Sahaar et al. (2024) Estimating Rootzone Soil Moisture by Fusing Multiple Remote Sensing Products with Machine Learning

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

Identification

• Journal: Remote Sensing
• Year: 2024
• Authors: Shukran A. Sahaar, Jeffrey D. Niemann
• DOI: 10.3390/rs16193699

Short Summary

A machine learning framework integrating remote sensing and soil data was developed to estimate soil moisture across the coterminous US at five depths, finding that the XGBoost model provided the highest accuracy (R up to 0.86) and significantly outperformed the standard SMAP Level 4 product.

Objective

• Develop and evaluate a machine learning framework for estimating soil moisture at multiple depths (0–5 cm, 0–10 cm, 0–20 cm, 0–50 cm, and 0–100 cm) across the coterminous United States.
• Determine the optimal machine learning algorithm and key environmental drivers for accurate soil moisture estimation.

Study Configuration

• Spatial Scale: Coterminous United States (CONUS).
• Temporal Scale: Time series analysis (specific duration not specified).

Methodology and Data

• Models used: Feed-forward artificial neural network, Random Forest, Extreme Gradient Boosting (XGBoost), Categorical Boosting, and Light Gradient Boosting Machine.
• Data sources:
  • Soil Moisture Active Passive (SMAP) (surface soil moisture).
  • Global Precipitation Measurement (GPM) (precipitation).
  • Ecosystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) (evapotranspiration).
  • Moderate Resolution Imaging Spectroradiometer (MODIS) (vegetation data).
  • Gridded National Soil Survey Geographic (gNATSGO) (soil properties).
  • National Land Cover Database (NLCD) (land cover).
  • In situ soil moisture observations (validation data).

Main Results

• XGBoost exhibited the best overall performance among the five tested machine learning algorithms for estimating soil moisture across all depths.
• Performance generally improved with depth. For the deepest layer (0–100 cm), XGBoost achieved a correlation coefficient (R) of 0.86, a Root Mean Squared Error (RMSE) of 0.024 cm³/cm³, a Nash–Sutcliffe Efficiency (NSE) of 0.694, and a Kling–Gupta Efficiency (KGE) of 0.696.
• For the shallowest layer (0–5 cm), XGBoost achieved R = 0.76, RMSE = 0.039 cm³/cm³, NSE = 0.551, and KGE = 0.511.
• XGBoost consistently outperformed the operational SMAP Level 4 product in representing the time series of soil moisture across validation networks.
• Key factors influencing the soil moisture estimation were identified as elevation, clay content, aridity index, and antecedent soil moisture derived from SMAP.

Contributions

• Development of a comprehensive machine learning framework integrating diverse remote sensing and soil property datasets for estimating soil moisture profiles across CONUS.
• Identification of XGBoost as the optimal boosting algorithm for deep soil moisture estimation (up to 100 cm) in this multi-source data environment.
• Demonstration that the developed machine learning approach provides higher accuracy and better time series representation than the existing operational SMAP Level 4 product.

Funding

• Not specified in the paper text.

Citation

@article{Sahaar2024Estimating,
  author = {Sahaar, Shukran A. and Niemann, Jeffrey D.},
  title = {Estimating Rootzone Soil Moisture by Fusing Multiple Remote Sensing Products with Machine Learning},
  journal = {Remote Sensing},
  year = {2024},
  doi = {10.3390/rs16193699},
  url = {https://doi.org/10.3390/rs16193699}
}

Generated by BiblioAssistant using gemini-flash-latest (Google API)