Reducing the Moisture Effect and Improving the Prediction of Soil Organic Matter With VIS-NIR Spectroscopy in Black Soil Area

Soil organic matter (SOM) is rich in black soil area. It has a significant effect on soil health and luxuriant vegetation. The agricultural production activities can thus be guided by measurements of SOM at different periods. Most of previous work estimated SOM using reflectance spectra measurements on dried soil samples, to remove the effect of soil moisture content (SMC). However, it is time-consuming and almost conduct-prohibitive in the field. In this study, we utilized the continuous wavelet transform (CWT) on visible (VIS) and near infrared (NIR) spectra to test its effect on eliminating the SMC influence, and proposed to use the PCA-RF method coupled with CWT to predict SOM from the wet samples. Multi-scale coefficients can amplify the response of spectra to SOM in varying degrees, while improving the correlation of characteristic wavebands and minimizing the moisture interference on SOM specific wavelength. By analysing the multi-scale coefficients, we found that wavelengths ranging around the peaks of $580nm$ , $820nm$ , and especially the narrow region around $1400nm$ are highly correlated regions to SOM. Furthermore, the accuracy of SOM estimation models illustrated the effectiveness of the CWT. Results of the validation model using the dataset of wet samples on CWT scale 6 ( $R^{2}$ = 0.84, $Mse$ = 0.23%, and $RPD$ = 2.53) can be statistically equivalent to dataset of dried samples ( $R^{2}$ = 0.86, $Mse$ = 0.20%, and $RPD$ = 2.68). Combined with the PCA-RF method, SOM estimation can be perfectly performed with fewer features as input variables and has a great improved. The best prediction of validation model was on scale 6 with features extracted from 4 PCs, compared with the EPO-PLSR method ( $R^{2}$ = 0.85, $Mse$ = 0.25%, and $RPD$ = 2.54), the proposed method has a better result ( $R^{2}$ = 0.94, $Mse$ = 0.09%, and $RPD$ = 4.08).