基于Sentinel-2数据的林分平均树高和平均胸径估测研究

Research on Estimating Average Tree Height and Average DBH of Forest Stand Based on Sentinel-2 Data

基于Sentinel-2遥感数据,结合多种特征重要性评估方法与机器学习模型,实现了对杉木平均胸径和平均树高参数的定量估计。研究结果表明,基于随机森林回归算法及RF特征重要性评价指标的模型,在平均树高和平均胸径的估计中表现优异,显示出较高预测精度和稳定性,两者的确定系数(R^(2))和相对均方根误差(RRMSE)分别为0.42、18.05%和0.59、18.20%。值得注意的是,不同特征重要性评价指标的应用,在特征筛选阶段引发了特征集构成的多样性,进而对模型性能产生了较为显著的影响,凸显了特征选择策略的重要性。本研究不仅提供了一种高效的森林资源调查监测方法,还为后续相关研究提供了宝贵的参考。

This study integrated Sentinel-2 remote sensing data with various feature importance assessment methods and machine learning models to achieve accurate estimation of the average diameter at breast height(DBH)and average tree height parameters of Cunninghamia lanceolata.The results indicated that the model based on random forest regression algorithm and RF feature importance evaluation index performed the best in estimating both average tree height and DBH,showing high prediction accuracy and stability.The relative root mean square error(RRMSE)and coefficient of determination(R^(2))for estimates were 18.05%and 0.42 for average tree height,18.20%and 0.59 for average DBH,respectively.Notably,the application of different feature importance evaluation indicators led to diverse feature sets during the feature selection phase,which had a relatively significant impact on model performance,high lighting the importance of feature selection strategies.This study not only provides an efficient forest resource surveying and monitoring method,but also provides valuable reference for subsequent related research.