Forest above ground biomass inversion using machine learning and sentinel data

doi:10.13466/j.cnki.lczyyj.2025.04.011

Abstract

Abstract:

To investigate the potential of synergistic active-passive remote sensing for estimating forest aboveground biomass (AGB),with the central urban area of Guangzhou as the study area,31 multi-source remote sensing features (including 6 SAR features and 25 optical features) were extracted through Sentinel-1 SAR data and Sentinel-2 multispectral imagery.Combined with field-measured AGB,six machine-learning (ML) regression models (Random Forest,Support Vector Machine,Extreme Gradient Boosting,k-Nearest Neighbors regression,AdaBoost,and Linear Regression) were used to develop forest AGB inversion models.The results showed that:1) The Visible Atmospherically Resistant Index Green (VIGreen) performed prominently in forest AGB inversion,ranking fifth in feature importance in the Random Forest (RF) model;different polarization combinations also contributed significantly to AGB inversion;2) Across different dataset combinations,the RF model achieved the highest accuracy among the six regression models;3) Models using only optical data outperformed those using only SAR data;4) Fusion of SAR and optical data produced substantially higher AGB inversion accuracy than using SAR or optical data alone:compared with SAR-only features,the coefficient of determination (R²) increased by 0.48 and the root mean square error (RMSE) decreased by 3.73;compared with optical-only features,R² increased by 0.14 and RMSE decreased by 2.08.Therefore,ML approaches that integrate optical and SAR data can effectively improve the accuracy of forest AGB inversion.

Key words: forest biomass inversion, sentinel data, machine learning, random forest

CLC Number:

S718.5

LIU Gao, XIE Zeqi, ZHOU Jianhao, LIAO Lipeng. Forest above ground biomass inversion using machine learning and sentinel data[J]. Forest and Grassland Resources Research, 2025, (4): 101-111.

Figures/Tables 10

Tab.1

Tab.2

Tab.3

Characteristic factors

特征参数		公式	释义
VV			垂直发射-垂直接收的SAR极化信号,用于表征地表散射特性
VH			垂直发射-水平接收的SAR极化信号,对植被结构和体积散射敏感
VV+VH		V_V+V_H	V_V和V_H极化信号的和,增强对地表综合散射特征的表征
VV-VH		V_V - V_H	V_V和V_H极化信号的差,突出植被与土壤散射的差异
VV/VH		V_V/V_H	V_V与V_H极化信号的比值,能反映植被结构和生物量的变化
VV×VH		V_V×V_H	V_V与V_H极化信号的乘积,放大植被和地表散射的交互效应
超蓝光			超蓝光波段(通常约400~450 nm),对植被色素和大气散射敏感
蓝光			蓝光波段(约450~500 nm),能反映叶绿素吸收和植被覆盖信息
绿光			绿光波段(约500~570 nm),与植被绿色反射峰相关
红光			红光波段(约620~670 nm),植被叶绿素强烈吸收
红边1			红边波段1(约680~710 nm),植被反射急剧增加的过渡区
红边2			红边波段2(约710~740 nm),进一步反映红边特征
红边3			红边波段3(约740~780 nm),红边高反射区
近红外1			近红外波段1(约780~850 nm),植被高反射区
近红外2			近红外波段2(约850~1 000 nm),进一步增强植被反射特征
短波红外1			短波红外波段1(约1 400~1 600 nm),对水分和植被干物质敏感
短波红外2			短波红外波段2(约2 100~2 300 nm),反映土壤和植被水分含量
短波红外3			短波红外波段3(约2 300~2 500 nm),进一步增强水分和植被特征提取
大气阻挡抗植被指数		$R N I R - R R E D + β × (R B L U E - R R E D) R N I R + R R E D - β × (R B L U E - R R E D)$	通过减少大气影响增强植被监测精度
植被聚焦指数		$R N I R R G R E E N$ -1	突出近红外与绿光差异,反映植被覆盖和健康状态
增强型植被指数	2.5×(R_NIR-R_RED)			增强植被反射特征,减少土壤和大气噪声
绿度植被指数	$R G R E E N R R E D$			绿光与红光比值,反映植被绿色程度和叶绿素含量
改进型土壤调整植被指数	$12$ ×[2×(R_NIR+1)- $(2 × R N I R + 1) × 2 - 8 × (R N I R - R R E D)$ ]			减少土壤背景影响
重归一化差异红边指数	$(R N I R - R B A N D 5) (N R I R + R B A N D 5)$			利用红边波段与近红外的差异,监测植被叶绿素含量和健康状态
归一化植被指数	$R N I R - R R E D R N I R + R R E D$			经典植被指数,反映植被覆盖和生长状态
归一化水体指数	(R_NIR-R_ISWIR)(R_NIR+R_ISWIR)			突出水体与植被的差异,用于水体分布和植被水分监测
优化的土壤调整植被指数	$1.16 × (R N I R - R R E D) R N I R + R R E D + 0.16$			优化土壤影响校正,提高稀疏植被区域的监测精度
红边位置指数	705+ $35 × (R R E D + R B A N D 7) / 2 - R B A N D 6 R B A N D 6 - R B A N D 5$			基于红边波段变化定位红边位置,反映植被应力和叶绿素含量动态
重归一化植被指数	$R N I R - R R E D R N I R + R R E D$			改进NDVI,增强对高植被覆盖区域的敏感性
比值植被指数	$R N I R R R E D$			近红外与红光比值,简单反映植被覆盖和健康状况
可见耐大气指数绿色	$R G R E E N - R R E D R G R E E N + R R E D$			利用绿光和红光差异,减少大气影响

Tab.3

Tab.4

Fig.1

Tab.5

Fig.2

Tab.6

Fig.3

Fig.4

References 27

[1]	SAINUDDIN F V, CHIRAKKAL S, ASOK S V, et al. A comparative study on the efficacy of dual-pol and full-pol ASAR data in radiative transfer modeling for forest above-ground biomass estimation[J]. International Journal of Remote Sensing,Taylor & Francis, 2024, 45(3):719-747.
[2]	ZHANG Junxiang, ZHOU Cui, ZHANG Gui, et al. A Novel Framework for Forest Above-Ground Biomass Inversion Using Multi-Source Remote Sensing and Deep Learning[J]. Forests,Multidisciplinary Digital Publishing Institute, 2024, 15(3):456-477.
[3]	CHEN Xinyang, YANG Keming, MA Jun, et al. Aboveground Biomass Inversion Based on Object-Oriented Classification and Pearson-mRMR-Machine Learning Model[J]. Remote Sensing,Multidisciplinary Digital Publishing Institute, 2024, 16(9):1537-1550.
[4]	ZHOU Wei, LI Haoran, XIE Lijun, et al. Remote sensing inversion of grassland aboveground biomass based on high accuracy surface modeling[J]. Ecological Indicators, 2021, 121:107215.
[5]	CUI Li, ZHANG Jiaying, DAI Yiqun, et al. Influence of BRDF Models and Solar Zenith Angles on Forest Above-Ground Biomass Derived from MODIS Multi-Angular Indices[J]. Forests,Multidisciplinary Digital Publishing Institute, 2024, 15(3):245-263.
[6]	李敏, 陈利, 李静泰, 等. 基于Sentinel-2数据的互花米草地上生物量反演[J]. 海洋环境科学, 2024, 43(3):386-397.
[7]	郝晴, 黄昌. 森林地上生物量遥感估算研究综述[J]. 植物生态学报, 2023, 47(10):1356-1374. doi: 10.17521/cjpe.2023.0008
[8]	YU Zhibo, WU Yang, ZHANG Ziyu, et al. A precise estimation framework for individual tree AGB of Pinus kesiya var.Langbianensis utilizing point cloud registration Optimization[J]. International Journal of Applied Earth Observation and Geoinformation, 2025, 140:104612.
[9]	熊向阳, 杨小周, 赵银超, 等. 基于超参数优化随机森林算法的森林生物量遥感反演[J]. 中南林业科技大学学报, 2024, 44(5):102-111.
[10]	李尉尉, 薛志泳, 朱建华, 等. 基于GEE平台多源遥感数据的海南岛红树林碳储量估算[J]. 自然资源遥感, 2025, 37(2):220-227.
[11]	WANG Chu, ZHANG Wangfei, JI Yongjie et al. Estimation of Aboveground Biomass for Different Forest Types Using Data from Sentinel-1,Sentinel-2,ALOS PALSAR-2,and GEDI[J]. Forests,Multidisciplinary Digital Publishing Institute, 2024, 15(1):215-238.
[12]	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.
[13]	熊珂, 邢元军, 和晓风, 等. 基于KNN算法的森林地上生物量遥感估测[J]. 林草资源研究, 2024(3):106-112.
[14]	LE TOAN T, QUEGAN S, DAVIDSON M W J, et al. The BIOMASS mission:mapping global forest biomass to better understand the terrestrial carbon cycle[J]. Remote Sensing of Environment, 2011, 115(11):2850-2860. doi: 10.1016/j.rse.2011.03.020
[15]	JI Yongjie, WANG Lu, ZHANG Wangfei, et al. Forest above-ground biomass estimation using X,C,L,and P band SAR polarimetric observations and different inversion models[J]. International Journal of Digital Earth,Taylor & Francis, 2024, 17(1):2310730.
[16]	SOJA M J, PERSSON H J, ULANDER L M H. Estimation of Forest Biomass From Two-Level Model Inversion of Single-Pass InSAR Data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2015, 53(9):5083-5099. doi: 10.1109/TGRS.36
[17]	BARRACHINA M, CRISTÓBAL J, TULLA A F. Estimating above-ground biomass on mountain meadows and pastures through remote sensing[J]. International Journal of Applied Earth Observation and Geoinformation, 2015, 38:184-192. doi: 10.1016/j.jag.2014.12.002
[18]	宋洁, 刘学录. 基于星载激光雷达数据的森林地上生物量估算方法比较[J]. 生态科学, 2024, 43(5):52-62.
[19]	JI S, DASHPUREV B, PHAN T N, et al. Above-ground biomass retrieval with multi-source data:Prediction and applicability analysis in Eastern Mongolia[J]. Land Degradation & Development, 2024, 35(9):2982-2992. doi: 10.1002/ldr.v35.9
[20]	李正英, 舒清态, 李圣娇, 等. 遥感影像与激光雷达数据协同反演森林生物量综述[J]. 世界林业研究, 2024, 37(6):48-53.
[21]	ZHANG Wangfei, ZHAO Lixian, LI Yun, et al. Forest Above-Ground Biomass Inversion Using Optical and SAR Images Based on a Multi-Step Feature Optimized Inversion Model[J]. Remote Sensing,Multidisciplinary Digital Publishing Institute, 2022, 14(7):1608-1624.
[22]	冯健昭, 潘永琪, 熊悦淞, 等. 基于mRMR-XGBoost的水稻关键生育期识别[J]. 农业工程学报, 2024, 40(15):111-118.
[23]	朱龙祥, 王自法, 张昕, 等. 基于GIS和多种机器学习算法的广东省森林火灾预测模型[J]. 林业工程学报, 2024, 9(3):159-167.
[24]	LIU Xiao, NEIGH C S R, PARDINI M, et al. Estimating forest height and above-ground biomass in tropical forests using P-band TomoSAR and GEDI observations[J]. International Journal of Remote Sensing,Taylor & Francis, 2024, 45(9):3129-3148.
[25]	VIÑA A, GITELSON A A. Sensitivity to Foliar Anthocyanin Content of Vegetation Indices Using Green Reflectance[J]. IEEE Geoscience and Remote Sensing Letters, 2011, 8(3):464-468. doi: 10.1109/LGRS.8859
[26]	LAI Xiahua, ZHAO Xiaomin, WANG Chen, et al. Remote sensing aboveground biomass inversion of four vegetation types in the nanji wetland[J]. Forests, 2025, 16(9):1376-1396. doi: 10.3390/f16091376
[27]	LIU Pan, REN Chunying, WANG Zongming, et al. Improving aboveground biomass estimation by integrating ICESat-2 LiDAR sampling and tree species stratification strategies using multi-source datasets[J]. Geo-spatial Information Science, 28(3):1-21.

数据源	获取时间	云量	数据量/副	数据来源
Sentinel-1	2017-07-15	<5%	2 469	https://sentinel.esa.int/web/sentinel/home
Sentinel-2	2017-07-10	<5%	3 581	https://sentinel.esa.int/web/sentinel/home

模型	部分关键参数	依据
RF^[20]	n_estimators=500 max_depth=15	扩大因子的网格搜索+ 十折交叉验证
SVM^[21]	kernel=‘RBF’ C=10 gamma=0.1	扩大因子的网格搜索+ 十折交叉验证
XGBoost^[22]	learning_rate=0.05 max_depth=8	扩大因子的网格搜索+ 十折交叉验证
KNN^[7]	n_neighbors=5 weights=‘distance’ p=2	扩大因子的网格搜索+ 十折交叉验证
Catboost^[23]	iterations=1 000 learning_rate=0.05 depth=8	扩大因子的网格搜索+ 十折交叉验证
LR^[23]	fit_intercept=True normalize=False solver=‘auto’	扩大因子的网格搜索+ 十折交叉验证

类型	预测数据	预测模型	R²	E_RMS
1	SAR	RF	0.35	6.50
		SVM	0.33	7.87
		GBR	0.31	7.47
		XGBoost	0.21	7.67
		CatBoost	0.31	8.48
		LR	0.32	6.33
2	光学	RF	0.69	4.85
		SVM	0.67	5.85
		GBR	0.68	4.39
		XGBoost	0.69	5.88
		CatBoost	0.63	2.79
		LR	0.69	5.29
3	SAR+光学	RF	0.83	2.77
		SVM	0.74	4.10
		GBR	0.73	4.28
		XGBoost	0.71	5.41
		CatBoost	0.16	5.92
		LR	0.47	4.60

类型	预测数据	预测模型	R²	E_RMS
1	SAR	RF	0.35	6.50
		SVM	0.33	7.87
		GBR	0.31	7.47
		XGBoost	0.21	7.67
		CatBoost	0.31	8.48
		LR	0.32	6.33
2	光学	RF	0.66	5.89
		SVM	0.63	6.25
		GBR	0.63	4.78
		XGBoost	0.66	5.97
		CatBoost	0.61	3.04
		LR	0.69	3.54
3	SAR+光学	RF	0.78	3.77
		SVM	0.74	4.10
		GBR	0.75	5.18
		XGBoost	0.73	6.41
		CatBoost	0.15	6.22
		LR	0.42	5.60