欢迎访问林业科学,今天是

林业科学 ›› 2022, Vol. 58 ›› Issue (2): 13-22.doi: 10.11707/j.1001-7488.20220202

• 前沿与重点: 森林碳汇专题 • 上一篇    下一篇

基于光学遥感的稀疏乔灌木地上部分生物量反演方法

石永磊1,2,王志慧2,*,李世明3,李春意1,肖培青2,张攀2,常晓格1   

  1. 1. 河南理工大学测绘与国土信息工程学院 焦作 454003
    2. 黄河水利科学研究院水利部黄土高原水土保持重点实验室 郑州 450003
    3. 中国林业科学研究院资源信息研究所 北京 100091
  • 收稿日期:2021-06-24 出版日期:2022-02-25 发布日期:2022-04-26
  • 通讯作者: 王志慧
  • 基金资助:
    国家自然科学基金项目(41701509);中央级公益性科研院所基本科研业务费专项资金项目(HKY-JBYW-2020-09);国家自然科学基金项目(41671507);河南省青年骨干教师项目(2019GGJS059)

A Method of Estimation Aboveground Biomass of Sparse Tree-Shrub Using Optical Remote Sensing

Yonglei Shi1,2,Zhihui Wang2,*,Shiming Li3,Chunyi Li1,Peiqing Xiao2,Pan Zhang2,Xiaoge Chang1   

  1. 1. School of Surveying and Land Information Engineering, Henan Polytechnic University Jiaozuo 454003
    2. Key Laboratory of Soil and Water Conservation in the Loess Plateau of Ministry of Water Resources, Yellow River Institute of Hydraulic Research Zhengzhou 450003
    3. Research Institute of Forest Resource Information Techniques, CAF Beijing 100091
  • Received:2021-06-24 Online:2022-02-25 Published:2022-04-26
  • Contact: Zhihui Wang

摘要:

目的: 基于无人机数据采用3种分层方案构建冠层盖度-乔灌木地上部分生物量模型以及基于Landsat8 OLI数据采用3种分层方案构建不同光谱指数-乔灌木地上部分生物量模型, 对比分析不同分层方案的乔灌木地上部分生物量模型精度, 以期为基于遥感数据的干旱区人工林乔灌木地上部分生物量高精度反演提供理论依据。方法: 在毛乌素沙地实地调查102块30 m×30 m样地, 基于高分辨率无人机影像, 利用面向对象的机器学习算法获取乔灌草植被覆盖度信息, 采用3种分层方案(不分层、基于乔木和灌木2种植被类型分层、基于5个树种分层)构建冠层覆盖度-乔灌木地上部分生物量模型。基于Landsat8 OLI影像, 使用6种光谱指数(NDVI、RVI、MSAVI、TCG、NDMI、NIRv), 结合无人机影像解译草本植被覆盖度, 采用3种分层方案(不分层、有无草本植被样地分层、3个草本植被覆盖度等级样地分层)构建不同光谱指数-乔灌木地上部分生物量模型。结果: 不分层的冠层覆盖度-乔灌木地上部分生物量模型鲁棒性最差(R2=0.22, n=102), 且估算精度最低(RMSE= 14.98 t·hm-2); 考虑乔木和灌木2种植被类型分层建模(RMSE = 7.44 t·hm-2)和5个树种分层建模(RMSE = 5.82 t·hm-2)的反演误差分别减少了50.32%和61.1%。在光谱指数-乔灌木地上部分生物量模型中, NIRv反演乔灌木地上部分生物量精度最高(3种分层方案平均RMSE = 7.25 t·hm-2), NDVI反演乔灌木地上部分生物量精度最低(3种分层方案平均RMSE = 9.43 t·hm-2)。不同光谱指数对稀疏乔灌木地上部分生物量变异的解释能力表现为NIRv>NDMI>TCG>MSAVI>RVI>NDVI。木本植被类型对光谱指数-乔灌木地上部分生物量模型精度的影响小于对冠层覆盖度-乔灌木地上部分生物量模型精度的影响。考虑草本植被覆盖度背景分层建模可使光谱指数-乔灌木地上部分生物量模型RMSE减少8.13%~16.62%, 不同光谱指数-乔灌木地上部分生物量模型精度对草本植被覆盖度背景的敏感性排序为NIRv>TCG> NDVI>MSAVI>RVI>NDMI。结论: 无人机高空间分辨率遥感可用于获取稀疏乔灌混交林树种类型及其草本植被信息等先验知识。在稀疏乔灌混交林区域, 木本植被类型对冠层覆盖度-乔灌木地上部分生物量模型精度影响较大, 至少需区分乔木和灌木两类植被才可保证该方法反演精度满足实用需求。基于Landsat-8 OLI卫星数据的考虑草本覆盖度的NIRv-乔灌木地上部分生物量模型分层建模方案适用于大区域稀疏乔灌木地上部分生物量遥感估算。

关键词: 稀疏乔灌木, 地上生物量, 分层建模, 光谱指数, 冠层覆盖度

Abstract:

Objective: Three stratification schemes were used to construct canopy coverage-aboveground biomass model of tree-shrub based on UAV data and aboveground biomass models of tree-shrub based on spectral indices of Lansat8 OLI data. Differences among stratification schemes in the precision of the aboveground tree-shrub biomass model were investigated to provide a scientific and theoretical basis for high-precision estimation of aboveground tree-shrub biomass of plantations in dryland based on remote sensing data. Method: A total of 102 plots were surveyed in Mu Us Sandy Land, information on vegetation coverage of tree, shrub and herbaceous were obtained from the high-resolution UAV images using object-oriented machine learning algorithm, and then the canopy coverage-aboveground biomass models of tree-shrub were developed with three stratification schemes (non- stratification, stratification-based two vegetation types of tree and shrub, stratification-based five tree species). Based on landsat-8 OLI images, 6 spectral indices (NDVI, RVI, MSAVI, TCG, NDMI, NIRv) were used to interpret the herbaceous coverage in combination with UAV images. Different spectral index-aboveground tree-shrub biomass models were developed using the 3 stratification schemes (non-stratification, stratification using plots with and without herbaceous vegetation, stratification using plots with three herbaceous coverage level). Finally, the precision of estimation was compared among the models with different stratification schemes. Results: The canopy coverage-aboveground tree-shrub biomass model without stratification has the worst robustness (R2=0.22, n=102) and the lowest estimation precision (RMSE=14.98 t·hm-2). The estimation errors were reduced by 50.32% and 61.1% for the models with stratification by 2 vegetation types of tree and shrub (RMSE=7.44 t·hm-2) and by 5 tree species (RMSE=5.82 t·hm-2), respectively. In the spectral index-aboveground tree-shrub biomass models, NIRv had the highest precision in estimation of aboveground tree-shrub biomass (the average RMSE of the 3 stratification schemes was 7.25 t·hm-2), and NDVI had the lowest precision (the average RMSE of the 3 stratification schemes was 9.43 t·hm-2). The explanatory ability of different spectral indices to variation of aboveground biomass of sparse tree-shrub was ranked as follows: NIRv > NDMI > TCG > MSAVI > RVI > NDVI. The effect of woody vegetation types on precision of spectral index-aboveground tree-shrub biomass model was less than that of the canopy coverage-aboveground tree-shrub biomass model. The RMSE of the spectral index-aboveground tree-shrub biomass models was reduced by 8.13%~16.62% considering the background of herbaceous coverage, and the sensitivity of precision to such background was ranked as follow: NIRv > TCG > NDVI > MSAVI > RVI > NDMI. Conclusion: The high spatial resolution remote sensing of UAV can be used to obtain prior knowledge of tree species types and herbaceous vegetation information for sparse tree-shrub mixed forest. The woody vegetation types have a great influence on the precision of canopy coverage-aboveground biomass model in sparse tree-shrub mixed forest, and at least the two vegetation types of tree and shrub should be distinguished to ensure the estimation precision of the method to meet the practical application requirements. The stratification model schemes of NIRv-aboveground biomass model derived from Landsat-8 OLI satellite data considering herbaceous vegetation coverage is suitable for remote sensing estimation of aboveground tree-shrub biomass in a large area.

Key words: sparse tree-shrub, aboveground biomass, stratification-based model, spectral index, canopy coverage

中图分类号: