Development of hyperspectral model for rapid monitoring of soil organic carbon under precision farming in the Indo-Gangetic Plains of Punjab, India

Abstract

Degrading soil quality at an alarming rate as a result of high input agriculture under continuous rice-wheat cropping system in the Indo-Gangetic alluvial Plains of Punjab (India), a major food growing region of south-east Asia, has ushered the need of precision farming for which rapid site specific monitoring of soil organic carbon (an indicator of soil quality) is needed. In this study, visible-near infrared reflectance spectroscopy was evaluated for rapid prediction of soil organic carbon (SOC) contents in soils of the Indo-Gangetic alluvial Plains of Punjab, India. A total of 800 surface soil samples (480 for calibration and 320 for validation) from farmers’ field representing the districts of Ludhiana, Moga, Gurdaspur and Bhatinda in Punjab State, India were collected, ensuring sufficient variation in SOC content. Reflectance spectra were obtained from air-dried samples (<2 mm size) under controlled laboratory conditions using a hyperspectral ASD FieldSpecPro spectroradiometer. Part of the same samples was used for SOC determination by Walkley and Black titration method. The SOC value in the study area varies from 4.0 to 18.1 g kg−1 (mean 7.9 g kg−1 and standard deviation of 2.2 g kg−1) among the soil samples. Partial least squares regression technique was employed to examine the relationships between SOC and the reflectance spectra; and to identify the wavelengths sensitive to SOC variation. Among 15 spectral transformations used for calibration, SGF-2-3 transformation (transformation to 1st derivative with second order polynomial smoothing with 3 points using Savitzky-Golay filter) was the best for SOC modeling in the IGP soils as it showed highest validation r2 (0.81) and RPD (2.30) and the lowest RMSEP (0.116) with 6 PLS factors. The most important wavelengths for SOC prediction were 460, 470 and 550 nm in the visible and 1400, 1420, 1920, 2040, 2210, 2270, 2320 and 2380 nm in the near-infrared region. At this juncture of much awaited second green revolution envisaged to be based on sustainability and precision agriculture in one hand and the increased availability of high resolution hyperspectral satellite data on the other hand; our findings regarding rapid evaluation of SOC through hyperspectral model are encouraging as it might assist in real time evaluation of pre and post-scenarios of soil quality and sustainability under precision farming system.