The modified external parameter orthogonalization with removed PC2 to remove effectively the moisture effect on the spectra

In situ visible–near infrared spectroscopy holds great potential in providing information supporting field applications, decision making and management in soil science, especially combined with the information present in the archived soil spectra. However, soil moisture can drastically affect the reflectance curve and reduce prediction accuracy. The external parameter orthogonalization (EPO) can remove the moisture effect but the effect of elimination is urgently needed to be improved. Herein, firstly, we implemented EPO on 50 bootstrapped calibration datasets, which generated from the local soil spectra library, with 255 combinations of PC₁ to PC₈, resulting in a total of 50*255 models in PLSR and cubist, respectively. Secondly, we calculated the mean prediction results from these 255 combinations and selected the top 10 validation performance results. Thirdly, we performed correlation analysis on SOM with each segment induced by singular value decomposition on the difference of laboratory and in situ spectra to determine which PC should be removed in the modified EPO. Results revealed that top 10 prediction results with principles without PC₂ and the moisture effect was mainly in PC₁, and PC₂ with significant correlation with SOM were removed from the EPO procedure. EPO with removing PC₂ (namely Modified EPO) improved the correlation of SOM with some optional bands that directly and indirectly were associated with SOM to improve the SOM prediction accuracy. Modified EPO improved the accuracy in predicting SOM with increased R² (9 %–44 % and 7 %–17 %) and root mean square error (1 %–9 % and 63 %–68 %) in the PLSR and Cubist model, respectively. Our study highlights the advantage of modified EPO in improving the elimination efficiency of water in spectra, and of PC analysis biplots in approximating the removed PCs.