应用Sentinel-2A卫星光谱与纹理信息的森林蓄积量估算

Estimation of Forest Stock Volume With Spectral and Textural Information from the Sentinel-2A

以Sentinel-2A为遥感数据源,以第九次森林资源清查数据为样地实测数据,对吉林省中东部的森林蓄积量进行反演。通过对遥感影像进行处理,获取影像的波段光谱值、植被指数,降维处理纹理特征以及地形因子;采用多元线性回归、偏最小二乘法、随机森林、支持向量机等构建了研究区的森林蓄积量估算模型,对检验样本做出了估测。结果表明:机器学习法在反演结果上均优于传统建模方法,随机森林法结果最优,相对误差为17.88%,方程精度为82.12%。

The forest stock volume inversion was conducted in the Middle East of Jilin Province with Sentinel-2A image and National Continuous Forest Inventory data.With band spectral values,vegetation indexes,topographic factors,and the data of textural measurements by reduced dimension,four different modeling methods which include multiple linear regression,partial least squares regression,support vector machine regression,random forest were applied to construct the forest stock volume estimation model.By using the test sample,predictions were made to figure out the optimal model.The two kinds of machine learning methods are superior to two kinds of traditional modeling methods;the random forest method showed the optimal results with the relative error of 17.88%and the accuracy of 82.12%.