Improving the estimation accuracy of soil organic matter based on the fusion of near-infrared and Raman spectroscopy using the outer-product analysis

Accurate estimation of soil organic matter (SOM) content is of great significance for advancing precision agriculture and assessing carbon storage. Proximal sensing techniques, such as near-infrared spectroscopy (NIR) and Raman spectroscopy, provide effective means for rapidly acquiring soil information. However, quantitative estimation of soil parameters using Raman spectroscopy has been challenged by inaccurate estimation results, which has restricted the widespread application of Raman spectroscopy in SOM estimation. The fusion of complementary information from multi-sensor data has been considered as one of the feasible solutions to address the poor results of single-sensor estimation. Therefore, the study on SOM estimation based on spectral data fusion was carried out by evaluating the effects on estimation performance under different fusion strategies. In this study, 258 soil samples from the North China, along with their corresponding near-infrared spectra and Raman spectra were collected and the spectral data was fused by two strategies involved direct concatenation (DC) and outer-product analysis (OPA). The SOM estimation performance of random forest (RF) and partial least squares (PLS) models constructed based on independent spectra data (NIR spectra, Raman spectra before baseline correction, Raman spectra after baseline correction), spectral data fused by DC, and spectral data fused by OPA were evaluated, respectively. The results indicated that the fusion of near-infrared spectroscopy and Raman spectroscopy could improve the poor performance of using Raman spectroscopy independently for quantitative estimation of SOM; Furthermore, OPA was a more effective fusion strategy compared with DC, significantly improving the estimation accuracy of the model. In addition, the PLS model constructed based on OPA fused spectral data achieved the best estimation accuracy, with R2, RMSE, and RPD of 0.903, 2.594 g/kg, and 3.061 on the validation set, respectively. This study can provide a technical support for accurately estimating the content of SOM using proximal spectroscopy technologies, contributing to the improvement of soil management practices in the context of precision agriculture.