On-site rapid quantitative assessment of wet soil nutrients based on laser-induced breakdown spectroscopy fingerprint spectral lines corrected by near-infrared spectroscopy

• A new on-site rapid optical detection method for wet soil nutrients was proposed. • NIR spectroscopy corrected LIBS to eliminate the effect of moisture on detection. • Soil nutrient detection models based on LIBS fingerprint lines correction were built. • Calibration models can detect wet soil nutrient rapidly and accurately on-site. Rapid acquisition of soil nutrient information is of great significance for precision agriculture, and sample moisture difference is a major challenge to achieve rapid on-site optical detection of soil nutrient. To realize the on-site rapid quantitative detection of wet soil nutrients, this study took the soil total potassium detection as an example, and the soil moisture content information detected by near-infrared (NIR) spectroscopy was used to correct the fingerprint spectrum lines of laser-induced breakdown spectroscopy (LIBS), so as to reduce the influence of water on sample LIBS detection. Firstly, for the two potassium detection LIBS fingerprint spectra with the wavelength of 766.49 nm and 769.90 nm, the exponential mapping relationship models between the correction factors of the above two LIBS fingerprint lines and the moisture content of wet soil were constructed, and the mapping models were y = 1.143 × e - 0.106 x and y = 1.106 × e - 0.102 x , respectively. Secondly, partial least squares regression (PLSR) was used to establish soil moisture NIR spectroscopy detection models of different preprocessed spectral data. Through comparison, the correlation coefficients of the calibration set ( R C ) and test set ( R T ) of the best soil moisture content NIR spectroscopy detection model were 0.9926 and 0.9871, and the root mean square errors of the calibration set (RMSEC) and test set (RMSET) were 0.84 % and 1.14 %, respectively. Finally, the calibration models of soil total potassium spectrum detection based on LIBS fingerprint lines data corrected by NIR spectroscopy were constructed. The determination coefficients ( R 2 ) of the calibration models established based on 766.49 nm and 769.90 nm fingerprint lines correction data were 0.9776 and 0.9788, respectively. The root mean square errors of cross-validation (RMSECV) were 0.0510 % and 0.0499 %, and the average relative errors (ARE) were 13.3424 % and 13.0500 %, respectively. In addition, the excellent detection performance of the above two calibration models was further verified by the rapid detection of actual wet soil samples. These preliminary research results showed that the on-site rapid optical quantitative detection of total potassium content in wet soil could be realized by the detection strategy based on LIBS fingerprint lines corrected by NIR spectroscopy. This study would provide a reference for the on-site rapid detection of samples with significant moisture differences such as wet soil.