| 引用本文： |
-
姜仕昆,谭伟,张雁,梅本清.联合无人机激光雷达和多光谱数据的马尾松单木地上生物量估算模型[J].广西科学,2026,33(3):545-555. [点击复制]
- JIANG Shikun,TAN Wei,ZHANG Yan,MEI Benqing.An Estimation Model of Individual Tree Aboveground Biomass of Pinus massoniana Based on Unmanned Aerial Vehicle LiDAR and Multispectral Data[J].Guangxi Sciences,2026,33(3):545-555. [点击复制]
|
|
| 摘要: |
| 本研究旨在探究联合无人机激光雷达(LiDAR)和多光谱数据估算马尾松Pinus massoniana单木地上生物量的潜力。基于冠层高度模型(CHM)应用分水岭算法分割出460株马尾松，进而提取其点云特征变量和光谱特征，经多元逐步回归结合方差膨胀因子(VIF)共线性诊断筛选建模因子，利用多元线性回归(MLR)和随机森林(RF)算法建立生物量模型。结果显示，单木点云特征变量、多光谱特征变量与马尾松单木地上生物量有密切关联，相关性显著(P＜0.05)。仅使用LiDAR数据的模型包含点云高度最大值(H.max)、高度偏斜度(H.s)、叶面积指数(LAI)、第8层点云切片密度(D7)，RF模型预测性能更优，其在检验数据中的表现为决定系数(R2)=0.76,均方根误差(RMSE)=40.21 kg/株，平均绝对误差(MAE)=33.28 kg/株。仅使用多光谱数据的模型包含绿光叶绿素指数(CIG)、叶片叶绿素指数(LCI)、修正的归一化植被指数(MNDVI)、红边优化土壤调节植被指数(REOSAVI)、红光波段反射率(B1)，MLR模型预测效果较好：R2=0.65、RMSE=48.70 kg/株、MAE=38.84 kg/株。联合两种数据的模型包含点云高度99分位数(H99)、LCI、CIG、LAI、B1、MNDVI、D7，RF模型预测性能最优：R2提升至0.82,RMSE、MAE分别降低至35.01、28.06 kg/株。将单木点云特征变量和多光谱影像的光谱特征相结合，相比单一数据源，能有效提升马尾松单木地上生物量模型的预测效果。 |
| 关键词: 激光雷达 马尾松 多光谱 机器学习 数据耦合 |
| DOI：10.13656/j.cnki.gxkx.20260708.009 |
| 投稿时间：2025-03-24修订日期：2025-04-24 |
| 基金项目:贵州省科技计划项目“基于机载高光谱传感器的马尾松单木生物量的反演模型构建研究”(黔科合基础MS〔2025〕638)，贵州省林业科研项目“基于激光雷达技术的马尾松森林结构参数估测研究”(黔林科合〔2022〕37号)和贵州省科技计划项目“贵州省野外科学观测研究站——马尾松资源培育、评价与监测研究”(黔科合平台YWZ〔2025〕006)资助。 |
|
| An Estimation Model of Individual Tree Aboveground Biomass of Pinus massoniana Based on Unmanned Aerial Vehicle LiDAR and Multispectral Data |
|
JIANG Shikun, TAN Wei, ZHANG Yan, MEI Benqing
|
| (College of Forestry, Guizhou University, Guiyang, Guizhou, 550025, China) |
| Abstract: |
| This article aims to explore the potential of combining unmanned aerial vehicle Light Laser Detection and Ranging (LiDAR) and multispectral data to estimate the individual tree aboveground biomass of Pinus massoniana.On the basis of the Canopy Height Model (CHM),the watershed algorithm was adopted to segment 460 P.massoniana plants,and their point cloud variables and spectral features were extracted.The modeling factors were screened through stepwise regression combined with Variance Inflation Factor (VIF) collinearity diagnosis,and a biomass model was established via Multiple Linear Regression (MLR) and Random Forest (RF) algorithms.The results showed that individual tree point cloud variables and multispectral characteristic variables had correlations with the individual tree aboveground biomass of P.massoniana (P＜0.05).The single tree point cloud variables used for modeling include the maximum point cloud height (H.max),height skewness(H.s),Leaf Area Index (LAI),and eighth layer point cloud slice density(D7).Among the models,the RF model performs better in the test data,with the coefficient of determination (R2)=0.76,Root Mean Square Error (RMSE)=40.21 kg/plant, and Mean Absolute Error (MAE)=33.28 kg/plant.The multispectral features used for modeling include Chlorophyll Index-Green (CIG),Leaf Chlorophyll Index (LCI),Modified Normalized Difference Vegetation Index (MNDVI),Red Edge Optimized Soil Adjusted Vegetation Index (REOSAVI),and red light band reflectance (B1).Among the models,MLR performs better,with R2=0.65,RMSE=48.70 kg/plant,MAE=38.84 kg/plant.The model combining two types of data include point cloud height percentile (H99),LCI,CIG,LAI,B1,MNDVI,D7.Among the models,the predictive performance of RF model was the best,with R2 increased to 0.82,RMSE and MAE decreased to 35.01 kg/plant,and 28.06 kg/plant respectively.Combining individual tree point cloud variables with spectral features of multispectral images can effectively improve the predictive performance of the individual tree aboveground biomass model for P.massoniana compared with a single data source. |
| Key words: LiDAR Pinus massoniana multispectral machine learning data coupling |
