Prediction of the soil hydraulic properties from hyperspectral measurements: Two case studies in Campania.

Distributed three-dimensional Richards-based hydrological models describe the experimental field as a numerical grid discretized by a myriad of cells. To simulate soil water flow and solute transport processes, it is necessary to gain information in each model cell about the soil hydraulic properties (SHPs), namely the soil water retention (WRF) and hydraulic conductivity (HCF) functions. The use of soil spectroscopy in the visible, near-infrared, and shortwave infrared (Vis-NIR-SWIR) range is increasing at an unprecedented rate as a cost-effective and non-destructive technique to predict basic soil physico-chemical properties (bSPCPs, i.e. sand and clay contents, soil bulk density, pH, and soil organic carbon content), which are primary input variables in pedotransfer functions (PTFs) to estimate the SHPs. In this study, the spectral reflectance of 3,316 grinded oven-dry soil samples was measured in the vis-NIR-SWIR ranges (350–2500 nm) together with the soil physical/chemical properties in the administrative region of Campania (southern Italy). Stepwise multiple linear regression coupled with the bootstrap method was used to construct a soil spectral library (SSL) to estimate the sand, clay, organic matter, and bulk density. Transferability of the Campania SSL was evaluated in two independent areas of Campania: i) the Sele River plain, mainly covered by annual and perennial crops, and ii) the sub-catchment of Monteforte Cilento (MFC2) located in the Upper Alento River Catchment, which is representative of an agroforestry ecosystem. In the Sele River plain, a total of 88 soil samples were collected for spectral measurements and the determination of bSPCPs. In MFC2, a total of 160 soil samples were collected to perform spectral measurements and the determination of both bSPCs and SHPs. In MFC2, we tested the PTFs of Weynants et al. (2009) and Rawls and Brakensieck et al. (1985) based on the van Genuchten (vG) and Brooks-Corey (BC) soil hydraulic models, respectively when the bSPCPs were either measured or estimated from the spectral measurements. The prediction performance was evaluated in terms of root mean squared error (RMSE) and coefficient of determination (R2). Overall, our findings indicate that spectral data coupled with existing PTFs can provide useful information to predict the SHPs by exploiting an SSL developed over a large territory (Campania, southern Italy). Moreover, there is a need to extend and validate the derived transfer functions outside the region of Campania.