无人机激光雷达杉木人工林碳储量估测OA北大核心CSTPCD
Estimation of Forest Carbon Storage in Chinese Fir Plantation by UAV-Lidar
森林碳储量是衡量森林生态系统基本特征的重要指标,传统碳储量估测方式需要耗费大量的时间、人力、物力等,因此利用无人机搭载激光雷达获取机载激光数据并建立森林碳储量估测模型,为获取区域内森林碳储量提供参考,从而更好地对森林资源进行监测.以杉木人工林为研究对象,利用机载激光雷达点云数据获取点云特征变量,以此作为建模变量进行模型研建.通过不同筛选变量方式选择建模变量,分别建立非线性回归模型、线性回归模型和随机森林模型并进行模型评价,通过对比模型的决定系数(R2)、均方根误差(RMSE)和平均绝对误差(MAE)选出最优模型,用于后续研究.结果表明,1)3种模型拟合效果最佳的为随机森林模型,其R2为0.94,RMSE为0.53 t/hm2,MAE 为 0.44 t/hm2.非线性回归模型R2 为 0.71,RMSE 为 0.66 t/hm2,MAE 为0.56 t/hm2;线性回归模型 R2 为 0.67,RMSE 为 0.88 t/hm2,MAE 为 0.80 t/hm2.2)共计提取101个点云特征变量,通过变量筛选发现高度变量、密度变量无论是相关性还是重要性,均大于强度变量.3)对比是否加入优选变量对随机森林模型预测精度的影响,加入优选变量后模型R2没有变化,但RMSE与MAE小于未加入优选变量,精度有所提高.利用机载激光雷达点云数据获取的点云特征变量建立模型,对比非线性回归和线性回归模型,随机森林模型精度最高,用其估算得到研究区内碳储量为480.65 t/hm2,与实测值最相近.因此随机森林模型更适合于区域森林碳储量的估测.
Forest carbon storage is an important index to measure the basic characteristics of forest ecosys-tem.Traditional carbon storage estimation method requires a lot of time,manpower and material re-sources.In this study,unmanned aerial vehicle equipped with lidar was used to acquire airborne laser data and establish a forest carbon storage estimation model to provide reference for obtaining forest carbon stor-age within the region,so as to better monitor forest resources.Chinese fir plantation was taken as the re-search object,and the airborne lidar point cloud data were used to obtain the point cloud characteristic vari-ables,which were used as modeling variables for model establishment.The modeling variables were select-ed through different screening variables,and the nonlinear regression model,linear regression model and random forest model were established,respectively.The optimal model was selected by comparing R2,RMSE and MAE of the models for subsequent research.The results showed that 1)The best fitting effect of the three models was random forest model,whose R2,RMSE and MAE were 0.95,0.53 and 0.44 t/hm2,respectively.In the non-linear regression model,R2,RMSE,and MAE were 0.71,0.66 and 0.56 t/hm2,while in the linear regression model,R2,RMSE,and MAE were 0.67,0.88 and 0.80 t/hm2.2)In this study,a total of 101 point cloud characteristic variables were extracted.Through variable screening,it was found that height variables and density variables were greater than intensity variables in both correla-tion and importance.3)The effect of adding preferred variables on the accuracy of random forest was com-pared.After adding preferred variables,model R2 did not change,but RMSE and MAE were smaller than those without adding preferred variables.The point cloud characteristic variables obtained by airborne Li-DAR point cloud data were used to establish a model.Compared with nonlinear regression and linear re-gression models,the random forest model had the highest accuracy,and the carbon storage in the study are-a was estimated to be 480.65 t by using it,which was the closest to the measured value.Therefore,the sto-chastic forest model is more suitable for estimating regional forest carbon storage.
于艺;姚鸿文;温小荣;汪求来;叶金盛
南京林业大学南方现代林业协同创新中心,江苏南京 210037||南京林业大学林草学院、水土保持学院,江苏南京 210037浙江省森林资源监测中心,浙江杭州 310020广东省林业调查规划院,广东广州 510520
林学
碳储量机载激光雷达随机森林模型点云特征变量
carbon storageairborne lidarrandom forest modelpoint cloud characteristic variable
《西北林学院学报》 2024 (004)
131-137 / 7
广东省林业科技创新项目(2021KJCX001);国家重点研发计划(2016YFC0502704);江苏高校优势学科建设工程资助项目(PAPD).
评论