Zhao et al. (2026) A physics-guided sensor-to-model framework for real-time estimation and near-future forecasting of soil moisture

Identification

• Journal: Advances in Water Resources
• Year: 2026
• Date: 2026-01-22
• Authors: Haokai Zhao, Rohan Bhosale, Haruko Wainwright
• DOI: 10.1016/j.advwatres.2026.105221

Research Groups

• Department of Nuclear Science and Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
• Research Science Institute, Cambridge, MA 02139, USA

Short Summary

This study develops a physics-guided sensor-to-model framework, integrating an Ensemble Kalman Filter (EnKF) with a real-world environmental sensing network, for real-time soil moisture estimation and near-future forecasting. The framework significantly improves accuracy over calibrated hydrological models for estimation and outperforms data-driven benchmarks for 7-day forecasts.

Objective

• To develop and demonstrate a scalable, real-time soil moisture monitoring and forecasting framework by integrating an Ensemble Kalman Filter (EnKF)-based algorithm with a real-world environmental sensing network and physics-based hydrological simulations.

Study Configuration

• Spatial Scale: Localized, distributed across a real-world environmental sensing network.
• Temporal Scale: Real-time estimation and near-future forecasting (up to 7 days).

Methodology and Data

• Models used:
  • Ensemble Kalman Filter (EnKF)
  • FAO-56 Penman-Monteith Equation (for Reference evapotranspiration, ETo)
  • Physics-based hydrological simulations (calibrated via Monte Carlo simulations)
• Data sources:
  • Real-world environmental sensing network (for meteorological data and real-time sensor data streams)

Main Results

• For real-time estimation, the EnKF algorithm significantly improved accuracy over the calibrated hydrological model, achieving R² values of 0.98–0.99.
• For near-future forecasting (7-day predictions), the EnKF algorithm outperformed a data-driven LSTM benchmark, with R² values ranging from 0.78 to 0.82, particularly in predicting the timing of soil moisture responses to rainfall.
• The algorithm provides soil moisture estimations and forecasts across the entire vertical soil column, rather than being limited to discrete sensor depths.

Contributions

• Establishes a scalable, real-time soil moisture monitoring and forecasting framework by bridging sensor technologies and algorithmic modeling.
• Significantly improves real-time soil moisture estimation accuracy compared to calibrated hydrological models.
• Provides superior near-future soil moisture forecasting performance (up to 7 days) compared to data-driven benchmarks.
• Offers soil moisture estimations and forecasts across the entire vertical soil column, overcoming limitations of discrete sensor depths.
• Enhances early warning capabilities for drought and flood events and supports more efficient water use, contributing to climate resilience.

Funding

• The provided text does not contain information about funding projects or programs.

Citation

@article{Zhao2026physicsguided,
  author = {Zhao, Haokai and Bhosale, Rohan and Wainwright, Haruko},
  title = {A physics-guided sensor-to-model framework for real-time estimation and near-future forecasting of soil moisture},
  journal = {Advances in Water Resources},
  year = {2026},
  doi = {10.1016/j.advwatres.2026.105221},
  url = {https://doi.org/10.1016/j.advwatres.2026.105221}
}