VIS-NIR Reflectance Spectroscopy as an Alternative Method for Rapid Estimation of Soil Available Potassium

Abstract

Potassium (K) is an important macronutrient for crop plant and plays a crucial role in crop production. Therefore, accurate and rapid estimation of soil available K is necessary for judicious application of available K in an intensively cropped region. However, traditional soil chemical analysis for assessing soil available K is very much laborious, expensive and time consuming. The visible near-infrared (VIS-NIR) reflectance spectroscopy is considered as a promising alternative technique for rapid, non-destructive and ecofriendly estimation of available K and other soil properties. An experiment was carried out in an intensively cultivated region of Ludhiana district of Punjab to investigate the potential of VIS-NIR technique for accurate prediction of available K using multivariate model. A total of 170 georeferenced surface soil samples (0-15 cm) were collected from the study site for both chemical and spectral analysis of available K. A popular statistical technique namely, partial least square regression (PLSR) was employed to develop spectral model for K prediction. Important statistical diagnostics like coefficient of determination (R2 ), root mean square error (RMSE) and residual prediction deviation (RPD) were used to evaluate the efficacy of prediction model. The results showed that the R2 and RMSE and RPD values were 0.41, 0.09 and 1.44, respectively for independent validation dataset of PLSR model. The RPD value indicated acceptable prediction accuracy for soil available K with PLSR model. Comparatively lower performance of the studied prediction model could be ascribed to the less variation in the collected spectra of soil samples and the use of linear multivariate model. Therefore, the study suggested to explore advanced non-linear data mining techniques for achieving better prediction accuracy for soil available K