基于机载LiDAR的树冠形态特征因子对单木生物量估算的影响

Effect of Canopy Morphological Characteristics on IndividualTree Biomass Estimation Based on Airborne LiDAR

机载激光雷达遥感(light detection and ranging,LiDAR)是高精度单木参数反演的可靠手段,但单木生物量估算一直是研究的难点和热点。以马尾松、桉树为研究对象,基于从机载LiDAR数据中提取树高、冠幅为自变量,辅助以冠形率(crown shape ratio,CSR)、树冠率(crown rate,CR)等形态特征因子,采用随机森林(random forest,RF)、最小二乘支持向量机(least squares support vector machine,LS-SVM)、梯度提升回归树(gradient boosting decision tree,GBRT)机器学习构建生物量估算模型,对比分析各模型反演单木生物量的精度。结果表明:加入树冠形态特征因子可以有效提高生物量模型的精度;在3种模型中,RF模型效果最佳,未加入树冠形态特征因子的模型拟合结果R^(2)为0.77,rRMSE(relative root mean square error)为21.57%,加入树冠形态特征因子后,在不同的组合下,模型拟合的R^(2)分别为0.86、0.85、0.85,rRMSE分别为20.93%、20.17%、21.19%。

Airborne LiDAR(light detection and ranging)is a reliable method for high-precision inversion of individual tree parameters,but estimating individual tree biomass has always been a difficult and hot research topic.Taking pinus massoniana and Eucalyptus as research objects,based on the extraction of tree height and crown width from airborne LiDAR data as independent variables,assisted by morphological characteristic factors such as CSR(crown shape ratio)and CR(crown rate),biomass estimation models were constructed using RF(random forest),LS-SVM(least squares support vector machine)and GBRT(gradient lift regression tree)machine learning,and the accuracy of each model was compared and analyzed.The results indicate that adding crown shape feature factors can effectively improve the accuracy of biomass models.Among the three models,the RF model had the best performance.The fitting result R^(2) of the model without the addition of crown shape feature factors is 0.77,and the rRMSE is 21.57%.After the addition of crown shape feature factors,under different combinations,the fitting R^(2) of the model is 0.86,0.85,0.85,and the rRMSE was 20.93%,20.17%,and 21.19%,respectively.