Estimation of the electrical conductivity of saturated paste from soil–water extracts of coastal saline paddy soils using random forest and multiple regression models

Soil salinity is often measured with electrical conductivity (EC) of soil–water extracts (EC_w), and thus, conversion of EC_w to saturated paste EC (EC_e) is required for practical uses of EC_w. However, conversion factors (CF, the ratio of EC_e to EC_w) vary widely depending on soil properties such as soil texture and ionic composition. This study was conducted to develop a novel procedure to obtain soil-specific CF by performing random forest (RF) and multiple regression (MR) modeling using EC and soil properties.

Soil samples (n = 320) were collected from paddy fields in 16 reclaimed coastal areas of South Korea in 2018–2022. The samples were analyzed for EC (EC_1:5 and EC_e) and other soil properties including soil particle size, organic matter content, and exchangeable sodium percentage (ESP). The most important soil properties affecting the CF ratios were explored using RF modeling, and a MR equation was developed to estimate EC_e using EC_1:5 and the soil properties as parameters.

The RF model indicated that ESP and clay content are the key soil factors affecting CF variability, contributing 56.6% and 20.0% of the variability, respectively. Therefore, a multiple regression model was developed to predict CF using ESP and clay content, CF = 7.179 − 0.084 × Clay + 1.096 × ln(ESP). The predicted EC_e was linearly correlated with the measured EC_e (r² = 0.84) with a marginal error (3.1%).

Our study provides a novel method to estimate EC_e from EC_w by performing RF and MR modeling using soil-specific variables such as clay content and ESP.