Detection of total potassium and total phosphorus in soil in GAN NAN navel orange orchard using near infrared spectroscopy

Abstract: To study the distribution of soil nutrients and build soil models of the total potassium (TK) and the total phosphorus (TP) that could predict the measured value, the soil samples coming from GAN NAN navel orange orchard were collected. The precision of the soil moisture measurement using near-infrared spectra and quantitative analysis model method on the sample condition, soil samples were air-dried and sieved through 0.149 mm screen holes after grinding. The portable spectroradiometer of BRUKER TENSOR 37 with a full spectral wavelength of 400-2500nm, was used to scan the soil samples with diffuse reflectance spectroscopy and the data validity of the original spectra was averaged. Fifty-nine soil samples were selected, and thirty-seven soil samples were used to build the calibration model and nineteen were used to build the prediction model. Two kinds of data pretreatment methods including Savizky-Golay smoothing and the first order derivative were used to pretreat the soil sample spectra. The preprocessed by the combination of first-order derivative and moving average filter were used and the calibration models were developed by the partial least square regress (PLS), principal component regression (PCR) and least squares support vector machineLS-SVMbased on the spectral data and measured values, which the models could quickly and accurately estimate the soil contents of total potassium (TK) and the total phosphorus (TP) in the wave-number of 4000-7500cm-1. The model accuracy was evaluated using the correlation coefficient of prediction (RP), the root mean square error of prediction (RMSEP), and the residual predictive deviation (RPD). The results show that the least squares support vector machine (LS-SVM) model of total phosphorus (TP) gave the best results with the correlation coefficientRpof 0.884, the root mean square error of prediction (RMSEP) of 0.341, and the residual predictive deviation (RPD) of 2.59. The least squares support vector machine (LS-SVM) model of the TK gave the best result with the correlation coefficient (Rp) of 0.971 and the root mean square error of prediction (RMSEP) of 0.714, the residual predictive deviation (RPD) with the higher being the better, and the high value measured at 5.12. The experiments showed that the diffuse reflectance near infrared can be quickly and accurately estimated to the TK contents and the TP contents in soil samples using the least squares support vector machine method and this study provided a scientific basis for quickly detection of soil TP and TK by near infrared spectroscopy technology.