Comment on “Deep Learning Identification of the Governing Equation for Water Flow in Heterogeneous Soils From Data” by Song et al.
Water Resources Research · 2026-11 · 62 (11)
DOI: 10.1029/2025wr042159 ↗Kurz gefasst
Fachartikel in der Zeitschrift Water Resources Research, online erschienen am 12.11.2026. Verfasst von Chan Goo Lee und 2 weiteren Autor:innen. Die Arbeit nennt 6 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Korea Institute of Energy Technology Evaluation and Planning.
Abstract (Originalsprache)
This comment examines the Extended‐DeepGS framework proposed by Song et al. (2025, https://doi.org/10.1029/2024wr037786 ) to identify governing equations for soil moisture dynamics in heterogeneous soils. Extended‐DeepGS combines neural‐network based reconstruction and automatic differentiation with sparse regression. We clarify that these components have different but complementary roles. The neural network reconstructs the soil moisture field, reduces noise, interpolates between observations, and estimates derivatives. Sparse regression identifies the structure of the governing equation. We also examine how sensitive the identified partial differential equation (PDE) is to the ridge‐regularization parameter . A pseudo‐inverse ridge‐regression analysis of the sandy loam, loam, and silt loam cases shows that changing can change the selected active terms and the resulting equation structure. This result also applies to the value reported in Song et al. (2022, https://doi.org/10.1029/2022wr031926 ). An Extended‐DeepGS comparison for the loam case shows a similar pattern. Under the noise‐free conditions examined here, the pseudo‐inverse method recovered a low‐order PDE structure similar to that obtained with Extended‐DeepGS while requiring much less computation. These findings show the importance of separating the roles of deep learning and sparse regression. They also highlight the need to clearly report how is chosen and to test the stability of the identified equation structure.
Förderung
- Korea Institute of Energy Technology Evaluation and Planning
Zitieren
APA
Lee, C. G., Han, S. Y., & Yoon, G. H. (2026). Comment on “Deep Learning Identification of the Governing Equation for Water Flow in Heterogeneous Soils From Data” by Song et al. Water Resources Research, 62(11). https://doi.org/10.1029/2025wr042159
Vancouver
Lee CG, Han SY, Yoon GH. Comment on “Deep Learning Identification of the Governing Equation for Water Flow in Heterogeneous Soils From Data” by Song et al. Water Resources Research. 2026;62(11). doi:10.1029/2025wr042159
BibTeX
@article{Lee2026,
author = {Lee, Chan Goo and Han, Seog Young and Yoon, Gil Ho},
title = {Comment on “Deep Learning Identification of the Governing Equation for Water Flow in Heterogeneous Soils From Data” by Song et al},
journal = {Water Resources Research},
year = {2026},
volume = {62},
number = {11},
doi = {10.1029/2025wr042159}
}
EUVIMED