Zhu et al. (2026) Estimating daily seamless 20-m resolution evapotranspiration using data fusion and TSEB

Identification

• Journal: Agricultural Water Management
• Year: 2026
• Date: 2026-02-06
• Authors: Peng Zhu, Qisheng Han, Caixia Li, Hao Liu, Qingyao Zhao, Yaoming Ma, Mengru Yu, Shenglin Li, Jinglei Wang
• DOI: 10.1016/j.agwat.2026.110210

Research Groups

• Key Laboratory of Crop Water Use and Regulation, Ministry of Agriculture and Rural Affairs. Institute of Farmland Irrigation, Chinese Academy of Agriculture Sciences, Xinxiang 453003, China
• School of Information Engineering, Henan Institute of Science and Technology, Xinxiang 453003, China
• College of Horticulture and Landscape Architecture, Henan Institute of Science and Technology, Xinxiang 453003, China

Short Summary

This study developed an efficient framework integrating cloud-filling, a high-performance spatiotemporal fusion model, multi-source remote sensing data, and the Two-Source Energy Balance (TSEB) model to produce daily seamless evapotranspiration (ET) estimates at 20 m resolution. The framework achieved robust performance, with daily ET estimates showing a coefficient of determination (R²) of 0.56, a mean bias (BIAS) of –0.08 mm/d, and a root mean square error (RMSE) of 1.05 mm/d, providing a powerful tool for precision agricultural water resource management.

Objective

• To generate daily seamless land surface parameters (land surface temperature, leaf area index, fractional vegetation cover) at 20 m resolution through data fusion, integrating multi-source datasets (VIIRS, Sentinel-2/3, CLDAS LST) with cloud-filling algorithms and efficient spatiotemporal fusion methods.
• To evaluate the performance of the proposed framework in estimating instantaneous surface flux components and daily ET.
• To analyze the spatiotemporal variations of ET and assess the performance of different fusion strategies in ET estimation.

Study Configuration

• Spatial Scale: Regional scale (irrigation district), with output ET at 20 m spatial resolution.
• Temporal Scale: Daily seamless ET estimates for the critical growth period of winter wheat (March 1 to June 1) from 2019 to 2023.

Methodology and Data

• Models used:
  • Two-Source Energy Balance (TSEB) model (specifically TSEB-PT model for ET estimation)
  • Modified Neighborhood Similar Pixel Interpolator (MNSPI) (cloud-filling algorithm)
  • GPU-enabled Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (cuESTARFM) (spatiotemporal fusion)
  • Data Mining Sharpening (DMS) algorithm (LST downscaling)
• Data sources:
  • Satellite:
    • Visible Infrared Imaging Radiometer Suite (VIIRS): Daily 1 km surface reflectance products (VNP09GA).
    • Sentinel-2 Multispectral Instrument (MSI): 20 m Level-2A surface reflectance products.
    • Sentinel-3 Sea and Land Surface Temperature Radiometer (SLSTR): 1 km Level-2 LST products.
    • Landsat-8/9 Level-2 LST products (for validation).
    • Unmanned aerial vehicle (UAV) Thermal Infrared (TIR) imagery (for validation).
    • ESA WorldCover 10 m 2021 V200 product (land use data).
  • Reanalysis:
    • China Land Data Assimilation System (CLDAS-V2.0): 2 m air temperature, specific humidity, 10 m wind speed, incoming shortwave radiation, and LST at 0.0625° × 0.0625° resolution.
  • Observation (In-situ):
    • Eddy covariance (EC) system: Turbulent fluxes (sensible heat flux (H), latent heat flux (LE)).
    • Four-component radiometer: Net radiation (Rn).
    • Soil heat flux plates: Soil heat flux (G).
    • Soil temperature and moisture profile observations.
    • CNR4-observed LST data (for validation).
  • Other:
    • METRIC-EEFlux ET products (for comparison).

Main Results

• The GPU-accelerated cuESTARFM achieved a computational speedup of 540.6–652.76 times compared to the traditional CPU-based ESTARFM, reducing fusion time to approximately 40 seconds per task.
• The generated daily seamless 20 m Land Surface Temperature (LST) showed high accuracy:
  • Compared with UAV LST: R values of 0.79 and 0.76, with RMSE values of 2.47 K and 2.52 K, respectively.
  • Compared with CNR4 observations: R = 0.93, BIAS = –0.28 K, RMSE = 2.75 K.
• The proposed framework, using all available data, achieved the following validation metrics against ground observations:
  • Instantaneous latent heat flux (LE): R² = 0.77, BIAS = 2.99 W/m², RMSE = 74.61 W/m².
  • Daily ET: R² = 0.56, BIAS = –0.08 mm/d, RMSE = 1.05 mm/d.
• Comparison of daily 20 m ET estimation strategies (2021 data):
  • Proposed framework (fused surface parameter approach): R² = 0.77, BIAS = 0.25 mm/d, RMSE = 0.87 mm/d.
  • Fused ET approach: R² = 0.61, BIAS = 0.14 mm/d, RMSE = 0.91 mm/d.
  • ET₀-based interpolation: R² = 0.38, BIAS = 0.51 mm/d, RMSE = 1.59 mm/d.
• The proposed framework's 20 m ET estimates (R² = 0.54, BIAS = 0.51 mm/d, RMSE = 1.09 mm/d) outperformed the EEFlux product (R² = 0.43, BIAS = 0.43 mm/d, RMSE = 1.75 mm/d).

Contributions

• Developed an innovative and efficient framework for daily seamless 20 m ET estimation by integrating a cloud-filling algorithm (MNSPI), a high-performance GPU-enabled spatiotemporal fusion model (cuESTARFM), multi-source remote sensing data (VIIRS, Sentinel-2/3), reanalysis data (CLDAS), and the TSEB model.
• Achieved significant computational efficiency improvements (540-650 times faster) for large-scale image fusion, enabling near-real-time applications.
• Successfully generated high-precision, daily seamless 20 m land surface parameters and ET products, effectively overcoming the inherent trade-off between spatial and temporal resolution and mitigating cloud contamination issues.
• Demonstrated superior accuracy and spatial detail for field-scale ET monitoring compared to existing ET fusion and interpolation methods, and open-source products like EEFlux.
• Optimized data source selection by incorporating VIIRS and CLDAS LST, enhancing the framework's long-term applicability and stability beyond reliance on aging MODIS products.

Funding

• Agricultural Science and Technology Major Project

Citation

@article{Zhu2026Estimating,
  author = {Zhu, Peng and Han, Qisheng and Li, Caixia and Liu, Hao and Zhao, Qingyao and Ma, Yaoming and Yu, Mengru and Li, Shenglin and Wang, Jinglei},
  title = {Estimating daily seamless 20-m resolution evapotranspiration using data fusion and TSEB},
  journal = {Agricultural Water Management},
  year = {2026},
  doi = {10.1016/j.agwat.2026.110210},
  url = {https://doi.org/10.1016/j.agwat.2026.110210}
}