Research On Forest Volume Estimation Of Wangyedian Based On Multi-source SAR Data

Since its development,the application and research of synthetic aperture radar(Synthetic Aperture Radar,SAR)in forestry has attracted wide attention at home and abroad.As active remote sensing,it has unique advantages,canopy penetration ability,and is not affected by climate.The emergence of this technology has brought a breakthrough in the exploration of forest vertical structure in forestry investigation.This paper uses the Cband GF-3 SAR and the L-band ALOS PALSAR2 which are both dualpolarization data,combined with topographical variables and tree species dummy variables,adopting support vector machine regression models(SVR),etc.,to invert the stand volume of Wangyedian farm and explore the advantages and potentiality of different SAR data in forest stock estimation.The research indicates:(1)There are significant differences in the correlation between different data sources and the stock volume of each forest stand in Wangyedian.ALOS-PALSAR2 data has a better correlation with the total forest stand volume and the correlation coefficients of two data are mostly between 0.30 and 0.51,compared with GF-3 SAR data.Regardless of GF-3 SAR data or ALOS-PALSAR2 data,backscattering coefficient of cross polarization(HV)are more sensitive to the total forest stand volume than that of copolarization(HH).Both SAR data show a high sensitivity to the stock volume of Pine.GF-3 SAR is particularly sensitive to the stock volume of Pine,the correlation between the ratio of HH and HV from which and the stock volume of Pine is close to 0.60;Both SAR data are lowly relative to the stock volume of Larch.The topographic variables,like S2N and Altitude,also have a good correlation with the stock volume of each forest stand in Wangyedian,and the correlation coefficient of the two data is mostly between 0.30 and 0.62.(2)Based on different data combinations,the SVR model was established to estimate the stock volume of each forest stand in Wangyedian,the inversion levels obtained from which were also significantly different.The experimental results show that when the ALOS-PALS AR2 image after terrain radiation correction is combined with topographic variables to estimate the total stand volume,the best inversion level can be obtained(RMSE,RRMSE,accuracy P,and R2 are respectively 60.77 m3/ha,28.43%,74.35%,and 0.44),which is slightly better than the best inversion result of using GF-3 SAR data to estimate the total stand stock;the GF-3 SAR image before terrain radiation correction combined with topographic variables were used to estimate the stock volume of Pine in estimate,the inversion accuracy of which is the highest(the RMSE and RRMSE are 58.25m3/ha and 25.87%,respectively,and the accuracy P and R2 are 74.21%and 0.54,respectively);the estimation accuracy is not high that was obtained from estimating the stock volume of Larch based on different data combinations,of which the RMSE is over 63m3/ha,and the R2 is less than 0.30.The combination of two SAR data sources did not significantly improve the inversion accuracy when using SVR model to estimate the stock volume of each forest stand.(3)Selecting the SAR data corresponding to the best estimation result,adding its corresponding SAR texture features and adopting SVR model to estimate the stand stock volume can get a better inversion accuracy.The experiment shows that the factors that contribute mainly to the estimation accuracy are HHdB,CX2,HVMean,S2N,W2E2 and Altitude2.The use of multiple linear stepwise regression,random forest regression and support vector machine regression can achieve a pretty good estimation level entirely;Among them,the estimation results based on SVR model are the best,of which the RMSE and RRMSE are only 51.34 m3/ha and 21.51%,respectively,and the accuracy P and R2 are as high as 78.65%and 0.71,respectively.Compared with the previous best inversion accuracy,the result shows that the RMSE is reduced by 11.86%,and the accuracy P is increased by 5.98%.