Modeling soil solution electrical conductivity across Europe.

Area of Science:

• Environmental Science
• Soil Science
• Data Science

Background:

• Soil salinization negatively impacts soil health, biodiversity, and crop production.
• Accurate soil electrical conductivity (EC) assessment is vital for managing salinity.
• Predictive tools for soil EC at large scales are essential for sustainable agriculture.

Purpose of the Study:

• To develop and apply machine learning models for predicting soil EC in the EU and UK.
• To identify environmental factors influencing soil EC using feature selection.
• To create a spatial map of soil EC for the study area.

Main Methods:

• Utilized machine learning (XGBoost) with environmental predictors (soil, climate, topography, satellite data).
• Trained models on extensive soil EC data from LUCAS and WoSIS datasets.
• Employed forward feature selection to identify 17 optimal covariates from 34 initial predictors.

Main Results:

• The XGBoost model achieved R² values of 0.68 (training), 0.6 (internal testing), and 0.63 (validation).
• Estimated ~21.7 Mha of EU+UK land exceeded a predicted EC of 0.6 dS/m in 2018.
• Generated a 1km resolution gridded dataset of EC distribution and uncertainty for 2018.

Conclusions:

• Machine learning effectively predicts soil EC across the EU and UK.
• The developed model provides valuable spatial insights into soil salinity.
• Findings support improved soil management strategies and ecosystem health assessments.