Balouei et al. (2025) Developing a new high-resolution soil moisture index for local agricultural drought monitoring using Sentinel-1 data and an artificial neural network

Identification

• Journal: Spatial Information Research
• Year: 2025
• Date: 2025-12-22
• Authors: Fateme Balouei, Mostafa Kabolizadeh, Hamidreza Rabiei-Dastjerdi
• DOI: 10.1007/s41324-025-00663-8

Research Groups

• RS and GIS, Faculty of Earth Sciences, Shahid Chamran University of Ahvaz, Ahvaz, Iran
• Department of Remote Sensing and GIS, Faculty of Earth Sciences, Shahid Chamran University of Ahvaz, Ahvaz, Iran
• School of History and Geography, Faculty of Humanities and Social Sciences, Dublin City University (DCU), Dublin, Ireland

Short Summary

This study develops a novel 10-meter resolution Local-Scale Soil Moisture Condition Index (LS-SMCI) for agricultural drought monitoring in Khuzestan, Iran, utilizing Sentinel-1 SAR data and an Artificial Neural Network (ANN). The LS-SMCI significantly outperforms coarser global soil moisture products and demonstrates strong correlation with the Standardized Precipitation Index (SPI), providing precise local drought assessment.

Objective

• Can the artificial neural network designed in this study effectively simulate soil moisture with high accuracy and efficiency using Sentinel-1 data?
• Can the new drought index, derived at a 10-meter resolution, serve as a more robust alternative to the traditional Soil Moisture Condition Index (SMCI) based on coarse-resolution soil moisture products?
• Can a C-band radar at the depth of plant root (average 0–90 cm) measure soil moisture as an indicator of agricultural drought?

Study Configuration

• Spatial Scale: Local-Scale Soil Moisture Condition Index (LS-SMCI) developed at 10-meter resolution; validation using a 30 m x 30 m window around ground stations; regional monitoring at 90-meter resolution; study area covers approximately 12,000 hectares of agricultural land in Khuzestan Province, Iran.
• Temporal Scale: Sentinel-1 data and LS-SMCI calculated for 2017–2023; Standardized Precipitation Index (SPI) calculated from 1997–2023; validation performed over a 7-year overlapping period (2017–2023) using 3-, 6-, and 9-month SPI scales; monthly soil moisture predictions.

Methodology and Data

• Models used:
  • Artificial Neural Network (ANN): Feed-forward neural network with a single hidden layer (48 neurons selected for final model), trained using TensorFlow/Keras with Adam optimizer.
  • Soil Moisture Condition Index (SMCI).
  • Standardized Precipitation Index (SPI).
• Data sources:
  • Satellite:
    • Sentinel-1 (C-band SAR, 5.405 GHz, Ground Range Detected (GRD) data, Interferometric Wide Swath (IW) mode, VH cross-polarization, VV co-polarization, 10-meter resolution). Preprocessing included thermal noise removal, radiometric calibration, terrain correction (SRTM 30-meter Digital Elevation Model), conversion to decibels, 5x5 speckle filtering, and multi-temporal averaging (30-day periods) using Google Earth Engine (GEE).
    • Famine Early Warning Systems Network (FEWS NET) Land Data Assimilation System (FLDAS) (0.10° spatial resolution, monthly).
    • Global Land Data Assimilation System Version 2 (GLDAS-2.1) (0.25° spatial resolution, monthly).
    • Soil Moisture Active Passive (SMAP) (approximately 9 km spatial resolution, daily aggregated).
  • Observation/In-situ:
    • Gravimetric soil moisture measurements from 80 sugarcane farms in the Amirkabir region, Khuzestan, Iran (at 0–30 cm, 30–60 cm, 60–90 cm depths, and their average).
    • Monthly precipitation data from six meteorological stations in Khuzestan (1997–2023) sourced from Iran’s Meteorological Organization.

Main Results

• The developed LS-SMCI index showed a strong correlation with the SPI index, with Pearson correlation coefficients (R) reaching 0.68 (for November, 3-month SPI) and Root Mean Square Error (RMSE) values generally below 0.1.
• LS-SMCI significantly outperformed traditional SMCI indices derived from coarser global products (FLDAS, GLDAS, SMAP), with correlation differences ranging from 3% to 75%.
• The ANN model, utilizing Sentinel-1 VV and VH bands and their ratios, effectively simulated soil moisture with R values of 0.83 during training and 0.74 during validation.
• Sensitivity analysis revealed that polarimetric ratios (e.g., VV/VH, (VH-VV)/(VH+VV)) exhibited higher sensitivity to soil moisture changes than single-band variables (VV, VH).
• The 10-meter resolution of LS-SMCI provided significantly greater spatial detail and variability for local drought monitoring compared to the 9 km, 11 km, and 27 km resolutions of SMAP, FLDAS, and GLDAS, respectively.
• The LS-SMCI index successfully captured historical drought and flood events, such as the 2019 floods in Khuzestan, demonstrating its ability to reflect complex soil moisture dynamics and the temporal lag between precipitation and soil infiltration.
• Analysis of drought severity from 2017 to 2023 indicated fluctuating patterns, with 2018 experiencing the most severe drought (48.26% of the area in the "extremely dry" category) and 2020 showing significant improvement (34.16% "extremely wet").

Contributions

• Development of a novel, high-resolution (10-meter) Local-Scale Soil Moisture Condition Index (LS-SMCI) for agricultural drought monitoring, offering a more accurate and practical alternative to traditional SMCI based on coarse-resolution products.
• Innovative application of an Artificial Neural Network (ANN) to simulate soil moisture using Sentinel-1 SAR VV and VH bands and their derived ratios, overcoming limitations of optical indices (e.g., cloud cover) for continuous monitoring.
• Comprehensive validation demonstrating the LS-SMCI's superior accuracy and spatial detail compared to existing global soil moisture products (FLDAS, GLDAS, SMAP) and its strong correlation with the ground-based Standardized Precipitation Index (SPI).
• Confirmation of C-band radar's effectiveness in measuring soil moisture at relevant agricultural root depths (average 0–90 cm) as an indicator for agricultural drought.
• Provision of a robust tool for precise local drought monitoring, aiding farmers and water managers in identifying affected areas and implementing timely resilience strategies.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Balouei2025Developing,
  author = {Balouei, Fateme and Kabolizadeh, Mostafa and Rabiei-Dastjerdi, Hamidreza},
  title = {Developing a new high-resolution soil moisture index for local agricultural drought monitoring using Sentinel-1 data and an artificial neural network},
  journal = {Spatial Information Research},
  year = {2025},
  doi = {10.1007/s41324-025-00663-8},
  url = {https://doi.org/10.1007/s41324-025-00663-8}
}