Search

• Data from a simple color sensor and VisNIR DRS were combined to predict soil OC. • Variables measured by the color sensor showed negative and non-linear effects on OC. • PLSR produced satisfactory OC prediction using DRS spectra alone. • The combined model outperformed model from either sensor. Contemporary soil characterization is increasingly dependent upon proximal sensor data whereby high sample throughput and low-cost analysis are realized. Recent research studies have shown that combined sensor platforms generally offer greater predictive model stability and increased accuracy than the use of sensors in isolation. In this study, data from an inexpensive ($350 USD) Nix Pro color sensor, which measures the true color of an object by using red, blue, and green filters, was used with diffuse reflectance spectroscopy (DRS) (590–2490 nm) to predict soil organic carbon (OC) content in highly disturbed, landfill soils of India ex-situ. Generalized additive model (GAM) and partial least squares regression (PLSR) were applied to model DRS and Nix Pro data, respectively, both independently and by combining model predictions using a bilinear regression. Results showed that the combined model outperformed either sensor independently where the 30% external test set achieved a validation R2 of 0.95, residual prediction deviation (RPD) of 4.54, and the ratio of performance to interquartile range of 6.25 relative to laboratory-measured OC data. In contrary, the GAM-OC model using Nix Pro data alone and the PLSR-OC model using DRS data alone exhibited validation R2 values of 0.58 and 0.81, respectively. In sum, the addition of the inexpensive Nix Pro sensor substantively improved the prediction of soil OC relative to the use of DRS in isolation. Future studies should evaluate the effectiveness of such an approach on a wider variety of soil types (e.g., colors), its effectiveness in-situ under variable moisture conditions, and in possible combination with other proximal sensing platforms. [ABSTRACT FROM AUTHOR]