RAPID PREDICTION OF AVAILABLE K CONTENT IN SOIL USING NEAR-INFRARED SPECTROSCOPY

Abstract

Soil NPK testing has been widely used for fertilizer recommendation of annual crops. Soil analyses of available K content by chemical methods are sufficiently accurate, but they are expensive, time consuming and labour intensive. Recently, fertilization recommendation to the farmers is based on reduced number of soil samples due to high price of analyses. For this reason a rapid and cost-effective soil analysis is needed for soil quality assessment. Near infrared spectroscopy (NIRS) could provide a possible alternative. The objective of this study was to investigate the possibilities of NIRS for prediction of available potassium content in different soil units. A total 191 samples from four soil type- Chernozems, Vertisols, Luvisols and Fluvisols were analyzed for available K by conventional chemical method. NIR spectra of all samples were obtained by using an InfraAlayzer 450 spectrophotometer within the range 1445-2348 nm and portable FQA-NIRGun scanning spectrophotometer in shortwave NIR range from 600 to 1100 nm. SIMCA- soft independent modeling of class analogy was performed to classify samples, according to soil type. MLR and PLS regression were used for calibration models development for available K determination. The best model was obtained for samples of Chernozems with correlation coefficient R=0,94, standard error of calibration SEC=2,30 mg/100g, and the ratio of the standard variation of the reference data to the SEC, indicating the performance of the calibration RPD=3,1. Accuracy of determination of K content for models for separated soil units, developed by spectral data in short-wave NIR region, was better to accuracy of models, based on filter-type instrument in classical NIR region. Correlation coefficients for the global calibrations containing the samples of all soil units’ decrease and SEC or SEP increase compared to calibrations for separate soil units. The values of R were between 0,76 and 0,79 and RPD were between 1,5-1,7 for the both calibration and test set. Calibration models for each soil type increase accuracy of determination of available potassium content.