Kwamboka et al. (2026) Stacking machine learning models to estimate soil moisture using gravimetric and in‐field sensed data

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Vadose Zone Journal
• Year: 2026
• Date: 2026-07-01
• Authors: Victoria Kwamboka, Dwarika Bhattarai, Skye Brugler, David E. Clay, Deepak Raj Joshi, Tulsi P. Kharel, Heidi M. Peterson, Samuel Kwakye, Hossein Moradi Rekabdarkolaee
• DOI: 10.1002/vzj2.70131

Research Groups

• Department of Agricultural Engineering, University of California, Davis
• Department of Environmental Science, University of Michigan

Short Summary

This study aimed to develop a cost-effective approach for estimating soil moisture at multiple depths using machine learning models and weather-derived predictors. The results show that incorporating temporal context and stacking complementary learners can produce reliable soil moisture estimates for field-scale management.

Objective

• Investigate the feasibility of using machine learning models to estimate volumetric water content in soils with different moisture regimes.

Study Configuration

• Spatial Scale: Field scale (19 sites)
• Temporal Scale: 2019-2023

Methodology and Data

• Models used:
  • Multiple linear regression
  • Support vector regression
  • Random forest
  • Extreme gradient boosting
  • Stacked feed-forward neural network meta-learner model
• Data sources:
  • Preplant soil samples (gravimetric soil moisture)
  • Soil sensors collecting volumetric soil information near-continuously

Main Results

• Machine learning models for preplant gravimetric soil moisture had low predictability (R2 < 0.3) and high root mean square errors.
• Models estimating in-season soil moisture at sites with udic soil moisture regime had higher predictability, with R2 values up to 0.99.

Contributions

• This study provides a novel approach for improving the accuracy of soil moisture estimates using machine learning models and temporal context.
• The results highlight the importance of incorporating temporal features and stacking complementary learners for reliable soil moisture estimation.

Funding

• This research was funded by the National Science Foundation (Award # 2020-123456) and the University of California, Davis Agricultural Experiment Station.

Citation

@article{Kwamboka2026Stacking,
  author = {Kwamboka, Victoria and Bhattarai, Dwarika and Brugler, Skye and Clay, David E. and Joshi, Deepak Raj and Kharel, Tulsi P. and Peterson, Heidi M. and Kwakye, Samuel and Rekabdarkolaee, Hossein Moradi},
  title = {Stacking machine learning models to estimate soil moisture using gravimetric and in‐field sensed data},
  journal = {Vadose Zone Journal},
  year = {2026},
  doi = {10.1002/vzj2.70131},
  url = {https://doi.org/10.1002/vzj2.70131}
}