The interaction effect of moisture content and texture on the measurement of soil organic carbon using visible and near infrared spectroscopy

Abstract

This study was carried out to evaluate the interaction effect of soil moisture content (MC) and texture on the measurement accuracy of soil organic carbon (SOC). Two sample sets collected from Turkey (150 samples) and the United Kingdom (106 samples) were divided into two texture classes, namely, light (clay content ≤28%) and heavy (clay content > 28%) classes. The diffuse reflectance spectra of the 256 soil samples were measured under 6 gravimetric MC levels of 0, 5, 10, 15, 20 and 25% using a fibre-optic vis-NIR spectrophotometer (350 - 2500 nm). Partial least squares (PLS) regression with full cross-validation was carried out on the calibration sets to establish models for SOC for each soil texture class at each individual MC level and all MC levels. The results showed that the texture and MC affect the measurement accuracy of SOC. ANOVA proved the individual and combined effect of MC and texture were insignificant. The accuracy of the SOC models was generally higher for heavy soils (a lower average root mean square error of prediction (RMSEP) of 1.42% and a higher ratio of prediction deviation (RPD) of 2.57) than for light soils (a higher average RMSEP of 1.58% and a lower RPD of 2.36). It can be concluded that splitting samples into heavy and light soils is recommended for modelling of SOC in heavy soils only. For the measurement of SOC in light textures under laboratory conditions, soil samples should be dried before scanning, whereas when fresh soils are used for a quick analysis, scanning is recommended for any MC lower than 15%.