Quantifying Soil Carbon Levels in Permafrost-Region Soils Using Mid and Near Infrared Spectroscopy.

Diffuse-reflectance Fourier-transform mid-infrared spectroscopy (MidIR) and near infrared spectroscopy (NIR) were used to develop Partial Least Squares Regression (PLSR) calibrations for total carbon in permafrost-region soils encompassing multiple horizons at 28 sites across a latitudinal gradient (78.7858N to 55.3535N deg) in Alaska. Site vegetation types included coniferous, deciduous, and mixed forests; grasslands; shrubs; moist and dry acidic and non-acidic tundra; and moss/lichen. Soils were separated into two groups: organic (O) horizons and mineral (A, B, C, and E) horizons. The objectives of the study were to: 1) test whether NIR or MidIR spectral ranges produced the best calibrations, 2) determine if a single calibration can produce good predictions for soil horizons with widely different soil carbon ranges, and 3) determine whether calibration based on selected MidIR bands rather than entire MidIR spectra can reliably predict soil carbon while reducing the amount of spectral data. PLSR of all horizons together produced good calibrations for total soil C with both NIR and MidIR, but the results were better for MidIR. The R² and standard error of the estimates (SEE) were 0.97 and ±2.21 for MidIR, and 0.93 and ±4.83 for NIR. Total carbon in organic horizons ranged from 3.6 to 49.9%, whereas mineral horizon carbon ranged from 0.24 to 42.1%. When MidIR calibrations were derived separately for organic and mineral horizons, the predictions for organic horizons were good (0.97 R² and ±3.47 SEE) but even better for mineral horizons (0.97 R² and ±1.46 SEE). The MidIR regions between 3560-3180, 2980-2830, 1890-1840, 1750-1650, 1270-1000 cm^-1 had > 0.75 R² with soil total carbon. When these regions were selected exclusively for PLSR calibration of the complete sample set, the predictive capability did not deteriorate relative to using the full 4000-650 cm^-1 spectral range (0.97 R², ±2.27 SEE), suggesting this selective approach might increase computational efficiency.