面向山地森林区域的植被高度预测数据集

余翠琳; 钟梓炫; 庞弘毅; 丁煜晟; 赖涛; 黄海风; 王青松

doi:10.11999/JEIT250941

面向山地森林区域的植被高度预测数据集

doi: 10.11999/JEIT250941 cstr: 32379.14.JEIT250941

1.
中山大学电子与通信工程学院深圳 518107
2.
厦门大学电子科学与技术学院厦门 361005

基金项目: 国家自然科学基金(62273365)，“小米青年学者”项目

详细信息

作者简介:
余翠琳：女，博士，博士后，研究方向为机器学习交叉应用、多源信息融合和遥感数据处理

钟梓炫：男，硕士生，研究方向为遥感数据处理

庞弘毅：男，本科生，研究方向为遥感数据处理

丁煜晟：男，硕士生，研究方向为多源遥感数据融合和处理

赖涛：男，博士，副教授，研究方向为SAR成像雷达系统设计与信息处理

黄海风：男，博士，教授，研究方向为空间电子和智能感知领域关键技术

王青松：男，博士，教授，研究方向为遥感图像精化处理、智能视觉导航、协同探测感知与信息融合

通讯作者:
王青松　wangqs5@mail.sysu.edu.cn

中图分类号: TN957.52; TP7
计量
- 文章访问数: 37
- HTML全文浏览量: 18
- PDF下载量: 5
- 被引次数: 0
出版历程
- 收稿日期: 2025-09-22
- 修回日期: 2025-10-11
- 录用日期: 2025-11-03
- 网络出版日期: 2025-11-12

Vegetation Height Prediction Dataset Oriented to Mountainous Forest Areas

1.
School of Electronics and Communication Engineering, Sun Yat-sen University, Shenzhen 518107, China
2.
School of Electronic Science and Engineering, Xiamen University, Xiamen 361005, China

Funds: The National Natural Science Foundation of China (62273365), Xiaomi Young Talents Program

摘要

摘要: 植被高度是刻画森林垂直结构、碳储量和生态系统功能的重要参数，在生态学、气候变化和生物地理等领域具有广泛应用。随着人工智能尤其是大模型技术的发展，森林生态研究对大规模、标准化训练数据的需求愈加迫切。然而，目前公开数据仍缺乏覆盖广区域、统一规范的林冠高度预测数据集，限制了先进智能方法的应用。为此，该文构建了面向山地森林区域的植被高度预测数据集(VHP-Dataset)，融合多光谱遥感影像、数字高程模型(DEM)、植被覆盖度和覆盖类型等多源数据，以全球生态系统动力学调查(GEDI)冠层高度为目标变量，形成18维输入特征。该文介绍了数据集的构建流程，并通过空间分布、模型验证和特征重要性分析等实验进行评估。结果表明，VHP-Dataset能够有效支持监督学习建模，在多地貌、多区域的植被高度预测中展现出良好的科学性与适用性，为森林结构反演提供了标准化训练样本支撑。
- 植被高度预测 /
- 监督学习 /
- 多源遥感数据 /
- 数据集构建
Abstract: Objective Vegetation height is a key ecological parameter that reflects forest vertical structure, biomass, ecosystem functions, and biodiversity. Existing open-source vegetation height datasets are often sparse, unstable, and poorly suited to mountainous forest regions, which limits their utility for large-scale modeling. This study constructs the Vegetation Height Prediction Dataset (VHP-Dataset) to provide a standardized large-scale training resource that integrates multi-source remote sensing features and supports supervised learning tasks for vegetation height estimation. Methods The VHP-Dataset is constructed by integrating Landsat 8 multispectral imagery, the digital elevation model AW3D30 (ALOS World 3D, 30 m), land cover data CGLS-LC100 (Copernicus Global Land Service, Land Cover 100 m), and tree canopy cover data GFCC30TC (Global Forest Canopy Cover 30 m Tree Canopy). Canopy height from GEDI L2A (Global Ecosystem Dynamics Investigation, Level 2A) footprints is used as the target variable. A total of 18 input features is extracted, covering spatial location, spectral reflectance, topographic structure, vegetation indices, and vegetation cover information (Table 3, Fig. 4). For model validation, five representative approaches are applied: Extremely Randomized Trees (ExtraTree), Random Forest (RF), Artificial Neural Network (ANN), Broad Learning System (BLS), and Transformer. Model performance is assessed using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Standard Deviation (SD), and Coefficient of Determination (R²). Results and Discussions The experimental results show that the VHP-Dataset supports stable vegetation height prediction across regions and terrain conditions, which reflects its scientific validity and practical applicability. Model comparisons indicate that ExtraTree achieves the best performance in most regions, and Transformer performs well in specific areas, which confirms that the dataset is compatible with different approaches (Table 5). Stratified analyses show that prediction errors increase under high canopy cover and steep slope conditions, and predictions remain more stable at higher elevations (Fig. 6～9). These findings indicate that the dataset captures the effects of complex topography and canopy structure on model accuracy. Feature importance analysis shows that spatial location, topographic factors, and canopy cover indices are the primary drivers of prediction accuracy, while spectral and land cover information provide complementary contributions (Fig. 10). Conclusions The results show that the VHP-Dataset supports vegetation height prediction across regions and terrain types, which reflects its scientific validity and applicability. The dataset enables robust predictions with traditional machine learning methods such as tree-based models, and it also provides a foundation for deep learning approaches such as Transformers, which reflects broad methodological compatibility. Stratified analyses based on vegetation cover and terrain show the effects of complex canopy structures and topographic factors on prediction accuracy, and feature importance analysis identifies spatial location, topographic attributes, and canopy cover indices as the primary drivers. Overall, the VHP-Dataset fills the gap in large-scale high-quality datasets for vegetation height prediction in mountainous forests and provides a standardized benchmark for cross-regional model evaluation and comparison. This offers value for research on vegetation height prediction and forest ecosystem monitoring.
- Vegetation height prediction /
- Supervised learning /
- Multi-source remote sensing data /
- Dataset construction

HTML全文

图 1 不同研究区域在全球地理位置分布及植被类型展示

下载: 全尺寸图片幻灯片

图 2 数据集整体构建流程

下载: 全尺寸图片幻灯片

图 3 不同研究区域筛选后GEDI数据空间分布情况

下载: 全尺寸图片幻灯片

图 4 VHP-Dataset中除经度和纬度外特征的可视化

下载: 全尺寸图片幻灯片

图 5 不同研究区域植被预测高度分布情况

下载: 全尺寸图片幻灯片

图 6 不同研究区植被覆盖率与实测植被高度统计关系

下载: 全尺寸图片幻灯片

图 7 不同分层植被覆盖率的预测误差箱型图

下载: 全尺寸图片幻灯片

图 8 不同坡度范围预测误差箱型图

下载: 全尺寸图片幻灯片

图 9 不同高程范围预测误差箱型图

下载: 全尺寸图片幻灯片

图 10 VHP-Dataset特征重要性得分柱状图

下载: 全尺寸图片幻灯片

表 1 研究数据基本属性

产品名称	覆盖范围	坐标系	数据类型	用途
GEDI L2A	51.6°N$ \sim $51.6°S	WGS84	激光雷达光斑观测产品	提供冠层高度、地面高程及垂直结构参数，作为目标变量。
Landsat 8	全球	WGS84	多光谱与热红外遥感影像	提取光谱反射波段和归一化指数，用于表征植被生长状态。
AW3D30	84°N$ \sim $84°S	WGS84	DEM产品	提供高程和坡度等地形特征，支持复杂地形条件下建模。
CGLS-LC100	全球	WGS84	全球土地覆盖产品	提供地表覆盖类型，辅助冠层高度差异化建模。
GFCC30TC	全球	WGS84	森林覆盖度数据	提供0%$ \sim $100%植被覆盖百分比，刻画森林密度和连续性。

下载: 导出CSV

表 2 GEDI L2A有效光斑筛选条件

筛选条件	说明
sensitivity>0.9	Sensitivity为信号灵敏度，为获取质量较高的光斑数据，通常在陆地区域采用不小于0.9的灵敏度阈值，因此剔除了灵敏度低于0.9的光斑。
quality_flag=1	quality_flag为质量标记值，当质量标记值为“1”时，表明该波形在能量、灵敏度、振幅以及实时地表跟踪质量等方面达到了设定标准，可视为有效信号数据。
degrade_flag=0	degrade_flag为退化状态，当状态退化标志为“1”时，说明指向或定位精度下降。为确保数据质量，剔除了标志值为1的光斑，仅保留标志为0的有效光斑数据。
beam>4	beam表示光束，GEDI L2A数据共包含8个波束通道，其中第1至第4波束为覆盖波束，第5$ \sim $8波束为全功率波束。相较之下，覆盖波束在穿透高密度森林冠层时存在一定局限，仅能在树冠覆盖率不超过95%的林地条件下实现有效穿透。因此，为提升数据穿透能力与精度，通常优先采用全功率波束进行分析。
Solar_elevation<0	Solar_elevation是太阳高度角，表示太阳位于地平线以下，代表获取夜间的GEDI数据。

下载: 导出CSV

表 3 数据集特征与目标变量

数据源	类型	特征	目标变量
GEDI L2A	空间信息	经度和纬度	RH95 冠层高度
Landsat 8	光谱波段	Band 2，Band 3，Band 4，Band 5，Band 6，Band 7
Landsat 8	归一化指数	NDVI, NDWI, NDII, NDGI, NBR
AW3D30	地形结构	高程，坡度，和坡向
CGLS-LC100	植被类型	植被覆盖类型
GFCC30TC	植被指数	植被覆盖指数

下载: 导出CSV

表 4 Landsat-8光谱波段基本信息

波段	名称	波长(μm)	分辨率(m)	作用
Band 2	蓝光波段	0.450$ \sim $0.515	30	用于水体识别和大气校正。
Band 3	绿光波段	0.525$ \sim $0.600	30	用于植被监测和土地覆盖分类等。
Band 4	红光波段	0.630$ \sim $0.680	30	用于植被反射和农作物健康监测等。
Band 5	近红外波段	0.845$ \sim $0.885	30	用于对植被高度敏感，用于植被指数计算。
Band 6	短波红外1	1.560$ \sim $1.660	30	用于土壤湿度、云识别和地质调查。
Band 7	短波红外2	2.100$ \sim $2.300	30	用于地表特征分类、植被区分和热异常检测。

下载: 导出CSV

表 5 VHP-Dataset在五种代表性方法下的评估结果

研究区域	方法	评估指标
研究区域	方法	RMSE(m)	MAE(m)	SD(m)	R²
美国GP森林	Transformer	8.114	6.249	8.065	0.594
	ANN	8.399	6.566	8.384	0.572
	RF	9.036	7.215	8.984	0.508
	BLS	8.512	6.683	8.512	0.557
	ExtraTree	8.116	6.263	3.335	0.597
美国GWJ-M森林	Transformer	6.443	4.971	6.398	0.409
	ANN	6.761	5.238	6.695	0.350
	RF	6.815	5.325	6.735	0.341
	BLS	6.616	5.139	6.616	0.375
	ExtraTree	6.351	4.883	5.311	0.417
新西兰ENP森林	Transformer	5.866	4.347	4.041	0.308
	ANN	5.430	4.403	5.332	0.321
	RF	4.852	3.850	4.534	0.460
	BLS	4.852	3.760	4.852	0.441
	ExtraTree	4.755	3.731	2.931	0.482
德国TF森林	Transformer	7.872	5.709	4.743	0.478
	ANN	7.002	5.424	6.888	0.537
	RF	6.595	5.114	5.942	0.584
	BLS	6.460	4.864	6.460	0.601
	ExtraTree	6.217	4.606	5.166	0.638

下载: 导出CSV

参考文献(27)

[1]	MALAMBO L and POPESCU S. Mapping vegetation canopy height across the contiguous United States using ICESat-2 and ancillary datasets[J]. Remote Sensing of Environment, 2024, 309: 114226. doi: 10.1016/j.rse.2024.114226.
[2]	ADRAH E, WONG J P, and YIN He. Integrating GEDI, Sentinel-2, and Sentinel-1 imagery for tree crops mapping[J]. Remote Sensing of Environment, 2025, 319: 114644. doi: 10.1016/j.rse.2025.114644.
[3]	CHANG Bingtao, XIONG Hao, LI Yuan, et al. ALCSF: An adaptive and anti-noise filtering method for extracting ground and top of canopy from ICESat-2 LiDAR data along single tracks[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2024, 215: 80–98. doi: 10.1016/j.isprsjprs.2024.07.002.
[4]	BEAUDOIN A, BERNIER P Y, VILLEMAIRE P, et al. Tracking forest attributes across Canada between 2001 and 2011 using a k nearest neighbors mapping approach applied to MODIS imagery[J]. Canadian Journal of Forest Research, 2018, 48(1): 85–93. doi: 10.1139/cjfr-2017-0184.
[5]	PICKSTONE B J, GRAHAM H A, and CUNLIFFE A M. Estimating canopy height in tropical forests: Integrating airborne LiDAR and multi-spectral optical data with machine learning[J]. Sustainable Environment, 2025, 11(1): 2469406. doi: 10.1080/27658511.2025.2469406.
[6]	BENHALIMA N, OUARZEDDINE M, SOUISSI B, et al. Integrating PolInSAR and GEDI data with machine learning for forest canopy height predicting in Pongara National Park, Gabon[J]. International Journal of Remote Sensing, 2025, 46(18): 6875–6896. doi: 10.1080/01431161.2025.2549131.
[7]	POURSHAMSI M, XIA Junshi, YOKOYA N, et al. Tropical forest canopy height estimation from combined polarimetric SAR and LiDAR using machine-learning[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2021, 172: 79–94. doi: 10.1016/j.isprsjprs.2020.11.008.
[8]	LANG N, KALISCHEK N, ARMSTON J, et al. Global canopy height regression and uncertainty estimation from GEDI LIDAR waveforms with deep ensembles[J]. Remote Sensing of Environment, 2022, 268: 112760. doi: 10.1016/j.rse.2021.112760.
[9]	TOLAN J, YANG H I, NOSARZEWSKI B, et al. Very high resolution canopy height maps from RGB imagery using self-supervised vision transformer and convolutional decoder trained on aerial lidar[J]. Remote Sensing of Environment, 2024, 300: 113888. doi: 10.1016/j.rse.2023.113888.
[10]	LANG N, JETZ W, SCHINDLER K, et al. A high-resolution canopy height model of the Earth[J]. Nature Ecology & Evolution, 2023, 7(11): 1778–1789. doi: 10.1038/s41559-023-02206-6.
[11]	LEI Yuqi, WANG Yuanjia, WANG Guilong, et al. Estimating forest canopy height based on GEDI lidar data and multi-source remote sensing images[C]. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Changsha, China, 2024: 297–303. doi: 10.5194/isprs-archives-XLVIII-1-2024-297-2024.
[12]	钱亚冠, 孔亚鑫, 陈科成, 等. 利用频谱衰减增强深度神经网络对抗迁移攻击[J]. 电子与信息学报, 2025, 47(10): 3847–3857. doi: 10.11999/JEIT250157. QIAN Yaguan, KONG Yaxin, CHEN Kecheng, et al. Adversarial transferability attack on deep neural networks through spectral coefficient decay[J]. Journal of Electronics & Information Technology, 2025, 47(10): 3847–3857. doi: 10.11999/JEIT250157.
[13]	DEMIR I, KOPERSKI K, LINDENBAUM D, et al. DeepGlobe 2018: A challenge to parse the earth through satellite images[C]. The IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, Salt Lake City, USA, 2018: 172–181. doi: 10.1109/CVPRW.2018.00031.
[14]	CHEN Hao and SHI Zhenwei. A spatial-temporal attention-based method and a new dataset for remote sensing image change detection[J]. Remote Sensing, 2020, 12(10): 1662. doi: 10.3390/rs12101662.
[15]	BURNS P, HAKKENBERG C R, and GOETZ S J. Multi-resolution gridded maps of vegetation structure from GEDI[J]. Scientific Data, 2024, 11(1): 881. doi: 10.1038/s41597-024-03668-4.
[16]	MIURA Y, SHAMSUDDUHA M, SUPPASRI A, et al. A global multi-sensor dataset of surface water indices from landsat-8 and sentinel-2 satellite measurements[J]. Scientific Data, 2025, 12(1): 1253. doi: 10.1038/s41597-025-05562-z.
[17]	HUANG Huabing, CHEN Peimin, XU Xiaoqing, et al. Estimating building height in China from ALOS AW3D30[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2022, 185: 146–157. doi: 10.1016/j.isprsjprs.2022.01.022.
[18]	XU Yidi, YU Le, FENG Duole, et al. Comparisons of three recent moderate resolution African land cover datasets: CGLS-LC100, ESA-S2-LC20, and FROM-GLC-Africa30[J]. International Journal of Remote Sensing, 2019, 40(16): 6185–6202. doi: 10.1080/01431161.2019.1587207.
[19]	TOWNSHEND J. Global Forest cover change (GFCC) tree cover multi-year global 30 m V003[R]. 2016. doi: 10.5067/MEaSUREs/GFCC/GFCC30TC.003.
[20]	LIU Xiaoqiang, SU Yanjun, HU Tianyu, et al. Neural network guided interpolation for mapping canopy height of China’s forests by integrating GEDI and ICESat-2 data[J]. Remote Sensing of Environment, 2022, 269: 112844. doi: 10.1016/j.rse.2021.112844.
[21]	DUNCANSON L, NEUENSCHWANDER A, HANCOCK S, et al. Biomass estimation from simulated GEDI, ICESat-2 and NISAR across environmental gradients in Sonoma County, California[J]. Remote Sensing of Environment, 2020, 242: 111779. doi: 10.1016/j.rse.2020.111779.
[22]	LIU Aobo, CHENG Xiao, and CHEN Zhuoqi. Performance evaluation of GEDI and ICESat-2 laser altimeter data for terrain and canopy height retrievals[J]. Remote Sensing of Environment, 2021, 264: 112571. doi: 10.1016/j.rse.2021.112571.
[23]	ARSHAD T and ZHANG Junping. Hierarchical attention transformer for hyperspectral image classification[J]. IEEE Geoscience and Remote Sensing Letters, 2024, 21: 5504605. doi: 10.1109/LGRS.2024.3379509.
[24]	KURANI A, DOSHI P, VAKHARIA A, et al. A comprehensive comparative study of artificial neural network (ANN) and support vector machines (SVM) on stock forecasting[J]. Annals of Data Science, 2023, 10(1): 183–208. doi: 10.1007/s40745-021-00344-x.
[25]	SALMAN H A, KALAKECH A, and STEITI A. Random forest algorithm overview[J]. Babylonian Journal of Machine Learning, 2024, 2024: 69–79. doi: 10.58496/BJML/2024/007.
[26]	DUAN Junwei. Broadfusion: A novel two-stage multifocus image fusion approach with human visual system embedded broad learning system[J]. Knowledge-Based Systems, 2025, 326: 114030. doi: 10.1016/j.knosys.2025.114030.
[27]	SAMAT A, PERSELLO C, LIU Sicong, et al. Classification of VHR multispectral images using extratrees and maximally stable extremal region-guided morphological profile[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2018, 11(9): 3179–3195. doi: 10.1109/JSTARS.2018.2824354.