Comparing atomic spectroscopy, molecular spectroscopy and multi-source spectroscopy synergetic fusion for quantitation of total potassium in culture substrates

The rapid assessment of total potassium in a culture substrate is of great significance for scientific planting and reducing agricultural non-point-source pollution. The rapid assessment of total potassium in a culture substrate is of great significance for scientific planting and reducing agricultural non-point-source pollution. In this study, laser-induced breakdown spectroscopy (LIBS) and near-infrared spectroscopy (NIRS) were explored for the rapid detection of culture substrate total potassium at atomic and molecular scales, respectively. In addition, a new method based on atomic and molecular spectral information synergetic fusion for detection was also proposed. Single-spectrum detection models and LIBS–NIRS synergetic fusion prediction models were established. The results showed that LIBS univariate calibration curves and the NIRS detection model showed poor detection performance. The detection accuracy of models constructed from LIBS full variables, UVE-filtered LIBS variables and two LIBS variables with strong spectral line characteristics was significantly improved, but there was still great room for improvement. High-precision detection would be realized through LIBS–NIRS synergetic fusion. The simplest and optimal model would be constructed by combining the strong spectral line characteristics of LIBS and the NIRS spectral characteristics of SPA screening, and this model was conducive to the development of special detection equipment based on a photomultiplier tube as the core component of signal detection. At this moment, the number of model input variables was only 9, the determination coefficient and root mean square error of the calibration set were 0.9910 and 0.8523 g kg −1 , respectively, and the corresponding values of the prediction set were 0.9900 and 0.8802 g kg −1 . The fusion of LIBS and NIRS was feasible, and it could improve the robustness of the prediction model.