Synergistic Retrieval Of Forest Aboveground Biomass Using GEDI And Multispectral Data

Forests are complex ecosystems consisting of plants,animals,microorganisms,and non-living components.They are of great significance to the Earth and humanity.The height of the forest canopy and aboveground biomass are important indicators for assessing the structure and function of forest ecosystems,which directly reflect forest productivity and ecosystem status.Accurate estimation of forest canopy height can help us better understand forest structure and biomass distribution,providing valuable references for forest management and protection.This paper presents a study based on remote sensing data including airborne LiDAR,Landsat series OLI and OLI-2 data,GEDI L2A data,SRTM data,and aboveground biomass density products.Multiple remote sensing factors were extracted to build multiple linear regression,random forest,neural network,and support vector machine aboveground biomass inversion models,which achieved remote sensing inversion of forest aboveground biomass in the study area.The research contents and conclusions are as follows:(1)Data preprocessing.According to the needs of forest aboveground biomass inversion,various data were introduced and preprocessed.The airborne LiDAR data extracted the digital surface model(DSM)and digital terrain model(DTM)and calculated the canopy height model(CHM)by subtracting the two.OLI and OLI-2 data were calibrated for radiometric and geometric correction,and band reflectance,band combinations,and texture factors of sample points were extracted.GEDI L2A data extracted ground elevation and relative height indicators through a series of filtering conditions.SRTM data extracted slope,aspect,and terrain factors of ground elevation.The aboveground biomass density product and airborne LiDAR data were used to develop a model to infer the aboveground biomass,and the coefficient of determination R~2reached 0.74,which was used to predict the aboveground biomass in the study area in2022 and participate in the accuracy verification of the aboveground biomass inversion model.(2)Ground elevation inversion.Firstly,the height reference conversion was performed on the DTM of airborne LiDAR data to use the same height reference as GEDI.Then,the"elev＿lowestmode"parameter of the six algorithms of GEDI L2A data was used to invert ground elevation.The results showed that the R~2 of the six algorithms were all above 0.99,with algorithm a2 having the highest accuracy and algorithm a5 having the lowest accuracy,with an RMSE of 9.00 m.An optimized algorithm proposed in this paper further improved the accuracy of ground elevation inversion based on algorithm a2,with an MAE of 1.58 m,RMSE of 3.67 m,and r RMSE of 1.02%.(3)Canopy height inversion in forest based on GEDI L2A data.The 90th percentile CHM from airborne LiDAR data was used as validation data.Through analysis of the relative height inversion results obtained by the default algorithm of GEDI L2A data,it was found that the height value at the 97%cumulative return energy(rh97)had the highest accuracy.Based on preliminary exploration,this paper analyzed the canopy height inversion results of six algorithms in GEDI L2A data at rh97,and found that algorithm a2 had the highest accuracy,with an R~2 value of 0.44,MAE of 3.34 m,RMSE of 5.20 m,and r RMSE of 24.54%,while a4 and a5 had the lowest accuracy.After removing algorithms a4 and a5,the average rh97 of the remaining four algorithms was calculated,and the accuracy of forest canopy inversion was further improved.The inversion indicators were R~2 value of 0.47,MAE of 3.29 m,RMSE of 5.07 m,and r RMSE of23.09%.(4)Synergistic inversion of forest aboveground biomass based on multiple data sources.This paper discusses the method of synergistic inversion of biomass based on multiple data sources.For single data sources,GEDI,OLI,and OLI-2,the neural network inversion model of GEDI had the highest accuracy among the four regression inversion models,with an R~2 value of 0.68,MAE of 14.30 t/hm~2,RMSE of 19.42 t/hm~2,and r RMSE of 10.65%,while the improvement in data quality of OLI-2 compared to OLI did not lead to higher inversion accuracy.Based on the synergistic inversion of multiple data sources,this paper investigated four inversion models using GEDI&OLI,as well as GEDI&OLI-2.The results showed that for GEDI&OLI,the neural network regression model with tanh as the activation function had the highest inversion accuracy,with an R~2 value of0.71,MAE of 13.85 t/hm~2,RMSE of 18.53 t/hm~2,and r RMSE of 10.16%.For GEDI&OLI-2,the support vector machine model with RBF as the kernel function had the highest inversion accuracy,with an R~2 value of 0.72,MAE of 13.25 t/hm~2,RMSE of 18.39 t/hm~2,and r RMSE of 10.08%.Through synergistic inversion of multiple data sources,the accuracy of forest aboveground biomass inversion was improved compared to that of the inversion models based on single data sources.In future research,the feasibility of more data combinations can be explored to conduct regional and global-scale forest studies.