Xie et al. (2025) Determination of dry soil layer and vertical variations in soil thermal properties through interpretation of heat pulse signals using deep learning-based data assimilation

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-12-10
• Authors: Xiaoting Xie, Lijie Li, Yili Lu, Lin Liu, Tusheng Ren
• DOI: 10.1016/j.jhydrol.2025.134775

Research Groups

• Department of Geographic Science, Faculty of Arts and Sciences, Beijing Normal University at Zhuhai, Zhuhai, China
• National Engineering Laboratory for Site Remediation Technologies (NEL-SRT), Beijing, China
• BCEG Environmental Remediation Co., Ltd., Beijing, China
• College of Land Science and Technology, China Agricultural University, Beijing, China

Short Summary

This study evaluates a deep learning-based data assimilation (DA(DL)) method for estimating vertical variations in soil thermal properties from heat pulse measurements to characterize dry soil layer (DSL) dynamics. It demonstrates that DA(DL) outperforms conventional methods in handling nonlinearity and successfully captures DSL evolution under natural evaporation conditions.

Objective

• To evaluate a deep learning (DL)-based data assimilation (DA) method, DA(DL), for estimating vertical variations in soil thermal properties from heat pulse (HP) measurements, particularly to characterize dry soil layer (DSL) dynamics under radiative boundary conditions.

Study Configuration

• Spatial Scale: Point-scale measurements at specific depths, resolving vertical variability across entire soil profiles.
• Temporal Scale: Dynamic evolution of soil thermal properties and dry soil layer over time, under natural soil evaporation conditions.

Methodology and Data

• Models used: Deep learning-based data assimilation (DA(DL)), Ensemble Smoother with DL-based update (ES(DL)), Ensemble Smoother with Kalman-based update (ES(K)).
• Data sources: Heat pulse (HP) measurements, numerical experiments (controlled scenarios), and field experiments under natural soil evaporation conditions.

Main Results

• Numerical experiments revealed that DA(DL) and ES(DL) consistently outperformed ES(K) in estimating soil thermal conductivity (λ) and volumetric heat capacity (C), with average RMSEs of 0.17 W m⁻¹ K⁻¹ (λ) and 0.26 MJ m⁻³ K⁻¹ (C) for DA(DL), compared to 0.28 W m⁻¹ K⁻¹ (λ) and 0.31 MJ m⁻³ K⁻¹ (C) for ES(K).
• The superior performance of DA(DL) and ES(DL) is attributed to their enhanced capacity to handle strong nonlinearity and non-Gaussian distributions.
• Field experiments validated DA(DL)'s effectiveness in estimating soil thermal properties and successfully captured the dynamic evolution of the dry soil layer (DSL), with inferred transitions closely aligning with observed evaporation front progression.

Contributions

• Proposes and validates a novel deep learning-based data assimilation framework (DA(DL)) for interpreting heat pulse signals to resolve vertical variations in soil thermal properties.
• Demonstrates the superior capability of DL-based data assimilation methods over conventional Kalman-based methods in handling complex, heterogeneous soil conditions and strong nonlinearities.
• Provides an innovative methodology for identifying and characterizing the spatiotemporal dynamics of the dry soil layer (DSL), advancing the understanding of subsurface energy and moisture transport.

Funding

• Not specified in the provided text.

Citation

@article{Xie2025Determination,
  author = {Xie, Xiaoting and Li, Lijie and Lu, Yili and Liu, Lin and Ren, Tusheng},
  title = {Determination of dry soil layer and vertical variations in soil thermal properties through interpretation of heat pulse signals using deep learning-based data assimilation},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134775},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134775}
}