Forest Above-ground Biomass Estimation Using Feature Selection Based On Remote Sensing Data

Forest above-ground biomass(AGB)plays an important role in terrestrial system carbon cycles and climatic variations.According to previous studies in forest AGB estimation,it is conventional to improve estimation accuracy using multi-source remote sensing data and their derived features.But these methods are always apt to introduce the problems of high-dimension of data,information redundancy,model overfitting.Therefore,it is urgent to develop a steady,high-efficient method in feature selection.Aiming at the over-fitting problem caused by information redundancy from multi-source remote sensing data and their derived high-dimensional features,this study is to effectively pre-select the optimal feature combination to optimize the k-nearest neighbor(k-NN)for regional forest above-ground biomass(AGB)estimation.In this study,a novel features selection method for k-NN algorithm(KNN-FIFS)was proposed.This method iteratively pre-select the optimal features which determined by the minimum root mean square error(RMSE)between the measured forest AGB values and the k-NN estimates using the leave-one-out(LOO)cross-validation.In this study,KNN-FIFS was used to estimate forest AGB using multi-source data,including Landsat-8 OLI and its vegetation indices,K-T transformations,texture metrics,topographic factors,HV polarization of P-band synthetic aperture radar(SAR)data,and forest inventory data over Genhe forest reserve located in Inner Mongolia.Afterwards,forest AGB estimation accuracy comparison was conducted among KNN-FIFS,stepwise multiple linear regression(SMLR),k-NN cooperated with Pearson correlation coefficient(Pearson+k-NN),random forest(RF+k-NN)and support vector machine(SVM).The results showed that KNN-FIFS(R2=0.77,RMSE=22.74 t·ha-1)performed better than SMLR(R2=0.53,RMSE=32.37 t·ha-1),Pearson+k-NN(R2=0.60,RMSE=30.58 t·ha-1),RF+k-NN(R2=0.63,RMSE=28.54 t·ha-1)and SVM(R2=0.66,RMSE=27.70 t·ha-1).More important,the KNN-FIFS algorithm can quickly pre-select the optimal features from the high-dimensional data and therefore largely improve the estimating efficiency.By this automatic feature selection,KNN-FIFS can pre-select the optimal feature combination to estimate regional forest AGB using multi-mode remote sensing data with high-dimensional information,which also provides the promising way to operationally apply the multi-source remote sensing data to estimate the other forest structural parameters.