|CHO, YONGJIN - University Of Missouri|
|SHERIDAN, ALEXANDER - University Of Missouri|
|Sudduth, Kenneth - Ken|
Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/27/2017
Publication Date: 10/24/2017
Citation: Cho, Y., Sheridan, A.H., Sudduth, K.A., Veum, K.S. 2017. Comparison of field and laboratory VNIR spectroscopy for profile soil property estimation. Transactions of the ASABE. 60(5):1503-1510.
Interpretive Summary: Quantification of profile soil properties such as soil carbon, nitrogen, bulk density, and texture is traditionally accomplished by collection of soil samples in the field and subsequent laboratory analysis. This process is inefficient and may be impractical when measurements are needed at many locations and depths across a field or landscape, for example to develop precision agriculture management plans, to quantify spatial variability in soil quality, or to drive spatially-explicit process-based models. Prior research has established laboratory-based visible and near-infrared (VNIR) reflectance spectroscopy as a more efficient alternative, but this approach still requires collection of soil samples. Field-deployable spectroscopy sensors that can collect data at multiple depths in the soil profile are a potential solution to this problem. This research evaluated one such sensor in comparison to standard laboratory measurements and laboratory-based spectroscopy. Soil samples and in-situ spectral scans were collected from a research site with variable soils. Generally, the most accurate spectral estimates of soil properties were obtained in the laboratory. Some soil properties, including organic carbon, total nitrogen, and bulk density were estimated in the field with errors at most 25% higher than those obtained in the laboratory. On the other hand, estimation errors of soil texture fractions were almost doubled for field estimates. Thus, considering efficiency advantages, field in-situ reflectance spectroscopy appears to be a viable alternative to laboratory spectroscopy for some, but perhaps not all soil properties. Further research is needed to verify these findings across a wider range of soil conditions. These results will benefit researchers and practitioners interested in using spectral sensing approaches to improve the efficiency of quantifying spatially-variable soil properties.
Technical Abstract: In-field, in-situ data collection with soil sensors has potential to improve the efficiency and accuracy of soil property estimates. Optical diffuse reflectance spectroscopy (DRS) has been used to estimate important soil properties, such as soil carbon, nitrogen, water content, and texture. Most previous work has focused on laboratory-based visible and near-infrared (VNIR) spectroscopy using dried soil. The objective of this research was to compare estimates of laboratory-measured soil properties from a laboratory DRS spectrometer and an in-situ profile DRS spectrometer. Soil cores were obtained to approximately 1 m depth from treatment blocks representing variability in topsoil depth located at the South Farm Research Center of the University of Missouri. Soil cores were split by horizon and samples were scanned with the laboratory DRS spectrometer in both field-moist and oven-dried conditions. In-situ profile DRS spectrometer scans were obtained at the same sampling locations. Soil properties measured in the laboratory from the cores were bulk density, total organic carbon (TOC), total nitrogen (TN), particulate organic matter carbon and nitrogen (POM-C and POM-N), water content, and texture fractions. The best estimates of TOC, TN, and bulk density were from the laboratory DRS spectra on dry soil (R2 = 0.94, 0.91, and 0.71, respectively). Estimation errors with the field DRS system were at most 25% higher for these soil properties. For POM-C and POM-N, the field system provided estimates of similar accuracy to the best (dry soil) laboratory measurements. Clay and silt texture fraction estimates were most accurate from laboratory DRS spectra on field-moist soil (R2 = 0.91 and 0.93, respectively). Estimation errors were almost doubled with the field DRS system. Considering the efficiency advantages, field, in-situ DRS appears to be a viable alternative to laboratory DRS for TOC, TN, POM-C, POM-N and bulk density estimates, but perhaps not for soil texture estimates.