Machine learning modeling and interpretation of saline soil thermal conductivity

Soil thermal conductivity is a key parameter in geothermal design, buried infrastructure, frozen-ground engineering, agricultural heat-transfer assessment, and subsurface energy systems. This study developed an interpretable multi-model machine-learning framework for predicting soil thermal conductivity from 903 literature-derived observations and 15 predictors describing temperature, texture, bulk density, volumetric water content, salt concentration, salt molar mass, and ionic identity. Eleven regression algorithms were optimized by five-fold cross-validation within a common 90% development subset and compared on an untouched 10% random hold-out. Extra Trees yielded the lowest observed random-holdout error (R² = 0.9816, RMSE = 0.0625 W m⁻¹ K⁻¹, MAE = 0.0446 W m⁻¹ K⁻¹, AARE = 3.93%, and A20 = 98.90%), with CatBoost and XGBoost providing similarly high point performance. Because the database was assembled from 20 independent literature sources, source-aware validation was added. Pooled leave-one-study-out prediction decreased to R² = 0.2367 and RMSE = 0.3983 W m⁻¹ K⁻¹. A stricter retrospective external source validation was then performed by withholding all 100 observations from the two 2024 sources before any refitting or tuning and developing Extra Trees only on the remaining 803 observations from 18 earlier sources. The resulting external-source performance was R² = −3.3720, RMSE = 0.5855 W m⁻¹ K⁻¹, MAE = 0.5438 W m⁻¹ K⁻¹, AARE = 51.14%, and A20 = 12.00%, demonstrating substantial between-study domain shift. Feature ablation showed that a physical-only Extra Trees model achieved test R² = 0.9551; adding salt concentration increased R² to 0.9679, adding salt molar mass increased it to 0.9813, and the full chemistry representation reached R² = 0.9816. The results therefore support high-accuracy interpolation within represented experimental domains but do not support universal transfer to unseen studies. Source-specific or local validation is required before engineering deployment.