Machine learning facilitates connections between soil thermal conductivity, soil water content, and soil matric potential
Xiangwei Wang,Yanchen Gao,Jiagui Hou,Jiahui Yang,Kathleen Smits,Hailong He
DOI: https://doi.org/10.1016/j.jhydrol.2024.130950
IF: 6.4
2024-04-01
Journal of Hydrology
Abstract:Properly understanding the strongly coupled thermal and hydraulic properties of porous media is crucial to interests in different fields, e.g., agriculture, building physics and meteorology. Previous studies generally consider the thermal and hydraulic properties independently, specifically, the soil thermal conductivity curve and the soil water retention characteristics curve as methods to connect such process are unavailable or untested. In an effort to test methods based on machine learning (ML) to effectively integrate the soil thermal conductivity (λ), matric potential (ψ) and water content (θ), 2328 measurements from 43 soils were compiled using ML algorithms. Five ML algorithms, including random forest-RF, gradient boosting decision tree-GBDT, support vector machine-SVM, multiple linear regression-LR and backpropagation neural network-BPNN, based on the balanced dataset with synthetic minority over-sampling technique were trained, validated and tested to establish the λ(ψ) and λ(θ) relationships. Results demonstrate that, ML algorithms perform better than empirically based methods to optimize the thermal and hydraulic relationship. The selection of ML algorithms has a significant influence on the coupling relationship and should be taken into account. GBDT performed superior for small soil hydrothermal datasets (Testing set: Nash-Sutcliff efficiency-NSE = 0.9999 and Root Mean Square Error-RMSE = 0.0144 W m−1 K−1; independent verification: NSE = 0.7122 and RMSE = 0.2726 W m−1 K−1), while BPNN is limited by the amount of data with NSE = 0.9970, RMSE = 0.0790 W m−1 K−1 of Testing set.
geosciences, multidisciplinary,water resources,engineering, civil