Dong et al. (2026) A fully automated OPTRAM (aOPTRAM) for soil moisture retrieval: Evaluating multiple fitting functions, vegetation indices, land-cover types, and scales

Identification

• Journal: Remote Sensing of Environment
• Year: 2026
• Date: 2026-03-23
• Authors: Zhe Dong, Micha Silver, Gregory S. Okin, Arnon Karnieli
• DOI: 10.1016/j.rse.2026.115380

Research Groups

• The Albert Katz International School for Desert Studies, The Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Sde Boker Campus, Israel
• The Jacob Blaustein Center for Scientific Cooperation, The Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Sde Boker Campus, Israel
• Institute of Environment and Sustainability, University of California, Los Angeles, USA
• The Remote Sensing Laboratory, The Wyler Department of Dryland Agriculture, French Associate Institute for Agriculture and Biotechnology of Drylands, The Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Sde Boker Campus, Israel

Short Summary

This study introduces a fully automated Optical Trapezoid Model (aOPTRAM) for high-resolution soil moisture retrieval, systematically evaluating its performance across diverse ecosystems using Sentinel-2 imagery and in-situ data. It demonstrates that aOPTRAM, without manual calibration, achieves performance comparable to optimal OPTRAM, providing a fast and robust framework for monitoring soil moisture in heterogeneous landscapes.

Objective

• To develop and systematically evaluate a fully automated Optical Trapezoid Model (aOPTRAM) for high-resolution soil moisture retrieval.
• To assess aOPTRAM performance by testing different vegetation indices (NDVI, SAVI), edge-fitting functions (linear, exponential, polynomial), land-cover types (bare soil, shrubland, grassland, tree cover), and spatial scales (network vs. station).

Study Configuration

• Spatial Scale: Sentinel-2 imagery at 10 m spatial resolution; in-situ soil moisture measurements at 5 cm depth; network-scale boundaries encompassing entire soil moisture networks; station-scale boundaries defined by 1 km radius buffers around stations, filtered for homogeneous 10 m land-cover classes.
• Temporal Scale: Sentinel-2A/B imagery with a 5-day revisit frequency; observation periods varying from 1 to 6 years per network (e.g., 2019.01–2025.01); in-situ data resampled to daily values; monthly median composites of clear-sky observations; analysis restricted to local growing seasons.

Methodology and Data

• Models used: Optical Trapezoid Model (OPTRAM); Automated OPTRAM (aOPTRAM) incorporating an Adaptive Sliding Window (ASW) edge-detection algorithm; Optimal OPTRAM (calibration-based benchmark). Edge-fitting functions included linear, exponential, and second-degree polynomial forms. Vegetation indices used were Normalized Difference Vegetation Index (NDVI) and Soil Adjusted Vegetation Index (SAVI).
• Data sources: Sentinel-2A/B Level-2A surface reflectance products from Google Earth Engine; in-situ soil moisture measurements from 10 networks (130 stations) including the International Soil Moisture Network (ISMN) and a local Mashash network; ESA WorldCover 2021 land-cover data (10 m resolution); MOD15A2H Leaf Area Index (LAI) dataset (500 m resolution); Köppen-Geiger climate classification.

Main Results

• aOPTRAM performance is sensitive to the choice of vegetation index, land cover type, edge functional form, and spatial scale of analysis.
• SAVI with linear edges performs best in shrublands (e.g., JR: KGE = 0.726, ubRMSE = 0.018 m^3/m^3).
• NDVI with exponential edges is generally preferred in grasslands and tree cover (e.g., SMN-SDR: KGE = 0.651, ubRMSE = 0.041 m^3/m^3; FR-Aqui: KGE = 0.414, ubRMSE = 0.071 m^3/m^3).
• Bare-soil conditions remain challenging due to a small Shortwave Infrared Transformed Reflectance (STR) dynamic range, leading to weak wet–dry separability and amplified retrieval uncertainty (MinQin: KGE ≈ 0.161, Mashash: KGE ≈ 0.123).
• Station-scale edge estimation consistently improves accuracy for vegetated areas compared to network-scale estimation (e.g., SMN-SDR KGE increases from 0.651 to 0.724) and better captures seasonal soil moisture dynamics, though bare-soil performance degrades.
• aOPTRAM achieves performance broadly comparable to the optimal OPTRAM benchmark (calibration-based) despite being fully automated and independent of in-situ data for edge identification (e.g., ΔKGE up to 0.138, ΔubRMSE up to -0.026 m^3/m^3).
• A strong negative relationship was observed between aOPTRAM skill (KGE) and mean growing-season Leaf Area Index (LAI), indicating reduced performance in denser vegetation.

Contributions

• Developed a fully automated OPTRAM (aOPTRAM) framework that integrates an enhanced Adaptive Sliding Window (ASW) algorithm for robust wet/dry edge detection and fitting, eliminating the need for manual intervention or in-situ calibration.
• Conducted a systematic and comprehensive evaluation of OPTRAM performance across diverse ecosystems, vegetation indices, edge-fitting functions, and spatial scales, providing clear guidance on optimal configurations.
• Demonstrated that aOPTRAM achieves high accuracy comparable to optimally calibrated OPTRAM, establishing it as a fast, reproducible, and transferable framework for high-resolution soil moisture monitoring.
• Provided novel insights into the scale-dependency of OPTRAM performance and the inherent physical limitations of optical-only models in bare soil and dense vegetation.

Funding

• Israeli Council for Higher Education for the Tuning for Deserts project.
• Sol Leshin Program for collaboration between Ben-Gurion University of the Negev (BGU) and University of California, Los Angeles (UCLA).

Citation

@article{Dong2026fully,
  author = {Dong, Zhe and Silver, Micha and Okin, Gregory S. and Karnieli, Arnon},
  title = {A fully automated OPTRAM (aOPTRAM) for soil moisture retrieval: Evaluating multiple fitting functions, vegetation indices, land-cover types, and scales},
  journal = {Remote Sensing of Environment},
  year = {2026},
  doi = {10.1016/j.rse.2026.115380},
  url = {https://doi.org/10.1016/j.rse.2026.115380}
}