Estimation Of Forest Growing Stock Based On Landsat-8 Remote Sensing Image

Forest growing stock is an important factor in forest investigation and an important index for evaluating forest quantity and quality.In this paper,Longquan City in Zhejiang Province is selected as the research area,and the information of Digital Elevation Model(DEM),remote sensing image of Landsat-8,the inventory for forest management planning are used as the basic data.Estimation model of forest growing stock is established by combining spectral information,vegetation index,texture features extracted from remote sensing images with other topographic and measured data,through the methods of Multivariable Linear Regression,Partial Least Squares Regression and General Regression Neural Network.Then predicting and testing the models to obtain a more reliable and stable method.The main contents and results are as follows:(1)The data of DEM and remote sensing information of Landsat-8 is extracted to overlay with the inventory data of forest management planning through ENVI 5.3 and ArcGIS 10.2.The unit growing stock is the dependent variable,and other 18 factors are regarded as independent variable,including Elevation,Slope,Aspect,Band2,Band3,Band4,Band5,Band6,Band7,Normalized Differential Vegetation Index,Ratio Vegetation Index,Difference Vegetation Index,Enhanced Vegetation Index,Red Index,the depth of soil layer,the depth of soil humus layer,tree age,canopy density.(2)The models are established for predicting the total forest growing stock in Longquan through the three methods of Multivariable Linear Regression,Partial Least Squares Regression and General Regression Neural Network by SPSS 20 and MATLAB R2012 a.The accuracy of forecasting is 69.73%,71.13%,74.42%,respectively.Eight texture feature factors which extracted from 3×3 texture window from B8 are added into the General Regression Neural Network model.The accuracy of forecasting is up to 74.96%,which is the best result in this study.(3)Research shows that good data preprocessing plays an important role in modeling later,and screening appropriate indicators can make the model more stable.The General Regression Neural Network model has a good effect in the research,and the addition of texture factor further improves the experimental results,which has made a progress without distinguishing biological characteristics.