Research On Estimation And Spatial Scaling Of Forest Biomass Based On Multi-source Remote Sensing Data

Scientific and accurate estimates of regional forest biomass is significant for studies of global carbon cycle,forest productivity and climate change.The accuracy of regional forest biomass estimation based on multi-source remote sensing still needs to be improved.Firstly,multi-spectral satellite images of SPOT5,large footprint Light Detection and Ranging?LiDAR?-GLAS data,MODIS data,related auxiliary data and the data of simple plots are collected.Then,by optimizing the selection method of SPOT5 parameters,the estimation accuracy of the maximum crown height estimation model of the forest within the spot based on GLAS waveform parameters is improved.The maximum crown height distribution map of the forest is obtained by combining SPOT5 remote sensing data;the combined forest crown height data and SPOT5 Remote sensing data was used to establish a regional forest biomass estimation model.The best estimation model was applied to MODIS and SPOT5 remote sensing data,and a scale conversion model based on remote sensing for forest biomass estimation was established.The results showed that the combined multi-source remote sensing data is beneficial to improve the estimation accuracy of forest biomass.Through scale conversion research,large-scale and high-precision forest biomass estimation results based on low-resolution remote sensing images can be obtained.The main results and conclusions are as follows:?1?The estimation accuracy of forest maximum canopy height estimation model based on waveform parameters of GLAS footprint is improved by optimizing the selection method of SPOT5 parameters.The results showed that when 20m is used as the threshold of terrain fluctuation to divide the terrain of the study area,improved topographic index model by introducing texture parameters,and the accuracy of the improved model is optimal in both gentle and inclined areas.Compared with the topographic index model,by the Leave-one-out Cross-Validation the improved model increases the R² from 0.777 to 0.952 in the sloped region,and the RMSE decreases from2.90 m to 1.35 m,and the accuracy is improved significantly.The texture feature can reflect the change of forest canopy height indirectly by reflecting the changes of canopy structure of different forest ages,thus reducing the influence of canopy structure difference on GLAS waveform parameters and improving the estimation accuracy of forest canopy height.?2?Utilizing Partial Least Squares Regression?PLS?,Support Vector Regression Model?SVR?and Random Forest Model?RF?,the model of estimation model for the maximum canopy height of regional forests is established by combining the maximum canopy height of the forest acquired by GLAS waveform data,SPOT5 remote sensing data and forest leaf area index data.The results showed that the best estimation accuracy can be obtained by using random forest model,including broad-leaved forest R²=0.90,RMSE=1.69 m,coniferous forest R²=0.93,RMSE=1.85 m,coniferous and broad-leaved mixed forest R²=0.95,RMSE=1.59 m.Studies have shown that The research showed that the texture feature is more sensitive to the change of forest structure parameters than the spectral parameters.Use the Random Forest Model can effectively simulate the linear or nonlinear relationship between texture and forest crown height,and is suitable for estimate the regional forest canopy height.?3?Using the complementary advantages of Multi-source remote Sensing data and utilizing Partial Least Squares Regression?PLS?,Support Vector Regression Model?SVR?and Random Forest Model?RF?,the model of estimation regional forest biomass is established by combining the forest maximum canopy height data and optical remote sensing data.The results showed that the maximum crown height of forest was the most sensitive to forest biomass,and the Random Forest Model had the highest estimation accuracy,including broad-leaved forest R²=0.93,RMSE=14.39 Mg?hm^-2;coniferous forest R²=0.93,RMSE=19.80 Mg?hm^-2 Coniferous and broad-leaved mixed forest R²=0.94,RMSE=26.61 Mg?hm^-2.Studies have shown that the Random Forest Model which combines forest crown height and optical remote sensing data,has good adaptability to the complex variability of forest biomass,and can effectively estimate the regional forest biomass.?4?Obtaining forest biomass distribution maps based on MODIS and SPOT5remote sensing data in the study area through the best forest biomass estimation model,and the scale effect between them was compared and analyzed.The texture parametrization approaches and Random Forest Model algorithm approach based on mixed pixels will be checked to establishment of an estimation model for correction factors of forest biomass scale effect,and the scale effect correction of forest biomass based on MODIS data estimation was carried out.The results showed that the correlation coefficient R between the corrected MODIS biomass and SPOT5 biomass increased from 0.61 to 0.90,and the RMSE decreased from 172.3 Mg?hm^-2 to 24.0Mg?hm^-2.The improvement effect was extremely significant.Studies have shown that based on the texture parametrization approaches combined with Random Forest Models can effectively obtain scale effect correction factors between forest biomass at different scales,thus improving the accuracy of forest biomass estimation for low-resolution remote sensing images,and achieving a wide range of forest biomass,higher accuracy estimates.