基于Sentinel-2与机载激光雷达数据的误差变量联立方程组森林蓄积量反演研究

Study on the forest volume inversion based on the simultaneous equations of error variables of Sentinel-2 and airborne Lidar data

【目的】研究通过提取Sentinel-2中的特征变量与机载激光雷达(Light detection and ranging,LiDAR)中的冠层高度模型(Canopy height model,CHM),探索使用误差变量联立方程组反演森林蓄积量制图的新方法。【方法】以广西壮族自治区国有高峰林场的界牌与东升分场为研究区,机载LiDAR和Sentinel-2影像为数据源,利用皮尔森相关系数与方差膨胀因子(Variance inflation factor,VIF)结合线性逐步回归进行遥感特征变量筛选。通过VIF判断和线性逐步回归保留后的遥感特征变量与LiDAR提取的CHM,分别选用普通回归模型(多元线性逐步回归与Logistic模型)、误差变量联立方程组、随机森林(Random forest,RF)、kNN(k-Nearest Neighbor,kNN)4种反演方法开展森林蓄积量反演,并利用地面实测数据对反演结果进行验证。【结果】1)在普通回归模型中,Logistic模型的反演精度(RRMSE=30.41%)优于MLR模型的反演精度(RRMSE=30.53%);2)在误差变量联立方程组反演方法中,MLR-Logistic联立模型精度(RRMSE=29.29%)优于Logistic-Logistic、MLR-MLR与Logistic-MLR联立模型(RRMSE分别为29.40%、29.60%与29.66%);3)在4种反演方法中,误差变量联立方程组反演结果精度最高(R2=0.60),显著优于普通回归模型方法、随机森林与kNN反演方法(R2分别为0.56、0.39与0.28)。【结论】误差变量联立方程组反演方法更适用于森林蓄积量遥感估测,其反演精度最高,且获得的蓄积量空间连续分布结果与实际接近,制图效果最好,表明误差变量联立方程组反演森林蓄积量制图方法是可行的。

【Objective】By extracting the feature variables in sentienl-2 and the canopy height model(CHM)in airborne Lidar,a novel method for mapping the inversion of forest volume using simultaneous equations of error variables was explored in this research.【Method】Taking Jiepai branch and Dongsheng branch of the state-owned peak forest farm in Guangxi Zhuang Autonomous Region as study area,the paper used airborne Lidar and Sentinel-2 images as data resources.The first step of this dissertation was to screen the remote sensing characteristic variables by using Pearson correlation coefficient method and variation inflation factor analysis method combined with linear stepwise regression.Secondly,the appropriate factors,in company with canopy height model extracted from Lidar,participated in modeling as independent variables.Combined with field data,four kind of volume inversion models were established,such as ordinary regression(multiple linear regression and logistic model),error variable simultaneous equations,random forest regression and kNN regression which were used to verify the accuracy of the model results.【Result】1)When applying ordinary regression inversion method,the inversion accuracy of logistic model(RRMSE=30.41%)was better than that of MLR model(RRMSE=30.53%);2)When applying error variable simultaneous equation group inversion method,the MLR-logistic modeling had the highest accuracy(RRMSE=29.29%),followed by Logistic-Logistic modeling,MLR-MLR modeling and logistic-MLR modeling(RRMSE were 29.40%,29.60%and 29.66%respectively);3)Among the four volume inversion models,the error variable simultaneous equation group inversion method had the highest accuracy(R2=0.60),which was significantly better than ordinary regression inversion method,RF regression inversion method and KNN regression inversion method(R2 were 0.56,0.39 and 0.28 respectively).【Conclusion】Logistic model was more suitable for forest volume inversion than multiple linear regression model,as its accuracy of volume estimation was better than that of multiple linear regression model.Among the four inversion methods,the simultaneous equations of error variable inversion method had the highest inversion accuracy.The inversion of volume using the simultaneous equations of error variable inversion method in the study area showed that the spatial distribution of volume was basically consistent with the actual situation,accompanied by the best mapping effect,which demonstrated that it was feasible to use simultaneous equations of error variables to retrieve forest volume mapping.