Zhu et al. (2026) DeepProfile: An inverse fusion framework for root zone soil moisture profile estimation

Identification

• Journal: Remote Sensing of Environment
• Year: 2026
• Date: 2026-04-08
• Authors: Liujun Zhu, Yi Tan, Shanshui Yuan, Junliang Jin, Zhengyang Tang, Jeffrey P. Walker
• DOI: 10.1016/j.rse.2026.115408

Research Groups

• State Key Laboratory of Water Disaster Prevention, Yangtze Institute for Conservation and Development, Hohai University, Nanjing 210098, China
• Department of Civil and Environmental Engineering, Monash University, Clayton, VIC 3800, Australia
• College of Hydrology and Water Resources, Hohai University, Nanjing 210098, China
• China Yangtze Power Co., Ltd. (CYPC), Wuhan, Hubei 430000, China

Short Summary

This study introduces DeepProfile, an inverse fusion framework that integrates multiple heterogeneous root zone soil moisture (RZSM) products to estimate continuous soil moisture profiles down to 1 meter, demonstrating strong global agreement with in-situ measurements.

Objective

• To develop a scalable and transferable framework (DeepProfile) for estimating continuous soil moisture profiles throughout the top 1 meter layer of soil by integrating widely used RZSM products with inconsistent accuracy and disparate vertical discretizations.

Study Configuration

• Spatial Scale: Global, validated against 2373 in-situ stations across 45 global networks.
• Temporal Scale: Continuous monitoring period implied by input data products (SMAP L4, GLDAS v2, ERA5-land).

Methodology and Data

• Models used: DeepProfile framework, which optimizes the integral of a polynomial profile to match heterogeneous layers using location-specific triple collocation weights.
• Data sources:
  • Input RZSM products: Soil Moisture Active Passive level 4 (SMAP L4), Global Land Data Assimilation System (GLDAS) version 2, ERA5-land (fifth generation European Centre for Medium-Range Weather Forecasts atmospheric reanalysis).
  • Validation data: In-situ measurements from 2373 stations across 45 global networks.
  • SMAP near-surface soil moisture (used as a critical input for top 0.5 m).

Main Results

• Strong agreement with in-situ measurements was observed in near-surface and intermediate layers (≤0.5 m).
• Median Root Mean Square Error (RMSE) values were below 0.06 m³/m³ for layers ≤0.5 m.
• Correlation coefficients (R) exceeded 0.72 for layers ≤0.5 m.
• The model showed promising performance at deeper layers (>0.5 m) with R > 0.65.
• Accuracy declined with increasing depth, attributed to weaker observational constraints.
• The inclusion of SMAP near-surface soil moisture was critical for satisfactory results in the top 0.5 m.

Contributions

• Proposes DeepProfile, a novel inverse fusion framework for estimating continuous soil moisture profiles from 0 to 1 meter depth.
• Integrates heterogeneous RZSM products without requiring harmonized inputs or identical vertical/spatiotemporal coverage.
• Generates a continuous analytical profile, allowing for flexible depth extraction.
• Utilizes location-specific triple collocation weights to optimally balance contributions from different parent products.
• Offers a scalable and transferable solution with potential applications in hydrological modeling, drought monitoring, and weather forecasting.

Funding

[No funding information was provided in the article text.]

Citation

@article{Zhu2026DeepProfile,
  author = {Zhu, Liujun and Tan, Yi and Yuan, Shanshui and Jin, Junliang and Tang, Zhengyang and Walker, Jeffrey P.},
  title = {DeepProfile: An inverse fusion framework for root zone soil moisture profile estimation},
  journal = {Remote Sensing of Environment},
  year = {2026},
  doi = {10.1016/j.rse.2026.115408},
  url = {https://doi.org/10.1016/j.rse.2026.115408}
}