Estimation Of Aboveground Biomass Of Regional Forest Based On Multi-source Remote Sensing

As the main body of the terrestrial biosphere,forest biomass is an important indicator to measure the productivity of ecosystems,and it’s also an important basis for studying the material cycle of forest ecosystems.It also plays a pivotal role in the research of global warming and climate change.Taking Linhai County,Zhejiang Province as the study area,spectral information,vegetation index,texture feature factor of Sentinel-2 optical remote sensing data and backscattering coefficient of Sentinel-1 SAR were extracted.Moreover,the inventory data for forest management inventory planning and digital elevation model data were extracted.In order to explores the potential of multi-source data and machine learning methods in forest above-ground biomass(AGB)estimation,and analyze the main factors to influence forest AGB of seven dominant tree species.This study uses the feature selection approach based on recursive feature elimination to establish the AGB estimation model of dominant tree species.And it also uses the coefficient of determination(R~2)and root mean square error(RMSE)to evaluate the performance of the models,comparative and validate the estimation results of Random Forest(RF),Adaptive Boosting(Ada Boost)and Categorical Features Gradient Boosting(CatBoost).The results shows that multi-source data can provide importance factors for AGB and effectively improve model accuracy.Combination of recursive feature elimination method can reduce model complexity and accelerate the speed up learning rate.In terms of effect of biomass of the ABG,the CatBoost is better than RF and RF is better than Ada Boost.For different dominant tree species,the main influencing factors of AGB are different,but there are also common influences such as age,canopy density and elevation.