1	SM2		SM3		SM4		SM5			SM6		SM7
相关性		COR1	COR2		COR3		COR4		COR5			COR6		COR7
SAR因子 (12)		HH	HV		HH/(HV+HH)		HV/(HV+HH)		HH/HV			HV/HH
SAR因子 (12)		HH-HV	HV-HH		HV×HH		HV+HH		(HH-HV)/(HV+HH)			(HV-HH)/(HV+HH)

数据源	郁闭度
Landsat 8	不分级	Band6	3_b3_sm	3_b4_cor	3_b6_cor	5_b2_hom	5_b5_hom	5_b5_sm	7_b3_ent	7_b3_mean
	低	5_b4_cor	3_b3_sm	3_b1_con	7_b6_sm	3_b2_con	3_b7_con
	中	Band3	3_b3_cor	3_b2_hom	3_b7_cor	3_b1_sm	5_b7_sm	PC3_A
	高	3_ent_rvi	3_b6_cor	3_b7_sm	N_Albedo	7_b1_var	3_b1_cor	7_b5_var	pca_b2
Landsat 8+ Alos-palsar 2	不分级	Band6	3_b3_sm	3_b4_cor	3_b6_cor	5_b2_hom	5_b5_hom	5_b5_sm	7_b3_ent	7_b3_mean
	低	5_b4_cor	3_b3_sm	3_b1_con	7_b6_sm	3_b2_con	3_b7_con
	中	3_b3_cor	3_b6_cor	3_b2_hom	3_b7_cor	3_b1_sm	hv_hh	Band6	7_b2_cor	5_b1_con	3_b7_hom	5_b1_cor
	高	3_ent_rvi	3_b6_cor	3_b7_sm	N_Albedo	7_b1_var	3_b1_cor	7_b5_var	pca_b2	hh

方法	R²					RMSE					rRMSE/%
方法	线性	KNN	SVM	MLP	RF	线性	KNN	SVM	MLP	RF	线性	KNN	SVM	MLP	RF
L(不分级)	0.56	0.27	0.49	0.57	0.26	34.58	44.80	37.26	34.49	45.10	41.09	53.24	44.27	40.98	53.59
L(低)	0.92	-0.31	-0.09	0.70	0.04	4.43	18.18	16.59	8.74	15.52	21.16	86.87	79.28	41.78	74.18
L(中)	0.68	0.40	0.71	0.71	0.33	24.09	32.88	22.99	22.80	34.87	34.83	47.53	33.23	32.95	50.41
L(高)	0.46	0.15	0.32	0.43	-0.03	33.92	42.60	38.18	34.87	46.94	29.85	37.50	33.61	30.69	41.32
L+P(不分级)	0.56	0.27	0.49	0.57	0.26	34.58	44.80	37.26	34.49	45.10	41.09	53.24	44.27	40.98	53.59
L+P(低)	0.92	-0.31	-0.09	0.70	0.04	4.43	18.18	16.59	8.74	15.52	21.16	86.87	79.28	41.78	74.18
L+P(中)	0.75	0.31	0.64	0.75	0.16	21.27	35.41	25.63	21.18	39.02	30.74	51.19	37.05	30.61	56.40
L+P(高)	0.40	0.13	0.38	0.54	-0.01	35.69	43.13	36.43	31.28	46.34	31.41	37.96	32.06	27.53	40.79