欢迎访问林草资源研究

林业资源管理 ›› 2021›› Issue (6): 83-89.doi: 10.13466/j.cnki.lyzygl.2021.06.014

• 科学研究 • 上一篇    下一篇

高山松地上生物量估测与尺度转换研究

唐金灏1(), 张加龙1(), 陈立业2, 程滔3   

  1. 1.西南林业大学 林学院,昆明 650224
    2.芷江侗族自治县林业局,湖南 怀化 419100
    3.国家基础地理信息中心,北京 100830
  • 收稿日期:2021-09-02 修回日期:2021-10-28 出版日期:2021-12-28 发布日期:2022-01-12
  • 通讯作者: 张加龙
  • 作者简介:唐金灏(1994-),男,甘肃武威人,在读硕士,从事林业遥感方面的研究。Email: TJH@swfu.edu.cn
  • 基金资助:
    国家自然科学基金(31860207);2020年云南省高层次人才培养支持计划“青年拔尖人才”专项(81210468);西南林业大学科研启动基金(111932)

Research on Estimation of Aboveground Biomass and Scale Conversion for Pinus densata Mast

TANG Jinhao1(), ZHANG Jialong1(), CHEN Liye2, CHENG Tao3   

  1. 1. Faculty of Forestry,Southwest Forestry University,Kunming 650224,China
    2. Zhijiang Dong Autonomous County Forestry Bureau,Huaihua,Hunan 419100,China
    3. National Geomatics Center of China,Beijing 100830,China
  • Received:2021-09-02 Revised:2021-10-28 Online:2021-12-28 Published:2022-01-12
  • Contact: ZHANG Jialong

摘要:

为研究遥感影像尺度上推对森林地上生物量估测的影响,以香格里拉市为研究区,基于Landsat-8,Sentinel-2A,和Spot-7影像,采用最邻近像元法、双线性内插法、三次卷积插值法、局部平均法以及像元聚合法将原始影像转换至低空间分辨率影像,结合外业调查的高山松样地地上生物量数据,分别建立随机森林(RF)和梯度提升回归树(GBRT)生物量估测模型,并与目标尺度真实影像的建模效果进行对比。结果表明:经最邻近像元法上推后,基于Spot-7影像的RF和GBRT建模的预估精度(P)分别为76.65%和75.55%,基于Sentinel-2影像的RF和GBRT建模的P值分别为81.78%和72.74%,均优于其余4种尺度转换方法;经5种方法尺度上推后的影像构建的RF预估精度(81.78%~63.94%)总体优于GBRT预估精度(75.55%~61.03%);采用Sentinel-2A影像更适合尺度上推进行森林生物量估测。研究结果可为尺度转换方法和生物量估测模型的选取提供参考。

关键词: 多源遥感数据, 地上生物量, 尺度转换, 随机森林, 梯度提升回归树

Abstract:

In order to study the impact of scaling up of remote sensing images for forest biomass estimation,this paperused the Nearest Neighbor method,Bilinear Interpolation,Cubic Convolution Interpolation,Local Average method and Pixel Aggregation method to convert original Landsat-8,Sentinel-2A and Spot-7 images to low spatial resolution images in Shangri-La City. Then,this paper combined the aboveground biomass data of Pinus densata Mast. obtained from field surveys and the real image of target scale to establish the biomass estimation models of Random Forest (RF) and Gradient Boosted Regression Trees (GBRT) respectively,and compared the modeling effects of the two models. The results showed that: after being scaled up by the Nearest Neighbor method,the estimation accuracy (P) of RF and GBRT modeling based on Spot-7 images were 76.65% and 75.55%,respectively.,the P values of the two models based on Sentinel-2 images were 81.78% and 72.74% respectively,which were better than the other four scale conversion methods;The estimation accuracy of the RF estimation (81.78%~63.94%) of the image constructed by the five scale conversion methods was better than the estimation accuracy of GBRT (75.55%~61.03%). Sentinel-2A images were more suitable for scaling up to estimate forest biomass. The study results can provide reference for the selection of scale conversion methods and biomass estimation models.

Key words: multisource remote sensing data, aboveground biomass, scale conversion, random forests, gradient boosting regression tree

中图分类号: