Impact of Spectral Resolution and Signal-to-Noise Ratio in Vis–NIR Spectrometry on Soil Organic Matter Estimation

In the last few decades, Visible and near-infrared (Vis-NIR) spectroscopy has emerged as a quick tool for estimating soil properties. Currently, most studies address that different pretreatments and modeling approaches affect the prediction accuracy of soil properties. However, the parameters of the spectroscopic instrument also determine the prediction performance. The focus of this study is to evaluate the impacts of the two core parameters (spectral resolution, and SNR) of spectrometers on the prediction performance of soil organic matter (SOM). To this aim, the initial soil spectral dataset (which was measured by an ASD spectrometer) was converted to construct two independent simulated soil spectral datasets that can be individually varied in terms of spectral resolution and SNR level. To verify the consistency and generality of our results, we adopted ridge regression (RR), partial least squares regression (PLSR), support vector machine regression, (SVMR), and back propagation neural network (BPNN) to develop multivariate calibration models. Finally, the ANOVA technique was performed on the RMSE and [[EQUATION]] obtained through the independent validation set to identify the factors that significantly affect the predictive performance of SOM and to determine the lowest spectral resolution and SNR level was identified without significantly affecting the prediction accuracy. Our results showed that i) the significant effects on the predictive performance of SOM in descending order of magnitude is SNR levels > spectral resolution > prediction models, ii) there is no significant difference in the predictive performance when the spectral resolution is within 100 nm and iii) when the SNR level is higher than 15%SNR, the different SNR levels do not give significant differences in the predictive performance of the SOM. In conclusion, only knowing the impact of these instrument parameters on the estimation of SOM can assist us in decision-making on the development of a dedicated, low-cost, and rapid SOM measurement spectrometer.