A fast iterative features selection for the K-nearest neighbor

Abstract

Recently, multi-source remote sensing data and their derived features such as vegetation indices, texture metrics have been frequently applied to quantitatively estimate forest above-ground biomass (AGB). However, it is still challenging to efficiently select the optimal features for modeling the forest AGB. In this study, a fast, efficient and automatic method has been proposed, called as k-nearest neighbor with fast iterative features selection (KNN-FIFS). This method iteratively pre-select the optimal features which determined by the minimum root mean square error (RMSE) between the forest field data and the k-nearest neighbor (k-NN) estimates based on the leave-one-out (LOO) cross-validation. By use of KNN-FIFS and multisource data, including Landsat-8 OLI (operational land imager) and its vegetation indices, texture metrics, HV polarization of P-band Synthetic Aperture Radar (SAR) data (P HV ), and forest inventory data, were applied to estimate forest AGB over Genhe forest reserve located in Inner Mongolia, China. Afterwards, the model behaviors between KNN-FIFS and stepwise multiple linear regression (SMLR) methods were compared, which showed that the KNN-FIFS method (R2 = 0.77 and RMSE = 22.74 t·ha−1) was superior to the SMLR method (R2 = 0.53 and RMSE = 32.37 t·ha−1).