Application of a Handheld Near Infrared Spectrophotometer to Farm‐Scale Soil Carbon Monitoring
Jonathan Sanderman, Colleen Smith, José Lucas Safanelli, Keith Shepherd, Yufeng Ge, Sadia Mannan Mitu, Richard Ferguson
Abstract
ABSTRACT Recent advances in hardware technology have enabled the development of handheld sensors with comparable performance to laboratory‐grade near‐infrared (NIR) spectroradiometers. In this study, we explored the effect of the uncertainty from the NeoSpectra Scanner Handheld NIR Analyzer (Si‐Ware) on estimating farm‐level soil organic carbon (SOC) stocks at three small farms in Massachusetts, USA. A field campaign conducted in Falmouth, MA, collected 192 soil samples from three farms at depths of 0–10, 10–20 and 20–30 cm. All samples were scanned both in the field at field moisture and under laboratory conditions after being dried and sieved. Samples were analysed for SOC via elemental analysis, while bulk density was determined after weighing the dry fine earth sampled with cylindrical cores in the field. Several strategies for spectral prediction were tested for estimating SOC content and bulk density (BD) using both moist and dry scans, including testing the application of prebuilt models from the Open Soil Spectral Library. Cubist was used to train all models, and conformal prediction was used to estimate the prediction intervals to one standard deviation. The Cholesky decomposition algorithm allowed us to consider the correlation between variables over the three depth layers during uncertainty propagation with Monte Carlo to come up with robust estimates of field‐scale SOC stocks and uncertainty. This analysis revealed that spectroscopy predictions, although less precise, can detect the same statistical patterns in SOC stock across farms at a large cost savings compared with the traditional analytical methods.