期刊
GLOBAL ECOLOGY AND CONSERVATION
卷 33, 期 -, 页码 -出版社
ELSEVIER
DOI: 10.1016/j.gecco.2022.e01999
关键词
Grassland; Unmanned aerial vehicle; Aboveground biomass; Random forest; Vegetation; RGB image
资金
- National Natural Science Foundation of China, China [41801102]
- National Key Research and Development Program of China, China [2017YFA0604801]
This study aims to provide a non-destructive method for quickly obtaining data on grassland aboveground biomass (AGB) using unmanned aerial vehicles (UAVs) and machine learning techniques. By comparing different index combinations, it was found that the model combining horizontal and vertical indices performed the best. However, the lack of vegetation height information in areas with high vegetation coverage remains a limitation.
Remote sensing has become an indispensable method for estimating the regional-scale collection of grassland aboveground biomass (AGB). However, the lack of ground verification samples often reduces the inversion accuracy. This paper aimed to find a non-destructive method to quickly obtain grassland AGB at quadrat-scale through unmanned aerial vehicles (UAVs) in a large area. Thus, we proposed and assessed the vertical and horizontal indices from UAV RGB images as predictors of grassland AGB using the random forest (RF) machine learning technique. By comparing the performance of different indices combinations, we found that the model combing the horizontal and vertical indices (RFVH) performed best (R-2 = 0.78; RMSE = 24.80 g/m(2)), followed by the model using only horizontal indices (the RFH model; R-2=0.73; RMSE =26.54 g/ m(2)), and the last was the model using only the vertical index. However, the REVH model was unsuitable for collecting AGB samples in a large area because the UAVs with RGB cameras failed to obtain vegetation height information in areas with high vegetation coverage. In conclusion, the RFH model can be used to replace the traditional destructive method for collecting ground data over large regions for AGB satellite inversion.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据