Rabiei et al. (2025) Deep Learning-Based Short- and Mid-Term Surface and Subsurface Soil Moisture Projections from Remote Sensing and Digital Soil Maps

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-09-18
• Authors: Saman Rabiei, Ebrahim Babaeian, Sabine Grunwald
• DOI: 10.3390/rs17183219

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This study develops a convolutional long short-term memory (ConvLSTM) framework to generate short- and mid-term forecasts of surface and subsurface soil moisture across the contiguous U.S., demonstrating its skill in supporting large-scale drought and flood monitoring despite varying accuracy with lead time, soil texture, and land cover.

Objective

• To develop and evaluate a convolutional long short-term memory (ConvLSTM) network framework for producing short- (1, 3, and 7 days ahead) and mid-term (14 and 30 days ahead) forecasts of soil moisture at surface (0–0.1 m) and subsurface (0.1–0.4 m and 0.4–1.0 m) soil layers across the contiguous U.S.

Study Configuration

• Spatial Scale: Contiguous U.S.
• Temporal Scale: Forecasts for 1, 3, 7, 14, and 30 days ahead. Model trained on a five-year period (2018–2022), with training data from January 2018–January 2021, validation in 2021, and testing in 2022.

Methodology and Data

• Models used: Convolutional Long Short-Term Memory (ConvLSTM) network.
• Data sources:
  • Soil Moisture Active Passive (SMAP) level 3 ancillary covariables.
  • North American Land Data Assimilation System phase 2 (NLDAS-2) soil moisture product.
  • Shortwave infrared reflectance from Moderate Resolution Imaging Spectroradiometer (MODIS).
  • Terrain features (e.g., elevation, slope, curvature).
  • Soil texture and bulk density maps from the Soil Landscape of the United States (SOLUS100) database.
  • In situ observations from the Soil Climate Analysis Network (SCAN) and the United States Climate Reference Network (USCRN) soil moisture networks for validation.

Main Results

• The accuracy of soil moisture forecasts decreased with increasing lead time, particularly in the surface (0–0.1 m) and upper subsurface (0.1–0.4 m) layers due to strong fluctuations from rainfall and evapotranspiration.
• Across all soil layers and lead times, the model achieved a median unbiased root mean square error (ubRMSE) of 0.04 cm³ cm⁻³ with a Pearson correlation coefficient of 0.61.
• Forecast accuracy was highest in coarse-textured soils, followed by medium- and fine-textured soils.
• Performance was strongest in grasslands and savannas and weakest in dense forests and shrublands.

Contributions

• Presents a novel ConvLSTM framework for large-scale, multi-layer, short- and mid-term soil moisture forecasting across the contiguous U.S.
• Provides a comprehensive evaluation of forecast accuracy across different soil depths, lead times, land cover types, and soil textures.
• Demonstrates the potential of the ConvLSTM framework to provide skillful soil moisture forecasts, offering valuable support for large-scale drought and flood monitoring.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Rabiei2025Deep,
  author = {Rabiei, Saman and Babaeian, Ebrahim and Grunwald, Sabine},
  title = {Deep Learning-Based Short- and Mid-Term Surface and Subsurface Soil Moisture Projections from Remote Sensing and Digital Soil Maps},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17183219},
  url = {https://doi.org/10.3390/rs17183219}
}