基于Sentinel-1A微波遥感数据的森林蓄积量估测

Estimation of Forest Volume Based on Sentinel-1A Microwave Remote Sensing Data

Sentinel-1A作为开源的集微波和光学数据一体的卫星数据源,给森林资源调查和监测提供了重要的数据源。为了探索Sentinel-1A数据在森林资源调查中的可用性,以Sentinel-1A为遥感数据源,结合地面样地森林蓄积量调查数据,采用随机森林法和偏最小二乘法两种模型对云南省普洱市思茅区的森林蓄积量进行预测以及遥感反演。通过对遥感影像进行一系列预处理,提取微波遥感数据VV和VH极化下的后向散射系数,并分别计算5个窗口(3m×3m、5m×5m、7m×7m、9m×9m、11m×11m)下的8种纹理特征,共计83个特征作为备选自变量,其中31个特征与蓄积量通过相关性检验,结合219块地面调查样地,采用随机森林法和偏最小二乘法两种算法,进行建模因子重要性分析,选择10个最优特征,建立随机森林蓄积量估测模型并进行定量反演。随机森林法回归模型在结果上优于偏最小二乘法回归模型,随机森林法的模型预测R 2为0.80,RMSE为30.14 m 3/hm 2;偏最小二乘法的模型预测R 2为0.70,RMSE为36.68 m 3/hm 2。随机森林法相较于偏最小二乘法在森林蓄积量预估及反演方面具有明显的适用性,有利于该模型在森林蓄积量定量估算和反演中的推广。

Sentinel-1A,as an open source satellite data source integrating microwave and optical data,provides an important data source for forest resource investigation and monitoring.In order to explore the availability of Sentinel-1A data in the investigation of forest resources,Sentinel-1A was used as the remote sensing data source,combined with the survey data of forest stocks on ground sample plots,and two models of random forest method and partial least squares method were applied to Puer City,Yunnan Province.The forest volume in Simao District is predicted and remotely sensed.Through a series of preprocessing of remote sensing images,the backscattering coefficients of microwave remote sensing data VV and VH polarization are extracted,and 8 texture features under 5 windows(3m×3m,5m×5m,7m×7m,9m×9m,11m×11m)are calculated,a total of 83 features as alternative independent variables,of which 31 features and accumulation passed the correlation test,combined with 219 ground survey plots,using two algorithms,random forest method and partial least squares method,modeling factor importance analysis,selecting 10 optimal features,establishing random forest accumulation estimation model and quantitative inversion.The stochastic forest method is superior to the partial least square method in the inversion results,the prediction R 2 of the stochastic forest regression model is 0.80,the RMSE is 30.14 m 3/hm 2,and the prediction R 2 of the partial least squares regression model is 0.70,and the RMSE is 36.68 m 3/hm 2.The stochastic forest method has obvious applicability in forest volume estimation and inversion compared with partial least squares method,which is beneficial to the generalization of the model in forest volume quantitative estimation and inversion.