UAS-based soil carbon mapping using VIS-NIR (480–1000 nm) multi-spectral imaging: Potential and limitations

Abstract

Traditional methods to assess the soil organic carbon (SOC) content based on soil sampling and analysis are time consuming and expensive, and the results are influenced by the sampling design. The aim of this study was to investigate the potential of UAS (Unmanned Aerial Systems) multi-spectral imagery (480–1000nm) for estimating the SOC content in bare cultivated soils at a high spatial resolution (12cm). We performed UAS analysis on the Hoosfield Spring Barley experiment at Rothamsted (UK) where adjacent plots with distinctly different SOC contents, due to different long-term management practices, provide a valuable resource to evaluate this approach. We acquired images (wavelength: 480–550–670–780–880–1000nm) at an altitude of 120m over an area of 2ha using a multi-spectral camera mounted on an UAS. The high-resolution images captured small-scale variations at the soil surface (e.g. shadows, tillage and wheels marks). After a projection in new dimensions by a PCA, we calibrated a support vector machine regression using observations from conventional soil sampling and SOC measurements. The performance of the calibration had a R2 of 0.98 and a RMSE of 0.17%C. A cross-validation showed that the model was robust, with an average R2 of 0.95 and a RMSE of 0.21%. An external validation dataset was used to evaluate the predicted spatial patterns of SOC content and a good fit with an RMSE 0.26%C was obtained. Although this study shows that the methodology has a clear potential for use in precision agriculture or monitoring important soil properties following changes in management, we also identify and discuss its limitations and current shortcomings.