南京林业大学学报(自然科学版) ›› 2020, Vol. 44 ›› Issue (3): 149-156.doi: 10.3969/j.issn.1000-2006.201811012

• 研究论文 • 上一篇    下一篇

基于Sentinel⁃1和Sentinel⁃2数据的杉木林地上生物量估算

潘磊1(), 孙玉军1(), 王轶夫1, 陈丽萍2, 曹元帅3   

  1. 1.国家林业和草原局森林资源与环境管理重点实验室,北京林业大学林学院,北京 100083
    2.惠州学院,广东 惠州 516007
    3.国家林业局华东林业调查规划设计院,浙江 杭州 310019
  • 收稿日期:2018-11-05 修回日期:2019-01-24 出版日期:2020-05-30 发布日期:2020-06-11
  • 通讯作者: 孙玉军
  • 作者简介:潘磊(1072344791@qq.com),博士生。
  • 基金资助:
    国家林业局“948”项目(2015-4-31)

Estimation of aboveground biomass in a Chinese fir (Cunninghamia lanceolata)forest combining data of Sentinel⁃1 and Sentinel⁃2

PAN Lei1(), SUN Yujun1(), WANG Yifu1, CHEN Liping2, CAO Yuanshuai3   

  1. 1.State Forestry and Grassland Administration Key Laboratory of Forest Resources & Environmental Management, College of Forest, Beijing Forestry University, Beijing 100083, China
    2.Huizhou University, Huizhou 516007, China
    3.East China Inventory and Planning Institute, State Forestry Administration, Hangzhou 310019, China
  • Received:2018-11-05 Revised:2019-01-24 Online:2020-05-30 Published:2020-06-11
  • Contact: SUN Yujun

摘要: 目的

雷达和光学遥感数据可以提供不同方面的信息,利用Sentinel?1与Sentinel?2联合估算亚热带地区森林地上生物量,探索光学数据与合成孔径雷达(SAR)数据结合对于提高森林地上生物量估测的优势。

方法

以福建省将乐国有林场杉木林为研究对象,以Sentinel?1 SAR数据和Sentinel?2光学数据为数据源,采用多元线性逐步回归方法进行建模,以决定系数(R2)、调整决定系数(R2adj)、均方根误差(RMSE)、方差膨胀因子(VIF)为模型评价指标,对比分析Sentinel?2光学数据与Sentinel?2结合Sentinel?1 SAR数据估算森林地上生物量的能力。

结果

基于Sentinel?2光学数据的森林地上生物量估算模型,其调整决定系数(R2adj)达到0.501、均方根误差(RMSE)为64.04 Mg/hm2;Sentinel?2光学数据结合Sentinel?1 SAR数据的森林地上生物量估算模型,其调整决定系数(R2adj)达到0.575、均方根误差(RMSE)为59.13 Mg/hm2,对比Sentinel?2估算模型,该模型精度有明显提高。

结论

Sentinel?2卫星的多光谱数据能够作为估算亚热带地区森林地上生物量的有效数据,加入Sentinel?1 SAR影像的极化纹理信息后,利用Sentinel?1雷达传感器的全天候获取数据能力与Sentinel?2多光谱传感器丰富的光谱波段信息特点,以及两者短重访周期的能力,能够有效提高估算森林地上生物量模型的精度。

关键词: 杉木林, 地上生物量, Sentinel?1, Sentinel?2, 纹理, 估测模型

Abstract: Objective

Forest above-ground biomass (AGB) is the main storage site of carbon, which plays an important role in the regional and global carbon cycle. In studies on AGB estimation based on remote sensing, saturation of optical and radar data is a well-known phenomenon, however, it is unclear so far as to how this problem may be reduced in order to improve AGB estimations. Optical and radar data offer different advantages for recording forest stand structures, thus optimized use of their features may help improve AGB estimation. This study was conducted to assess whether forest AGB estimations can be produced in a subtropical area based on data produced using Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 multispectral instruments.

Method

A stand of Chinese fir (Cunninghamia lanceolata) in a state-owned forest farm in Jiangle, Fujian Province, China, and Sentinel-1 SAR data and Sentinel-2 multispectral optical data were used. Backscatter coefficients of Sentinel-1 images in different polarization modes (VH, VV, VH/VV) and second-order texture measures of a gray level co-occurrence matrix in different window sizes (3×3, 5×5, 7×7 and 9×9) were recorded, and band values, vegetation indices, and second-order texture measures in 3 × 3 windows of Sentinel-2 image were extracted as remote sensing factors. Models were fitted using multiple linear stepwise regression. AGB estimation models based on Sentinel-2 remote sensing factors only and on cooperative Sentinel-2 and Sentinel-1 remote sensing were established. Determination coefficient (R2), adjusted determination coefficient (Radj2), root mean square error (RMSE), and variance inflation factor were selected for assessing the models. The ability of Sentinel-2 optical data and Sentinel-2 optical data combined with Sentinel-1 SAR data to estimate forest AGB was compared and analyzed.

Result

The results showed that when Sentinel-2 band values, vegetation indices and texture measures were used as dependent variables to predict AGB, the estimation effect was inadequate, and the Radj2 values of the models were 0.319, 0.303 and 0.455, respectively; an AGB estimation model based on all remote sensing factors of Sentinel-2 optical data (Model-1) produced an Radj2 value of 0.501, while RMSE was 64.04 Mg/hm2. When the Sentinel-1 remote sensing factor was added, the AGB estimation model based on Sentinel-2 optical data and Sentinel-1 SAR data was improved, with a model adjustment coefficient (Radj2) and RMSE were 0.575 and 59.13 Mg/hm2. Compared with Model-1, the accuracy of Model-2 was considerably improved, thus the second-order texture measures of Sentinel-1 VH polarimetric imaging plays an important role.

Conclusion

Our study demonstrates that multispectral optical data of the Sentinel-2 satellite can be used as reliable data source for estimating forest AGB in subtropical areas, and the combination of Sentinel-1 SAR data and Sentinel-2 multispectral optical data can substantially improve accuracy of estimating forest AGB. Estimations of forest ABG by combining multispectral optical remote sensing images and space-borne polarimetric SAR data offers considerable advantages. The results of the present study may help improve accuracy of forest AGB estimation, and it provides a reference basis for future studies on forest AGB using different remote sensing data sources.

Key words: Chinese fir forest, aboveground biomass, Sentinel-1, Sentinel-2, texture, regression model

中图分类号: