Rapid Assessment of Soil Traits in Hyperarid Areas via XRF and Locally Weighted PLSR

Baptiste Kerfriden^1*Stéphane Boivin¹Oscar Malou¹Yassine Fendane¹Hassan Boukcim¹Sami D. Almalki²Shauna K. Rees²Benjamin P. Y.-H. Lee²Ahmed Mohamed^2,3Abdalsamad Aldabaa^2,3

• ¹valorhiz, Montpellier, France
• ²Royal Commission for AlUla, AlUla, Saudi Arabia
• ³Desert Research Center, Al Matariyyah, Egypt

Effective soil characterization is crucial for a better understanding of ecosystem functions and for establishing ecological restoration strategies in degraded areas. However, measuring soil physical and chemical variables is usually cost-and time-consuming, which can be restrictive across large areas. X-ray fluorescence spectroscopy (XRF) has been successfully used for predicting soil variables, but has shown limits for some of them, such as soil texture in hyperarid environments. In this study, we tested the combination of centered log-ratio (CLR) transformation on XRF calculated atomic concentration data and locally weighted partial least squares regression (LWPLSR), for the prediction of soil properties in a hyperarid environment. Soil samples were collected across the AlUla region in Saudi Arabia for XRF spectra acquisition and physico-chemical analysis, such as texture, pH, carbonates content, electrical conductivity, cation exchange capacity (CEC), available macro-and micro-elements content, and soil carbon. LWPLSR construction was based on cross-validation over a calibration dataset to select the optimal number of latent variables. The models' performances were then evaluated on a validation dataset using the ratio of performance to deviation (RPD) or to inter-quartile (RPIQ), root mean square error of prediction (RMSEP), and the determination coefficient (R²). Accurate predictions were found for clay, silt, and sand content (R² = 0.96, 0.88 and 0.93, respectively), CEC (R² = 0.93), exchangeable CaO, MgO and K2O (R² = 0.89, 0.86 and 0.8, respectively), total carbonates content (R² = 0.81) and soil inorganic carbon (R² = 0.92). These findings highlight the potential of CLR transformation as an effective preprocessing method for XRF data and offer new insights into predicting soil physico-chemical properties in hyperarid environments.