Mitigating moisture-induced spectral distortions in soil spectroscopy through optimised reference panel calibration

Soil spectroscopy enables the rapid, non-destructive estimation of key soil properties, but it conventionally relies on air-dried samples measured against a white reference (W R , ≈100% reflectance) panel. When soils are wet, refractive-index and multiple-scattering effects darken the surface and suppress diagnostic absorptions, while the wide dynamic range of the W R panel further compresses these low-reflectance features; this reduces predictive accuracy and hinders the harmonisation of spectra acquired under different moisture conditions. Because each reference panel is treated as the 100% reference for its own measurement session, replacing W R with a lower-albedo panel is an acquisition-stage radiometric transformation that expands the effective dynamic range for low-reflectance (wet or dark) targets without modifying the soil sample itself. This study therefore re-evaluates calibration practice under wet conditions and tests whether lower-albedo gray (GR, mid-reflectance ≈50%) and black (BR, low-reflectance ≈20%) reference panels improve the prediction of soil constituents relative to W R , without drying. Using 3928 samples from the Israeli Legacy and Israeli Soil Spectral Libraries, robust dry-soil models (W R ) were first established for soil organic carbon (SOC), dithionite-extractable Fe (Fe_d), and CaCO₃, confirming high accuracy under air-dried conditions but substantial degradation at 20% moisture regardless of reference panel when trained only on dry spectra. To isolate calibration effects, 91 soils (training) were adjusted to 20% gravimetric moisture to develop wet-condition models, which were then applied to 223 independent soils at 20% moisture under W R , GR, and BR calibration. GR consistently yielded the highest performance (SOC R² = 0.66 vs. 0.48 for W R ; Fe_d R² = 0.75 vs. 0.67; CaCO₃ R² = 0.83 vs. 0.79), with concurrent RMSE reductions. When the GR-trained wet library (20%) was applied across 5–30% moisture, GR-based measurements maintained robust accuracy, whereas W R remained unstable and BR showed attribute-dependent variability. We conclude that constructing a dedicated wet spectral library at ≈ 20% gravimetric moisture using a gray reference panel provides a practical and transferable acquisition-stage strategy for wet-soil evaluation within the tested soils and moisture range. As the study is laboratory-based, the approach is presented as a proof of concept that warrants field validation.