Abstract:
Abstract: Soil stores more carbon in the terrestrial ecosystem than the combined vegetation and atmosphere. Soil organic carbon (SOC) as the key component of soil carbon pool is highly sensitive to earth surface evolution and anthropogenic-induced changes in climate and agricultural management practices. The spatiotemporal dynamics can exert important controls over soil productivity and ecosystem services. There is thus an increasing demand to quantify SOC at sufficiently high resolution and accuracy, thereby detecting localized soil degradation as well as ensuring sustainable agricultural management. Field-, airborne and satellite-based multi-platform Visible and Near-Infrared (Vis-NIR) reflectance spectroscopy has increasingly been used as a fast and effective tool to predict SOC, and thereby capture the variability at field to landscape scales. Comparing to the satellite-based remote sensing systems, commercially available portable Unmanned Aerial Vehicle (UAV) equipped with high-resolution Vis-NIR spectrometers can greatly improve the spatial resolution and acquisition efficiency of soil spectral information. It is also more flexible to carry out field surveys thanks to the small size, but applications of UAV-based spectroscopic assessment of SOC so far are still scarce. In this study, a UAV-compatible soil hyperspectral data acquisition platform was tested in two types of soil located in the Northeastern Black Soil Belt of China and the Belgian Loam Belt. The specific objectives were: 1) to test the ability of UAV-compatible Vis-NIR spectrometer for the accurate prediction of SOC content; and 2) to explore a spectral correction approach in a laboratory-based spectral model under field conditions. Soil hyperspectral data was gathered in a dark room and under natural sunlight. Subsequently, spectral-based SOC prediction models were developed using Partial Least Squares Regression (PLSR). Results show that: 1) PLSR models behaved excellent performances for both study sites using UAV-compatible spectral data (FX) from a dark room with the Relative Percent Difference (RPD) higher than 1.6 and R2≥0.65. 2) FX spectral data acquired under natural sunlight also achieved an acceptable PLSR model (RPD=1.48, R2=0.58) suitable for capturing the range of variation in SOC, although the accuracy slightly decreased, compared with the dark room. 3) A standard sand sample from Lucky Bay (Australia) was selected to correct and align the FX spectral data under two light conditions. The PLSR model using the laboratory spectra was directly applied to field spectra for the excellent performance (R2 = 0.53, RMSE= 0.29%, RPD = 1.45, RPIQ = 1.75). The spectral correction approach can offer promising potential in future applications to avoid the large sampling, when using UAV-based spectroscopy to rapidly assess SOC. This finding highlighted the UAV-based hyperspectral remote sensing to predict SOC in a fast, accurate and detailed fashion, providing technical reference in fields, such as digital soil mapping and precision agriculture. Future studies can explore the influence of soil surface roughness and moisture on the quality of soil spectral data acquired from UAV platforms, thereby correcting for the noise caused by external factors.