IMHE OpenIR
ICEsat与Landsat协同的山地森林地上生物量估算研究
Alternative TitleEstimation of Mountain Forest Aboveground Biomass by Integrating ICEsat and Landsat Data
卢学辉
Subtype博士
Thesis Advisor李爱农
2017
Degree Grantor中国科学院大学
Place of Conferral北京
Degree Discipline自然地理
KeywordGLAS Landsat 冠层高度 森林生物量 三维植被激光雷达模型 地形
Abstract森林是陆地生态系统的重要组成部分,在陆地碳循环和全球气候变化中起着至关重要的作用。森林生物量是衡量森林生产力的重要指标,准确的估算森生物量对于陆地碳循环和全球气候变化研究具有重要意义。遥感技术在估测森林生物量方面有广泛的应用。然而光学信号不具有穿透性只能获取冠层的水平分布信息,在反演森林垂直结构参数方面存在诸多局限。微波信号虽然具有一定的穿透力,但是在反演树高及生物量等参数时仍然存在着信号饱和散射机制复杂建模困难等问题。激光雷达对森林冠层具有很强的穿透力,在获取植被尤其是森林空间结构参数方面具有显著的优势。但是机载的小光斑激光雷达存在成本高、数据量大、覆盖范围有限等缺点。大光斑激光雷达系统能够获取光斑覆盖范围内森林空间结构信息,在反演区域森林树高、生物量和其他植被结构参数方面更有优势。本文选择贡嘎山地区作为研究,探讨了ICEsat与Landsat协同的山地森林冠层高度与地上生物量估算方法。文中首先基于三维植被激光雷达模型模拟了地形坡度对GLAS波形及树高反演精度的影响;然后采用粒子群-最小二乘法对GLAS波形进行了高斯分解,提取了GLAS激光点上的森林冠层高度;针对GLAS系统不具备成像能力的特点,融合了Landsat卫星数据建立了区域尺度的森林冠层高度反演模型;最终,讨论了联合区域森林冠层高度反演结果与光学遥感信息的森林生物量估算方法。研究结果表明,联合ICEsat激光雷达数据与Landsat多光谱数据,可以充分发挥激光雷达与光学遥感方式各自的优势,提高区域森林冠层高度与生物量反演的精度。本文主要结论如下:1、采用三维植被激光雷达模型模拟分析了坡度对不同密度与树高的林分的GLAS波形及树高反演精度的影响,波形模拟的结果表明:随着坡度的增加,冠层与地面回波的峰值降低,两者逐渐扩张融合。虽然整体波形同时向两端延伸,但是波形向信号开始方向的延伸更为显著;坡度从0°增加到10°时,冠层峰幅度下降约40%至60%,而地面峰幅度下降超过80%。冠层与地面回波完全融合时关键坡度角与冠层高度成正比,与林分密度成反比。文中模拟光斑尺寸为65m时九种林分样地的关键坡度角的范围是9°–20°。波形长度和地形坡度之间存在着近线性关系,随着坡度的增大,波形长度将会被低估。四分位能量高度值中,RH50稳定性最好,是估算冠层高度的最佳参数。在冠层高度基本一致时,前缘高度是表征地形坡度的良好指标,但林分密度较低、坡度较小的情况下前缘长度容易出现高估现象,而后缘长度在坡度较小时极不稳定,且其同时包含了冠层高度和地形坡度信息。而采用lead10和trail10描述GLAS波形边缘长度可以在一定程度上消除指标计算过程中的高估或低估的现象。采用直接估算反演冠层高度时,冠层高度误差与地形坡度基本成正比关系,当坡度为10°时,平均估算误差约为5.4 m,而坡度为15°平均估算误差达到9.4 m,因此直接估算法一般适用于坡度小于10°的地区。坡度校正法估算冠层高度时容易出现低估现象,冠层高度误差与林分密度和坡度都相关,林分密度越大、坡度越大时,森林冠层高度低估现象越严重。2、设计了一种基于粒子群-最小二乘法的GLAS波形分解算法,采用该方法对贡嘎山地区缓坡地区(坡度小于10°地区)GLAS波形数据进行了高斯分解和树高反演。解决了最小二乘法对GLAS波形进行高斯分解时存在的初值敏感性问题,同时保证了算法具有很高的收敛可靠性。波形分解结果显示绝大部分激光点拟合波形与实际回波波形吻合良好。树高估算结果与采用统计模型估算的树高非常接近(R2=0.974,RMSE=2.86m)。3、在进行GLAS冠层高度反演的基础上,联合TM光谱数据及其衍生的植被指数数据和地形数据,分别采用多元线性回归、主成分分析和随机森林方法构建冠层高度反演模型。三个模型的相关系数分别为0.588、0.559和0.669,通过模型交叉验证分析,随机森林模型是无偏模型,其预测能力更好,因此本文采用了随机森林模型估算整个研究区的森林冠层高度,并基于野外实测树高对演结果进行验证,实测树高与预测树高的相关系数为0.636,RMSE为9.35 m。空间推广的GLAS树高存在系统的高估现象,针对这一问题采用GLAS树高与野外实测树高之间存在线性关系式对反演树高进行校正。4、基于GLAS最大冠层高度与野外实测样地生物量数据,分别建立包含全部样地数据的混合模型和按照针叶林、阔叶林分类的估算模型。建模结果表明幂模型更能够表达树高-生物量之间关系,而分类模型预测能力优于混合模型。针叶林幂模型模型的决定系数R2达到了0.601,RMSE为36.13 t/ha;阔叶林幂模型的决定系数R2为0.244,均方根误差为20.84 t/ha,表明冠层高度对森林地上生物量具有重要的预测能力。而后,融合 GLAS 估算的最大冠层高度、TM 波段反射率、植被指数,利用多元线性回归(MLR)方法建立森林地上生物量反演模型。与树高-生物量模型相比,模型的相关系数有了较大的提高。对比阔叶林两种生物量反演模型发现,基于树高-生物模型反演阔叶林生物量空间分布趋势较为合理。研究表明基于GLAS树高空间分布结果,采用树高-生物量模型对森林生物量进行推广的方法可获取区域内合理生物量分布结果。 
Other AbstractAs an important part of terrestrial ecosystem, forest plays an important role in terrestrial carbon cycle and global climate change. Forest biomass is an important indicator to measure the ecosystem productivity. Accurate estimates of forest biomass is of great significance for the study of terrestrial carbon cycle and global climate change. In recent years, remote sensing technology has been extensively used in the estimation of forest biomass. The optical signal which has little penetrability can only reflects horizontal distribution information, so it has many limitations in the inversion of forest vertical structure parameters. Synthetic Aperture Radar (SAR) can penetrate the forest canopy to a certain degree, but its scattering mechanism is complex. As a new active remote sensing, lidar (Light Detect and Ranging) has a strong penetrating ability to the forest canopy, and has a significant advantage in obtaining the spatial structure parameters of vegetation. But small-footprint lidar are typically inappropriate for the large-scale forest inventory due to large data volumes and costs. The large-footprint lidar can obtain the information of forest spatial structure within its footprint, which makes it has more advantages in the inversion of regional forest spatial structure parameters. This paper explored the method to estimate forest canopy height and aboveground biomass by integrating ICEsat and landsat data in Gongga Mountains. Firstly, the three-dimensional (3-D) vegetation lidar model was used to simulate the terrain slope effect on GLAS waveform and canopy height retrieval. Then, a new waveform decomposition method based on particle swarm optimization-least square method (PSO-LSM) was used to decompose the GLAS waveform. The canopy height was calculated based on the decomposition results. Because the GLAS systems did not have the ability of imaging, regional forest canopy height and biomass were estimated by integrating multispectral data (TM) and GLAS canopy height. The results showed that the ICEsat/GLAS waveform data and Landsat multi spectral data can be combined to estimate regional forest canopy height and biomass with a good precision. The main conclusions in this paper were as follows:1. The 3D vegetation lidar model had been used to simulate the terrain effects on large footprint lidar waveform and canopy height retrieval. 1Model simulation showed that terrain slope broadened and decreased canopy and ground peaks. Although the terrain slope stretched waveform both in the signal-end and signal-start direction, the stretch in the signal-start direction was more significant due to the rapid decline of the ground return. 2The critical slope angle, where ground and vegetation returns are completely mixed, ranged from 9° to 20° in this study. It was in direct proportion to canopy height, and slightly decreased with the increased canopy cover. 3The variation of the GLAS waveform parameters with the slope was also investigated in this paper. The waveform extent was linear with terrain slope, but it may be underestimated over areas with steep slope, especially for the dense forest. Among the quartiles of waveform, RH50 was the most stable height index. The leading edge was a good indicator of the terrain slope for a homogeneous forest, and the trailing edge usually contained the information both of the canopy height and the topography relief. 4Both the direct method and slope-correction model developed by Yang et al. had been employed to estimate canopy height based on the simulation results. Error analysis showed that direct method could be used to retrieve tree height for areas up to a 10° slope. However, the simple physical slope correction could be used to could be used to areas within 30° slope, but there may be a lager underestimation for forests of dense canopy cover.2. This study presented a new waveform decomposition method based on particle swarm optimization-least square method (PSO-LSM). This method decomposed the GLAS waveform into a series of Gaussian components. The initial parameters of Gaussian were obtained using particle swarm optimization algorithm, and then they were optimized using the Levenburg-Marquardt method. The newly proposed method was applied to Gongga Mountain forest. The results indicated that the sum of Gaussian components could be an accurate representation of the original waveform. And the estimated canopy height matched well with the height calculated by the empirical model established by Hayashi et al. (2013) (R2=0.974, RMSE=2.86 m).3. The multispectral TM image and terrain data was used to build the extension model for GLAS canopy height. Correlation analysis showed that the correlation between GLAS canopy height and single factor was not high. Therefore, multiple linear regression, principal component analysis and random forest method were respectively used to build the extension model. The correlation coefficients of the three models were 0.588, 0.559 and 0.669, respectively. The model based on random forest was finally used for regional canopy height inversion. The precision of model was validated by field measured data. The correlation coefficient between the predicted canopy height and the actual canopy height was 0.636, RMSE was 9.35 m. The linear relationship between GLAS data and field measurement was used to correct the inversion results.4. Based on the extended results of GLAS canopy height, the relationship between forest canopy heights and aboveground biomass was modeled. The results showed the predictive ability of the classification model was better than that of the mixed model. For coniferous forest, the power law model between the aboveground biomass and the canopy height could get better effect. The R2 of the model reached 0.601, RMSE was 36.13 t/ha. For broad-leaved forest, the R2 of the power law model was 0.244, RMSE was 20.84 t/ha. It indicated that the canopy height was an important predictor for forest biomass. Then the multiple linear regression model (MLR) incorporating spectral indexes were established. Compared with the canopy height-biomass model, the correlation coefficient of the MLR model became larger. Comparing the two biomass inversion results of broad-leaved forest, it was found that the spatial distribution of the inversion results based on height-biomass model was more reasonable.The results showed that forest biomass could be better modeled using canopy height. 
Pages120
Language中文
Document Type学位论文
Identifierhttp://ir.imde.ac.cn/handle/131551/24607
Collection中国科学院水利部成都山地灾害与环境研究所
Affiliation中国科学院成都山地灾害与环境研究所
First Author Affilication中国科学院水利部成都山地灾害与环境研究所
Recommended Citation
GB/T 7714
卢学辉. ICEsat与Landsat协同的山地森林地上生物量估算研究[D]. 北京. 中国科学院大学,2017.
Files in This Item:
File Name/Size DocType Version Access License
ICEsat与Landsat协同的山地森(5746KB)学位论文 开放获取CC BY-NC-SAView Application Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[卢学辉]'s Articles
Baidu academic
Similar articles in Baidu academic
[卢学辉]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[卢学辉]'s Articles
Terms of Use
No data!
Social Bookmark/Share
File name: ICEsat与Landsat协同的山地森林地上生物量估算研究.pdf
Format: Adobe PDF
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.