loading page

Soil Water Movement Model Integrating Physical Information with Deep Learning
  • +3
  • Tao Wang,
  • Kuan Liu,
  • Jiaqi Zhai,
  • Yifei Zhao,
  • Longlong Zhang,
  • Lizhen Wang
Tao Wang
China Institute of Water Resources and Hydropower Research
Author Profile
Kuan Liu
China Institute of Water Resources and Hydropower Research
Author Profile
Jiaqi Zhai
China Institute of Water Resources and Hydropower Research

Corresponding Author:jiaqizhai@163.com

Author Profile
Yifei Zhao
China Institute of Water Resources and Hydropower Research
Author Profile
Longlong Zhang
China Institute of Water Resources and Hydropower Research
Author Profile
Lizhen Wang
China Institute of Water Resources and Hydropower Research
Author Profile

Abstract

The nonlinearity of soil moisture content, water potential, and unsaturated hydraulic conductivity in soil layers makes it difficult to simulate wetting-drying cycles using conventional means. We addressed this issue using physics-informed neural networks (PINNs). Based on the van Genuchten model, we solved the Richards equation with soil matric potential as the primary variable within the PINNs framework. The proposed approach was applied and validated in a typical deep-buried area at the Luancheng Experimental Station. We found that the PINNs method was as accurate as the finite difference method in simulating the vertical infiltration of soil moisture in a non-laboratory environment. Moreover, the model exhibited swift performance when the soil layer parameters were fine-tuned. Overall, this model accurately characterizes hydrological elements using minimal data, thereby providing a new approach for simulating related hydrological processes.
Submitted to Hydrological Processes
07 May 2024Reviewer(s) Assigned
04 Jul 2024Review(s) Completed, Editorial Evaluation Pending
16 Jul 2024Editorial Decision: Revise Major