Evaluation of different pedotransfer functions for the prediction of soil salinity using the electrical conductivity of the 1:5 soil:water extract (EC1:5)

Soil salinity is conventionally mirrored by the soil electrical conductivity (EC) of the saturated-paste extract (ECe), which consists of a labor-intensive and time-consuming procedure. In contrast, the EC of soil:water extracts at various ratios, most commonly EC1:5, has emerged as a simpler and time-efficient alternative for predicting ECe using pedotransfer functions (PTFs). Considering the various modeling approaches for the conversion of EC1:5 to ECe, the purpose of this study was to investigate the performance of two popular PTFs, including a Multiple Linear Regression (MLR) and a Random Forest (RF), in predicting topsoil (0–30 cm) salinity. To this aim, 173 soil samples were collected from three regions (I,II,III) of Greece and the ECe, EC1:5, soil texture, CaCO 3 , and soil organic matter (SOM) were determined. The examined PTFs were evaluated across a range of soil textures, salinity levels, and SOM up to 9.2%. According to the findings, RF outperformed MLR when all soils combined (R 2 = 0.983, RMSE = 0.439 dSm −1 , MAE = 0.261 dSm −1 ), whereas in the two regions (I, II) characterized by low SOM (≤ 2.5%), both MLR and RF models demonstrated similarly satisfactory predictive performance. In the third region (III), containing soils with SOM >2.5%, MLR achieved higher predictive accuracy than RF, with R 2 values from 0.592 to 0.935 and RMSE = 0.293 dSm −1 and MAE = 0.232 dSm −1 , while RF exhibited reduced generalization capability. Ultimately, with the presented comparative analysis of PTFs, the current research intends to assist land managers and policymakers in reliably predicting ECe by EC1:5 in similar soil conditions and support a more comprehensive and accurate soil salinity monitoring.