From Light Scan to Flow: Estimating Soil Hydraulic Properties Through VNIR-SWIR Spectroscopy and Calibration-Free PTFs.

This study presents a three-step approach for estimating soil-water retention (WRF) and hydraulic conductivity functions (HCF) by integrating VNIR-SWIR reflectance spectroscopy with semi-physical, calibration-free pedotransfer functions (PTFs). A total of 135 soil samples from the Alento Observatory (southern Italy) were analyzed. In the first step, diffuse reflectance spectroscopy was used to derive particle-size distribution (PSD) by combining five spectral pretreatment algorithms and three machine learning techniques. The best-performing model demonstrated excellent predictive accuracy, achieving a coefficient of determination (R2) of 0.945 and a Root Mean Square Error (RMSE) of 0.051 for PSD mass fractions. In the second step, two semi-physical PTFs—the Arya-Heitman (PTFWRF-AH) and Mohammadi-Vanclooster (PTFWRF-MV) models—were applied to estimate the WRF using spectrally-derived PSD together with measured soil bulk density and saturated water content. The PTFWRF-MV model outperformed the AH variant, yielding a lower RMSE (0.049 cm3 cm−3) and higher R2 (0.676). The third step involved estimating the HCF by applying the Arya and Heitman PTF alongside measured saturated hydraulic conductivity values. This process relied on the flow-similarity hypothesis, which assumes that water flow partitioning within pore domains is equivalent across idealized and natural-structure soils. However, the HCF predictions for both models exhibited uncertainties greater than one order of magnitude and R2 values under 0.50. These findings underscore the efficacy of spectroscopy for soil texture characterization, while highlighting persistent limitations in predicting hydraulic properties of structured soils, likely due to violations of flow-similarity assumptions.