Eucalyptus forest stock estimation study based on Landsat 8 OLI and Resource 3 stereo image
[Objective]To explore the potential of Landsat8 OLI data and stereo data in estimating Eucalyptus forest stock volume(FSV)and to accurately estimate eucalyptus FSV.[Method]This study used three Landsat8 OLI images and ZY-3 stereo data as remote sensing data sources and combined with a small amount of ground survey.The remote sensing estimation of Eucalyptus FSV was achieved by combining a small amount of ground survey data.There were three main types of remote sensing features used to estimate the FSV of eucalyptus:the first was the spectral features including vegetation index and single-band reflectance;the second was the eight texture features extracted based on the single-band of Landsat 8 OLI images;the third was the canopy height extracted based on the Resource 3 stereo image pair data and the open-source digital elevation model(DEM).DEM extracted from the canopy height model(CHM).The Boruta algorithm was used to characterize the three types of remote sensing features,and then three machine learning models,Random Forest(RF),K-Nearest neighbor(KNN)and Support vector machine(SVM),as well as the traditional multiple linear regression model(MLR),were established.And evaluated the model results with coefficient of determination(R2),root mean square error(RMSE)and relative root mean square error(rRMSE)as evaluation metrics.The model results were evaluated.[Result]The CHM extracted from ZY-3 stereo relative data and open-source DEM data had a strong correlation with the FSV of Eucalypts,with a Pearson correlation coefficient of 0.71.It was difficult to accurately estimate the FSV of Eucalypts using only spectral and texture features extracted from Landsat 8 OLI images,and the R2 of the estimated model was 0.29-0.38,rRMSE of 35.65%-43.30%,and a serious data saturation problem.When CHM was added to the variable set,the estimation accuracy of the model improved significantly,with R2 reaching 0.64-0.66 and rRMSE of 25.74%-26.41%.[Conclusion]There were serious data saturation problems when estimating Eucalyptus FSV using Landsat 8 OLI data,and it was difficult to effectively improve the estimation accuracy of forest stock using texture features with a spatial resolution of 30 m.More accurate CHM can be extracted using ZY-3 stereo image pair data and open-source DEM data,and the extracted CHM can significantly improve the saturation problem of optical data,thus improving the estimation accuracy of Eucalyptus FSV.
NETL
NSTL
万方数据
