Analyzing The Saturation Of Growing Stem Volume Based On ZY-3 Stereo And Multispectral Images In Planted Coniferous Forest

In recent years,the use of remote sensing data to estimate growing stem volume(GSV)has become an important technical tool.The spectral saturation effect tends to occur in the high GSV portion of forest stands during GSV estimation,which seriously hinders the accuracy of GSV estimation.At present,forest canopy height is considered as one of the main means to solve the spectral saturation effect of GSV.And the canopy height model extracted from stereoscopic imagery is an important means to obtain the forest canopy height.In this study,we used ZY-3 stereoscopic imagery combined with open-source DEM to extract canopy height model(CHM),and constructed five variable sets(CHM variable set,modified CHM variable set,multispectral data variable set,CHM joint multispectral data variable set and modified CHM joint multispectral data variable set)with four inversion models(multivariate linear model,support vector machine,K-neighborhood method,and random forest model)and three variable screening methods(stepwise regression model method,LASSO model method,and Pearson correlation coefficient method)used to estimate the accumulation of larch and Chinese pine in the study area.On this basis,the spectral saturation of different sets of variables was probed using the kriging spherical model,and the response process of ZY-3 stereo image pairs and multispectral data to the spectral saturation problem of GSV was analyzed,and the main findings were as follows.(1)Combining the DSM and DEM extracted by ZY-3 stereoscopic imagery,the accuracy of CHM can be effectively improved by the accurate alignment of plane position and elevation.The RMSE of the corrected CHM in the larch area is 2.02m and that of the corrected CHM in the Chinese pine area is 2.79m after the correction of the CHM data obtained with the DEM using the difference method combined with the correction model constructed by ZY-3 multispectral data,both of which have obtained good correction results.(2)Different feature screening methods and accumulation estimation models were used to invert the forest accumulation in the study area.rRMSE was distributed in the interval 22.2%-31.1%,especially the corrected CHM improved the accuracy of GSV most significantly.It was found that the types of inversion models and variable screening methods had little effect on the estimation results of the two plantation tree species.The optimal estimation model of larch R2=0.62,RMSE=52.3m3/ha,rRMSE=25.0%;the optimal estimation model of oleander R2=0.59,RMSE=48.9m3/ha,rRMSE=22.2%.(3)For the multiple feature sets constructed,the spectral saturation of different feature sets was estimated using a kriging spherical model,and the role of CHM information in improving the spectral saturation was clarified.The saturation of each model was further estimated by using the spherical model in the kriging interpolation model for the larch and Chinese pine GSV estimation results,and it was found that the variable selection method of the constructed model and the type of GSV inversion model had minimal effects on the saturation,while the saturation of the ZY-3 stereoscopic imagery combined with the ZY-3 multispectral data was higher than that of the ZY-3 multispectral data for the GSV estimation.The saturation of the optimal model for the five variable sets in the order of low to high is the multispectral data variable set,CHM variable set,modified CHM variable set,CHM combined multispectral data variable set and modified CHM combined multispectral data variable set;it can be seen from the final saturation estimation that the saturation of larch is improved from 150 m3/ha to 266 m3/ha,and the saturation of Chinese pine is improved from 220 m3/ha to 307 m3/ha,indicating that the saturation is significantly resolved by the ZY-3 stereoscopic imagery.