Soil organic carbon predictions in Subarctic Greenland by visible–near infrared spectroscopy

Abstract

Release of carbon from high-latitude soils to the atmosphere may have significant effects on Earth's climate. In this contribution, we evaluate visible-near-infrared spectroscopy (vis-NIRS) as a time-and cost-efficient tool for assessing soil organic carbon (SOC) concentrations in South Greenland. Soil samples were collected at two sites and analyzed with vis-NIRS. We used partial least square regression (PLS-R) modeling to predict SOC from vis-NIRS spectra referenced against in situ dry combustion measurements. The ability of our approach was validated in three setups: (1) calibration and validation data sets from the same location, (2) calibration and validation data sets from different locations, and (3) the same setup as in (2) with the calibration model enlarged with few samples from the opposite target area. Vis-NIRS predictions were successful in setup 1 (R 2 = 0.95, root mean square error of prediction [RMSEP] = 1.80 percent and R 2 = 0.82, RMSEP = 0.64 percent). Predictions in setup 2 had higher errors (R 2 = 0.90, RMSEP = 7.13 percent and R 2 = 0.78, RMSEP = 2.82 percent). In setup 3, the results were again improved (R 2 = 0.95, RMSEP = 2.03 percent and R 2 = 0.77, RMSEP = 2.14 percent). We conclude that vis-NIRS can obtain good results predicting SOC concentrations across two subarctic ecosystems, when the calibration models are augmented with few samples from the target site. Future efforts should be made toward determination of SOC stocks to constrain soil-atmosphere carbon exchange.

Water repellency prediction in high‐organic Greenlandic soils: Comparing vis–NIRS to pedotransfer functions

“…This agrees with results on OC vis–NIRS predictions from Ogrič et al. (2019), who obtained high accuracy predictions for area‐specific PLS‐R models in South Greenland, and Hermansen et al. (2016), who included both Danish and Greenlandic fields.…”

Soil spectroscopy with the use of chemometrics, machine learning and pre-processing techniques in soil diagnosis: Recent advances–A review

“…Applications of NIR and Vis-NIR spectral techniques with the PLS algorithm were found to be efficient for evaluating the soil organic carbon (SOC) in soils [14,[38][39][40][41][42][43][44][45][46], where the difference between all these studies are the type of soils, sample set representativeness, the number of soil samples, sample preparation, the preprocessing strategies and model validation methodologies.…”

Gas diffusion characteristics of agricultural soils from South Greenland

“…The soil resource is arguably the biggest unknown factor for the agricultural production in southwest Greenland and the majority of the soil-related research has been of an archeological and paleogeographical nature (Adderley & Simpson, 2006;Massa et al, 2012;Rutherford, 1995;Schofield et al, 2010), with only a few studies having investigated some basic physical properties of the soils. In general, the soils have been found to exhibit little to moderate soil formation (Jacobsen, 1987;Jakobsen, 1991;Rutherford, 1995), while being shallow, highly acidic, organic, and coarse-textured with a noteworthy low clay content (Adderley & Simpson, 2006;Caviezel et al, 2017;Ogrič et al, 2019). On the basis of a combination of chemical soil properties, texture, and ρ b , Caviezel et al (2017) evaluated the soil quality to be relatively poor in part of the agricultural area.…”