Research On Estimation Method Of Forest Volume Of Wangyedian Forest Farm Based On Multi-Source Remote Sensing Data

Forest volume is an important indicator for evaluating forest quality.The traditional ground surveys using artificial methods of sampling is time-consuming,labor-consuming and costly,which seriously restricted the efficiency of forest stock acquisition.With the rapid development of remote sensing,the combination of remote sensing images and ground samples has become an important method to estimate forest stock volume.In this paper,the Wangyedian Forest Farm in Inner Mongolia is used as the research area.Based on the stock volume survey of the samples,medium and high resolution optical remote sensing data(Landsat 8,GF-2)combined with C-band dual-polarization synthetic aperture radar data(Sentinel-1)are used.Established MultipleLinear Regression(MLR),Multi-Layer Perceptron(MLP),K-Nearest Neighbor(KNN),Support Vector Machine(SVM)and Random Forest(RF)model to estimate the forest stock volume of Wangyedian forest farm in Inner Mongolia,and made a spatial distribution map of forest stock volume.This study uses a combination of remote sensing data to estimate the forest stock volume,which effectively improves the estimation accuracy.The main research conclusions are as follows:(1)Feature variables such as texture factors and spectral factors extracted from different remote sensing data have different sensitivity to forest stock.The correlation between the texture factor of Landsat 8 image and the forest volume is slightly higher than the spectral factor.The correlation between the spectral factor of GF-2 image and the forest volume is slightly higher than the texture factor.The correlation between the ratio of backscattering coefficient and forest accumulation under different polarization modes extracted from Sentinel-1 image is slightly higher than the backscattering coefficient of single polarization.(2)Landsat 8 data is better than GF-2 data in estimating the forest volume when using a single optical remote sensing data to estimate the forest volume.The optimal model for estimating forest stock based on Landsat 8 is a multiple linear regression model,with an estimation accuracy(EA)of 69.06%.The optimal model for estimating forest stock based on GF-2 is a support vector machine regression model,with an estimation accuracy(EA)of 65.80%.(3)The addition of polarimetric SAR data can effectively improve the estimation accuracy of forest stock.The optimal model for estimating forest stock based on multi-source remote sensing data is a support vector machine regression model,with an estimation accuracy(EA)of 71.46%.With the addition of Sentinel-1,the average accuracy of estimation models reached 69.40%,an increase of 2.44%compared with Landsat 8 data,and an increase of 5.67%compared with GF-2 data.(4)Support Vector Machine(SVM)algorithm has shown great advantages in the five estimation models of forest stock volume.In the process of using optical remote sensing data and multi-source remote sensing data to estimate the forest stock,the support vector machine regression model has higher estimation accuracy than the other four models,indicating that Support Vector Machine(SVM)has strong generalization ability and has great potential in the estimation of forest stock volume.(5)The optimal model for estimating the forest stock volume using multi-source remote sensing data is the support vector machine regression model,and an inversion map of the forest stock volume in the study area has been made.The spatial distribution law of the forest volume of Wangyedian Forest Farm is obtained:the areas with high forest volume are mainly distributed in the west and southeast;the areas with low forest volume are mainly distributed in the northwest,central and northern areas where there have frequent human activities.