Aboveground Biomass Estimation Of Natural Secondary Forests Based On Ensemble Learning Algorithms

Forests are essential to our planet’s ecosystem,providing numerous ecological,social,and economic benefits.They play a critical role in mitigating climate change by sequestering atmospheric carbon dioxide and releasing oxygen.Forests also provide habitat for wildlife,regulate water cycles,and support human livelihoods.Therefore,it is necessary to accurately estimate the forest’s aboveground biomass,essential for forest management,conservation,and climate change mitigation.Based on the Landsat 8 OLI imagery,ALS data obtained from 2015,and the data of the fixed sample plots of natural secondary forests resources inventory of Maoershan Forest Farm in 2016,this study applied three topographic correction methods including Sun Canopy Sensor+C correction(SCS+C),Variable Empirical Coefficient Algorithm(VECA),and Minnaert to Landsat 8 OLI imagery,and then,utilized the five ensemble learning algorithms including simple averaging(SA),weighted averaging(WA),stacked generalization(SG),random forest(RF),and extreme gradient boosting(XGBoost)and adopted two cross-validation approaches including10-fold cross-validation and leave-one-out cross-validation to investigate the influence of topographic correction methods,ensemble learning models,and cross-validation approaches on the accuracy of the AGB estimation in natural secondary forests(NSFs).The results indicated that:(1)Among the three topographic correction methods,the SCS+C correction improved the accuracy of AGB estimation to a higher degree than other topographic correction methods;thus,the SCS+C correction was chosen as a suitable topographic correction method in Maoershan Forest Farm.(2)The SG algorithm based on decision tree(DT),k-nearest neighbor(KNN),support vector regression(SVR),and convolutional neural network(CNN)base models was the most influential ensemble learning algorithm and more accurate than other ensemble learning algorithms on the AGB estimation of NSFs.Compared to the four different algorithms,the SG algorithm had the highest accuracy regardless of the topographic correction methods and cross-validation approaches.Using the SCS+C correction and leave-one-out cross-validation approach,the SG had the best performance of R~2=0.98,RMSE=7.48 t/ha,r RMSE=0.05,MAE=3.25 t/ha based on Landsat 8 OLI imagery and ALS data.(3)For model validation approach,it was found that the AGB estimation accuracy using the leave-one-out cross-validation approach was significantly higher than that using the10-fold cross-validation approach based on the same prediction model.It indicated that the leave-one-out cross-validation approach may provide over-optimistic AGB estimates than 10-fold cross-validation approach.Therefore,it is important to consider both prediction models and validation approaches in AGB estimation.In practice,researchers should be cautious whether the high AGB estimation accuracy results from the specific prediction model or the validation approach.In summary,this study investigated that the effect of three topographic correction methods,five ensemble learning algorithms,and two cross-validation approaches on the AGB estimation of NSFs,and found that a combination of SCS+C correction,SG algorithm,and leave-one-out cross-validation approach provided the highest AGB estimation accuracy.The study provides the scientific foundation for the feasibility of the different topographic correction methods,ensemble learning algorithms and validation approaches for accurately estimating the AGB of NSFs in northeast China.