Azedou et al. (2025) Ensemble deep learning towards high-resolution soil-moisture mapping for enhanced water management in California's Central Valley

Identification

• Journal: Environmental Modelling & Software
• Year: 2025
• Authors: Ali Azedou, Aouatif Amine, Saïd Lahssini, Gordon Osterman, Mauricio Arboleda‐Zapata, Michael H. Cosh, Isaya Kisekka
• DOI: 10.1016/j.envsoft.2025.106824

Research Groups

• National School of Applied Sciences, Ibn Tofail University, Morocco.
• National School of Forestry Engineering, Salé, Morocco.
• USDA-ARS Sustainable Agricultural Water Systems, Davis, CA, USA.
• USDA-ARS, Beltsville, MD, USA.
• Department of Land, Air and Water Resources, University of California, Davis, USA.
• Department of Biological and Agricultural Engineering, University of California, Davis, USA.

Short Summary

This study develops an optimized ensemble deep learning framework to downscale NASA SMAP soil moisture data from 9 km to a 30 m spatial resolution across California’s Central Valley. The resulting high-resolution maps for surface and root-zone moisture provide critical data for precision irrigation scheduling and sustainable groundwater management.

Objective

• To develop and validate a novel ensemble deep learning architecture (combining DNN, CNN, and LSTM) to downscale SMAP surface and root-zone soil moisture observations to a 30 m resolution using multi-source ancillary environmental data.

Study Configuration

• Spatial Scale: California’s Central Valley, USA (approximately 51,800 km²).
• Temporal Scale: May 2023 to March 2024 (covering a full hydrologic year for training and seasonal analysis).

Methodology and Data

• Models used: Ensemble Deep Learning framework comprising Deep Neural Networks (DNN), 1D Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) networks, optimized via Bayesian hyperparameter tuning.
• Data sources:
  • Target: NASA SMAP Level 4 (9 km) surface and root-zone soil moisture.
  • Predictors: Sentinel-1 (SAR backscatter), Sentinel-2 (NDVI, SAVI, NDWI, LAI, EVI), CHIRPS (Precipitation), MODIS (LST, ET), USGS (DEM, slope, TWI), and POLARIS (Soil properties: organic matter, bulk density, texture).
  • Validation: In-situ measurements from Nuclear Magnetic Resonance (NMR) probes, Time-Domain Reflectometers (TDR), Neutron probes, and Cosmic-Ray Neutron Probes (CRNP).

Main Results

• Predictive Accuracy: The ensemble model achieved the highest performance with a correlation coefficient ($R$) of 0.789 for surface soil moisture (SSM) and 0.683 for root-zone soil moisture (RZSM).
• Error Metrics: The model recorded a Root Mean Square Error (RMSE) of 0.0281 cm³/cm³ for SSM and 0.0814 cm³/cm³ for RZSM, with Nash-Sutcliffe Efficiency (NSE) scores of 0.50 and 0.433, respectively.
• Variable Importance: SSM was primarily influenced by precipitation, land surface temperature (LST), and topography. RZSM was most sensitive to soil physical properties, specifically organic matter and silt content.
• Water Management Application: The 30 m resolution maps successfully captured seasonal variations across different crop types and were used to estimate plant-available water at the end of winter, identifying a potential storage of 4.6 km³ (3.75 million acre-feet) across the valley's irrigated lands.

Contributions

• Technical Innovation: Introduction of a specialized ensemble architecture that leverages the spatial feature extraction of CNNs and the temporal sequencing of LSTMs for soil moisture downscaling.
• Scale Advancement: Successfully bridged the gap between coarse satellite observations (9 km) and field-scale management requirements (30 m).
• Policy Support: Provides a data-driven tool for Groundwater Sustainability Agencies (GSAs) in California to refine Evapotranspiration of Applied Water (ETaw) estimates and implement demand-management strategies like delayed spring irrigation.

Funding

• USDA NIFA Award [2021-68012-35914].
• US-Morocco Fulbright Fellowship program.

Citation

@article{Azedou2025Ensemble,
  author = {Azedou, Ali and Amine, Aouatif and Lahssini, Saïd and Osterman, Gordon and Arboleda‐Zapata, Mauricio and Cosh, Michael H. and Kisekka, Isaya},
  title = {Ensemble deep learning towards high-resolution soil-moisture mapping for enhanced water management in California's Central Valley},
  journal = {Environmental Modelling & Software},
  year = {2025},
  doi = {10.1016/j.envsoft.2025.106824},
  url = {https://doi.org/10.1016/j.envsoft.2025.106824}
}

Generated by BiblioAssistant using gemini-3-flash-preview (Google API)