Izadgoshasb et al. (2025) Comparison of a Semiempirical Algorithm and an Artificial Neural Network for Soil Moisture Retrieval Using CYGNSS Reflectometry Data

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-11-03
• Authors: Hamed Izadgoshasb, E. Santi, Flavio Cordari, Leila Guerriero, Leonardo Chiavini, Veronica Ambrogioni, Nazzareno Pierdicca
• DOI: 10.3390/rs17213636

Research Groups

European Space Agency (ESA) - HydroGNSS Mission

Short Summary

This study develops and compares a novel physically based algorithm and an Artificial Neural Network (ANN) for soil moisture estimation using CYGNSS Level 1B data over land, finding that the ANN generally outperforms the semiempirical model, though the latter shows greater stability in data-scarce conditions.

Objective

• To utilize CYGNSS Level 1B products over land for soil moisture estimation.
• To develop and compare a novel physically based algorithm and an Artificial Neural Network (ANN) for soil moisture retrieval within the HydroGNSS mission framework.
• To evaluate the impact of static and dynamic auxiliary data on the accuracy of soil moisture retrievals.

Study Configuration

• Spatial Scale: Global
• Temporal Scale: Daily

Methodology and Data

• Models used: Physically based algorithm (inverting a semiempirical forward model), Artificial Neural Network (ANN).
• Data sources: CYGNSS Level 1B products, SMAP L3 global daily products (target soil moisture, Vegetation Water Content (VWC), Vegetation Optical Depth (VOD)), International Soil Moisture Network (ISMN) in situ measurements, MODIS (Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI)), auxiliary data (topography, Above Ground Biomass (AGB), land cover, surface roughness).

Main Results

• The Artificial Neural Network (ANN) approach generally outperformed the semiempirical model with a Root Mean Square Error (RMSE) of 0.047 m³ m⁻³ and a correlation coefficient (R) of 0.91.
• In a global stratification framework intersecting land cover classes with climate regimes, the ANN consistently outperformed the semiempirical model in most strata, achieving an RMSE of approximately 0.04 m³ m⁻³ and correlations above 0.8.
• The semiempirical model demonstrated greater stability in data-scarce conditions, highlighting its complementary strengths for the HydroGNSS mission.

Contributions

• Development of a novel physically based algorithm and an Artificial Neural Network (ANN) specifically for soil moisture estimation using CYGNSS Level 1B data for the HydroGNSS mission.
• Comprehensive comparison of the performance of physically based and data-driven approaches for soil moisture retrieval over land.
• Evaluation of the impact of various static (topography, AGB, land cover, surface roughness) and dynamic (VWC, VOD, NDVI, NDWI) auxiliary data on retrieval accuracy.
• Introduction of a global stratification framework to analyze algorithm performance across diverse land cover and climate regimes.
• Identification of complementary strengths between the two algorithms, providing insights for robust soil moisture product generation for the HydroGNSS mission.

Funding

• European Space Agency’s second Scout mission (HydroGNSS)

Citation

@article{Izadgoshasb2025Comparison,
  author = {Izadgoshasb, Hamed and Santi, E. and Cordari, Flavio and Guerriero, Leila and Chiavini, Leonardo and Ambrogioni, Veronica and Pierdicca, Nazzareno},
  title = {Comparison of a Semiempirical Algorithm and an Artificial Neural Network for Soil Moisture Retrieval Using CYGNSS Reflectometry Data},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17213636},
  url = {https://doi.org/10.3390/rs17213636}
}