深度学习方法下GEDI数据的天然云杉林地上生物量反演

doi:10.13866/j.azr.2023.09.11

干旱区研究 ›› 2023, Vol. 40 ›› Issue (9): 1472-1483.doi: 10.13866/j.azr.2023.09.11 cstr: 32277.14.j.azr.2023.09.11

深度学习方法下GEDI数据的天然云杉林地上生物量反演

孙丹阳^1,²(),魏建新^1,^2,³(),杨辽⁴,王杰⁵,唐宇琪⁴,巴比尔江·迪力夏提^2,³

1.新疆大学地理与遥感科学学院，新疆乌鲁木齐 830017
2.新疆激光雷达应用工程技术研究中心，新疆乌鲁木齐 830002
3.新疆维吾尔自治区自然资源信息中心，新疆乌鲁木齐 830002
4.中国科学院新疆生态与地理研究所，荒漠与绿洲生态国家重点实验室，新疆乌鲁木齐 830011
5.西华师范大学地理科学学院，四川南充 637002

收稿日期:2023-04-01 修回日期:2023-05-16 出版日期:2023-09-15 发布日期:2023-09-28
作者简介:孙丹阳（1998-），女，硕士研究生，研究方向为地信工程. E-mail: 1063209224@qq.com
基金资助:
新疆维吾尔自治区财政拨款项目(65000022P00790610001Q)

Study on using deep learning method to retrieve the biomass of natural Picea forest from GEDI data

SUN Danyang^1,²(),WEI Jianxin^1,^2,³(),YANG Liao⁴,WANG Jie⁵,TANG Yuqi⁴,Babierjiang DILIXIATI^2,³

1. College of Geography and Remote Sensing Sciences, Xinjiang University, Urumqi 830017, Xinjiang, China
2. Xinjiang LiDAR Application Engineering Technology Research Center, Urumqi 830002, Xinjiang, China
3. Xinjiang Uygur Autonomous Regions Natural Resources Information Center, Urumqi 830002, Xinjiang, China
4. State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, Urumqi 830011, Xinjiang, China
5. School of Geographical Sciences, China West Normal University, Nanchong 637002, Sichuan, China

Received:2023-04-01 Revised:2023-05-16 Published:2023-09-15 Online:2023-09-28

摘要/Abstract

摘要：

森林作为陆地最大碳库，对人类的生活与发展至关重要，精准掌握森林资源动态变化并对其进行现代化可持续发展已成为当下研究热点。本文以天山山脉的天然云杉林为研究对象，利用地面实测数据、直升机机载激光雷达点云数据以及全球生态系统动态调查激光雷达（Global Ecosystem Dynamics Investigation，GEDI）数据，构建多源融合数据框架，通过使用AutoKeras框架下的深度学习算法，实现GEDI数据的多个相对高度百分位数（Relative Height Percentile，RH）与其光斑内地上生物量的回归模型预测，验证GEDI数据在较大范围的地上生物量反演方面的可行性，主要结论如下：（1） GEDI数据用于森林地上生物量估测研究具有较高可行性，通过自动化深度学习算法，训练集、验证集、整体数据的决定系数（Coefficient of Determination，R²）分别为0.69、0.63和0.67，平均绝对误差（Mean Absolute Error，MAE）分为3.73 mg·hm^-2、4.22 mg·hm^-2和3.89 mg·hm^-2，具有较高的预测精度。（2）直升机激光雷达作为GEDI数据估算地上生物量的中间技术，整个研究区内的单木识别准确率高于0.75。最终本次研究通过多模态数据融合，定量化描述研究区单木基础结构参数的同时，验证GEDI数据在获取森林地上生物量方面的潜力，也为相近区域大面积的森林碳源汇、生物量、蓄积量估算、森林管理与经营、生物多样性保护等多个项目研究提供理论基础，具有一定的指导意义和基础数据支撑作用。

关键词: 天然云杉林, GEDI, LiDAR, 地上生物量, 深度学习

Abstract:

As the largest carbon reservoir on land, forests play a crucial role in human life and development. Understanding the dynamic changes in forest resources and modernizing their sustainable development is currently a significant research focus. This study focuses on natural Picea forests in the Tianshan Mountains and uses ground measurement data, helicopter airborne LiDAR point cloud data, and Global Ecosystem Dynamics Investigation (GEDI) data to construct a multisource fusion data framework. By utilizing deep learning algorithms within the AutoKeras framework, the study aims to predict the regression model of multiple relative height quantiles of GEDI data and their aboveground biomass in the study area, thereby validating the feasibility of GEDI data for large-scale aboveground biomass retrieval. The main conclusions are as follows: (1) GEDI data are highly feasible for estimating forest aboveground biomass. Through automated deep learning algorithms and training and verification sets, the overall data achieve a coefficient of determination (R²) of 0.69, 0.63, and 0.67, respectively, along with a mean absolute error of 3.73 mg·hm^-2, 4.22 mg·hm^-2, and 3.89 mg·hm^-2, demonstrating high prediction accuracy. (2) Helicopter LiDAR, an intermediate technology for estimating aboveground biomass using GEDI data, exhibits a single tree recognition accuracy of over 0.75 across the study area. The study successfully utilizes multimodal data fusion to quantitatively describe the structural parameters of the single tree foundation in the study area while verifying the potential of GEDI data for obtaining forest aboveground biomass. Moreover, the study provides a theoretical basis for estimating carbon sources and sinks, biomass, stock, forest management, biodiversity protection, and other projects in similar areas, offering essential guidance, and fundamental data support.

Key words: natural Picea forest, GEDI, LiDAR, aboveground biomass, deep learning

孙丹阳,魏建新,杨辽,王杰,唐宇琪,巴比尔江·迪力夏提. 深度学习方法下GEDI数据的天然云杉林地上生物量反演[J]. 干旱区研究, 2023, 40(9): 1472-1483.

SUN Danyang,WEI Jianxin,YANG Liao,WANG Jie,TANG Yuqi,Babierjiang DILIXIATI. Study on using deep learning method to retrieve the biomass of natural Picea forest from GEDI data[J]. Arid Zone Research, 2023, 40(9): 1472-1483.

图/表 13

图1

图2

图3

图4

图5

图6

图7

表1

图8

图9

表2

表3

图10

参考文献 34

[1]	Olson, Jerry S, Watts J A, et al. Carbon in Live Vegetation of Major World Ecosystems[M]. Teen: ORNL-5862, 1983: 15-25.
[2]	Hua Dengxin, Song Xiaoquan. Advances in lidar remote sensing techniques[J]. Infrared and Laser Engineering, 2008, 37(1): 21-27.
[3]	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, 41(3): 129-136.
	[Pang Yong, Li Zengyuan, Chen Erxue, et al. Lidar remote sensing technology and its application in forestry[J]. Scientia Silvae Sinicae, 2005, 41(3): 129-136.]
[4]	李增元, 刘清旺, 庞勇. 激光雷达森林参数反演研究进展[J]. 遥感学报, 2016, 20(5): 1138-1150.
	[Li Zengyuan, Liu Qingwang, Pang Yong. Review on forest parameters inversion using LiDAR[J]. Journal of Remote Sensing, 2016, 20(5): 1138-1150.]
[5]	胡国军, 方勇, 张丽. 星载激光雷达的发展与测绘应用前景分析[J]. 测绘技术装备, 2015, 17(2): 34-37.
	[Hu Guojun, Fang Yong, Zhang Li. Development of spaceborne lidar and prospect analysis of surveying and mapping applications[J]. Geomatics Technology and Equipment, 2015, 17(2): 34-37.]
[6]	谢栋平, 李国元, 赵严铭, 等. 美国GEDI天基激光测高系统及其应用[J]. 国际太空, 2018, 40(12): 39-44.
	[Xie Dongping, Li Guoyuan, Zhao Yanming, et al. GEDI space-based laser altimeter system and its application in the United State[J]. Space International, 2018, 40(12): 39-44.]
[7]	Adam M, Urbazaev M, Dubois C, et al. Accuracy assessment of GEDI terrain elevation and canopy height estimates in European temperate forests: Influence of environmental and acquisition parameters[J]. Remote Sensing, 2020, 12(23): 3948. doi: 10.3390/rs12233948
[8]	韩明辉. 基于星载激光雷达GEDI数据反演森林结构参数的研究[D]. 哈尔滨: 东北林业大学, 2022.
	[Han Minghui. Study on Forest Structure Parameters Inversion based on GEDI Data[D]. Harbin: Northeast Forestry University, 2022.]
[9]	Alireza H, Cheikh M, Annika K, et al. Deep learning for forest inventory and planning: A critical review on the remote sensing approaches so far and prospects for further applications[J]. Forestry: An International Journal of Forest Research, 2022, 95(4): 451-465. doi: 10.1093/forestry/cpac002
[10]	Cao Lin, She Guanghui, Dai Jinsong, et al. Status and prospects of the lidar-based forest biomass estimation[J]. Journal of Nanjing Forestry University, 2013, 56(3): 163-169.
[11]	Torre-Tojal L, Bastarrika A, Boyano A, et al. Above-ground biomass estimation from lidar data using random forest algorithms[J]. Journal of Computational Science, 2022, 58: 101517. doi: 10.1016/j.jocs.2021.101517
[12]	Quirós E, Polo M E, Fragoso-Campón L. GEDI elevation accuracy assessment: A case study of southwest Spain[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 5285-5299. doi: 10.1109/JSTARS.2021.3080711
[13]	Fayad I, Baghdadi N N, Alvares C A, et al. Assessment of Gedi’s lidar data for the estimation of canopy heights and wood volume of eucalyptus plantations in Brazil[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 7095-7110. doi: 10.1109/JSTARS.2021.3092836
[14]	Fayad I, Baghdadi N N, Alvares C A, et al. Terrain slope effect on forest height and wood volume estimation from GEDI data[J]. Remote Sensing, 2021, 13(11): 2136. doi: 10.3390/rs13112136
[15]	Healey S P, Yang Z, Gorelick N, et al. Highly local model calibration with a new GEDI lidar asset on Google Earth Engine reduces Landsat forest height signal saturation[J]. Remote Sensing, 2020, 12(17): 2840. doi: 10.3390/rs12172840
[16]	Fayad I, Baghdadi N, Bailly J S, et al. Analysis of GEDI elevation data accuracy for inland waterbodies altimetry[J]. Remote Sensing, 2020, 12(17): 2714. doi: 10.3390/rs12172714
[17]	Potapov P, Li X, Hernandez-Serna A, et al. Mapping global forest canopy height through integration of GEDI and Landsat data[J]. Remote Sensing of Environment, 2020, 253: 112165. doi: 10.1016/j.rse.2020.112165
[18]	林晓娟. 基于ICESat-2和GEDI森林冠层高度和森林地上生物量遥感诊断[D]. 北京: 中国科学院大学, 2021.
	[Lin Xiaojuan. Remote Sensing Diagnosis of Forest Canopy Height and Forest Aboveground Biomass based on ICESat-2 and GEDI[D]. Beijing: University of Chinese Academy of Sciences, 2021.]
[19]	朱笑笑. 基于ICESat-2和GEDI数据的中国30米分辨率森林高度反演研究[D]. 北京: 中国科学院大学, 2021.
	[Zhu Xiaoxiao. Forest Height Retrieval of China with a Resolution of 30 m Using ICESat-2 and GEDI Data[D]. Beijing: University of Chinese Academy of Sciences, 2021.]
[20]	Ngo Y N, Ho Tong Minh D, Baghdadi N, et al. Tropical forest top height by GEDI: From sparse coverage to continuous data[J]. Remote Sensing, 2023, 15(4): 975. doi: 10.3390/rs15040975
[21]	Schneider F D, Ferraz A, Hancock S, et al. Towards mapping the diversity of canopy structure from space with GEDI[J]. Environmental Research Letters, 2020, 15(11): 115006. doi: 10.1088/1748-9326/ab9e99
[22]	Roy D P, Kashongwe H B, Armston J. The Impact of geolocation uncertainty on GEDI tropical forest canopy height estimation and change monitoring[J]. Science of Remote Sensing, 2021, 4: 100024. doi: 10.1016/j.srs.2021.100024
[23]	Bauer L, Knapp N, Fischer R. Mapping Amazon forest productivity by fusing GEDI lidar waveforms with an individual-based forest model[J]. Remote Sensing, 2021, 13(22): 4540. doi: 10.3390/rs13224540
[24]	李明辉, 何风华, 刘云, 等. 天山云杉种群空间格局与动态[J]. 生态学报, 2005, 25(5): 1000-1006.
	[Li Minghui, He Fenghua, Liu Yun, et al. Spatial distribution pattern of tree individuals in the Schrenk spruce forest, Northwest China[J]. Acta Ecologica Sinica, 2005, 25(5): 1000-1006.]
[25]	王雅佩, 王振锡, 刘梦婷, 等. 基于无人机影像天山云杉林主伐迹地提取研究[J]. 新疆农业科学, 2019, 56(7): 1312-1324.
	[Wang Yapei, Wang Zhenxi, Liu Mengting, et al. Research on extraction of final felling area of Picea schrenkiana var tianshanica based on UAV image[J]. Xinjiang Agricultural Sciences, 2019, 56(7): 1312-1324.]
[26]	Soille P. Morphological Image Analysis-Principles and Applications[M]. Berlin: Springer-Verlag, 2003.
[27]	Dubayah R, Hofton M, Blair M J B, et al. GEDI L2A elevation and height metrics data global footprint level V001[DB/OL]. NASA EOSDIS Land Processes Distributed Active Archive Center, 2023-08-15, https://doi.org/10.5067/GEDI/GEDI02_A.001.
[28]	Dubayah R, Blair J B, Goetz S, et al. The global ecosystem dynamics investigation: High-resolution laser ranging of the earth’s forests and topography[J]. Science of Remote Sensing, 2020, 1: 100002. doi: 10.1016/j.srs.2020.100002
[29]	张绘芳, 高亚琪, 朱雅丽, 等. 新疆雪岭杉生物量模型对比研究[J]. 西北林学院学报, 2015, 30(6): 52-58.
	[Zhang Huifang, Gao Yaqi, Zhu Yali, et al. A comparative study on biomass models for Picea schrenkiana in Xinjiang[J]. Journal of Northwest Forestry University, 2015, 30(6): 52-58.]
[30]	Li Wenkai, Guo Qinghua, Jakubowski M K, et al. A new method for segmenting individual trees from the lidar point cloud[J]. Photogrammetric Engineering and Remote Sensing, 2012, 78: 75-84. doi: 10.14358/PERS.78.1.75
[31]	Hofton M, Blair J B, Story S, et al. Algorithm Theoretical Basis Document(ATBD) for GEDI transmit and receive waveform processing for L1 and L2 products[EB/OL]. Goddard Space Flight Centre, 2019. https://lpdaac.usgs.gov/documents/581/GEDI_WF_ATBD_v1.0.pdf.
[32]	Rishmawi K, Huang C, Schleeweis K, et al. Integration of VIIRS observations with GEDI-lidar measurements to monitor forest structure dynamics from 2013 to 2020 across the conterminous United States[J]. Remote Sensing, 2022, 14(10): 2320. doi: 10.3390/rs14102320
[33]	胡涛. 基于多源数据的孟家岗林场针叶林蓄积量估测研究[D]. 哈尔滨: 东北林业大学, 2022.
	[Hu Tao. Estimation of Coniferous Forest Volume in Mengjiagang Forest Farm Based on Multi-source data[D]. Harbin: Northeast Forestry University, 2022.]
[34]	Knapp N, Huth A, Fischer R. Tree crowns cause border effects in area-based biomass estimations from remote sensing[J]. Remote Sensing, 2021, 13(8): 1592. doi: 10.3390/rs13081592

网络层	输出大小	参数个数
输入	[（None，81）]	0
多类别编码层	（None，81）	0
归一化层	（None，81）	163
全连接层	（None，64）	5248
激活函数1	（None，64）	0
全连接层	（None，1024）	66560
激活函数2	（None，1024）	0
回归层	（None，1）	1025

评价指标	召回率（R）/%	准确度（P）/%	F值/%
样地1	67.40	93.93	78.48
样地2	78.57	96.70	82.24
样地3	89.40	84.29	86.77
平均值	78.46	88.16	82.50

深度学习方法下GEDI数据的天然云杉林地上生物量反演

Study on using deep learning method to retrieve the biomass of natural Picea forest from GEDI data

RichHTML

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 34

相关文章 7

Metrics

本文评价

推荐阅读 0

模型评价指标	数据集	R²	RMSE /(mg·hm^-2)	MAE /(mg·hm^-2)
深度学习（DL）	训练集	0.69	4.68	3.73
	验证集	0.63	5.31	4.22
	总体数据	0.67	4.88	3.89

[1]	戚曌, 闫峰, 席磊, 曹晓明, 邹佳秀, 冯益明. 鄂尔多斯高原砒砂岩区植被恢复潜力[J]. 干旱区研究, 2024, 41(9): 1583-1592.
[2]	邱春霞, 刘晓宏, 李豆, 张佳淼, 李朋飞. 机载LiDAR和模糊推理系统在黄土高原土壤侵蚀监测中的应用[J]. 干旱区研究, 2024, 41(8): 1331-1342.
[3]	李若楠, 李均力, 李爽爽, 刘帅琪, 都伟冰. 基于Sentinel-2的依连哈比尔尕冰川变化监测[J]. 干旱区研究, 2024, 41(6): 940-950.
[4]	梁博明, 刘新, 郝媛媛, 楚彬, 唐庄生. 基于5种植被指数的荒漠区植被生物量提取研究[J]. 干旱区研究, 2023, 40(4): 647-654.
[5]	王娇娇, 尹小君, 刘陕南, 王帝盟. 基于深度学习的玛纳斯土地利用时空格局变化与预测[J]. 干旱区研究, 2023, 40(1): 69-77.
[6]	赵生龙, 左小安, 张铜会, 吕朋, 岳平, 张晶. 乌拉特荒漠草原群落物种多样性和生物量关系对放牧强度的响应[J]. 干旱区研究, 2020, 37(1): 168-177.
[7]	李成秀, 李小雁, 杨太保, 李岳坦. 青海湖流域沙柳河草甸群落结构与数量特征[J]. 干旱区研究, 2013, 30(6): 1028-1035.