基于机载激光雷达和机器学习算法的林分平均胸径遥感估测
Remote sensing estimation of average diameter at breast height of forest stands based on airborne LiDAR and machine learning algorithms
[目的]针对森林调查中较大范围估测林分平均胸径精度不高的问题,提出了一种林分平均胸径遥感估测新方法。[方法]选取贵州省桂花国有林场马厂工区为研究区域,基于机载激光雷达点云数据结合样地调查数据,提取样地水平冠层特征变量和高度特征变量,使用方差膨胀因子分析和皮尔逊相关性检验进行变量筛选,构建基于机载激光雷达和机器学习算法的林分平均胸径估测模型。[结果]研究结果表明:(1)通过方差膨胀因子分析和皮尔逊相关性检验,点云特征变量如平均冠层高度和高度偏态与林分平均胸径具有较好的相关性。其中,平均冠层高度对模型构建的贡献最大,与林分平均胸径呈显著正相关。(2)在比较多元线性回归和机器学习模型后,我们发现机器学习模型优于多元线性回归模型,其中随机森林模型的拟合效果最好。随机森林模型的决定系数(R2)为 0.71,均方根误差(RMSE)为 2.50。(3)通过四种森林类型林分平均胸径预测值与实际值的差值进一步证实随机森林模型精度最高,拟合效果最好。[结论]综上,利用机载激光雷达点云数据提取点云特征变量,并构建基于机器学习算法的林分平均胸径估测模型是可行的。该方法的精度满足了森林资源调查的应用需求,可作为辅助林业调查工作的技术手段。关键词:机载激光雷达;林分平均胸径;随机森林回归;点云特征变量
[Objective]A new method for remotely estimating the average diameter at breast height(DBH)of forest stands was proposed to address the problem of low accuracy in large-scale estimation in forest surveys.[Method]Selecting the Machang work area of Guihua State owned Forest Farm in Guizhou Province as the research area,based on airborne LiDAR point cloud data combined with sample plot survey data,the horizontal canopy characteristic variables and height characteristic variables of the sample plot were extracted.Variance expansion factor analysis and Pearson correlation test were used to screen variables,and a forest average diameter estimation model based on airborne LiDAR and machine learning algorithm was constructed.[Result]The research results show that:(1)By analyzing the variance inflation factor and conducting Pearson correlation tests,we selected point cloud feature variables that have a good correlation with the DBH of forest stands,such as average canopy height and height skewness.Among them,the average canopy height contributes the most to the model construction and shows a significant positive correlation with the average DBH of forest stands,indicating its important influence on the variation of forest DBH.(2)After comparing multiple linear regression and machine learning models,we found that machine learning models outperform multiple linear regression models,with the random forest model achieving the best fit.The coefficient of determination(R2)for the random forest model is 0.71,and the root mean square error(RMSE)is 2.50.(3)The difference between the predicted average diameter at breast height(DBH)and the actual values further confirms that the random forest model has the highest accuracy and the best fitting effect among the four forest types.[Conclusion]In conclusion,it is feasible to use airborne LiDAR point cloud data to extract point cloud feature variables and build a model for remotely estimating the average DBH of forest stands based on machine learning algorithms.The accuracy of this method meets the application requirements of forest resource surveys and can serve as a technical means for assisting forestry survey work.
唐佳俊;柴宗政;
贵州大学林学院 贵阳 550025;贵州大学林学院 贵阳 550025;
airborne LiDAR average diameter at breast height of the forest stand random forest regression point cloud feature variables
215-224 / 10
评论