Han et al. (2025) Fusion of Multi-Source Evapotranspiration Products Via the Bayesian Three-Cornered Hat Method and its Application in Runoff Simulation for Semi-Arid Basins

Identification

• Journal: Water Resources Management
• Year: 2025
• Date: 2025-12-23
• Authors: Chuqiao Han, Jianghua Zheng, Wanqiang Han, Liang Liu, Congren Li, Wenjie Yu, Juan Yang, Jiale Wu
• DOI: 10.1007/s11269-025-04348-7

Research Groups

• College of Geography and Remote Sensing Sciences, Xinjiang University, China
• Xinjiang Key Laboratory of Oasis Ecology, Xinjiang University, China

Short Summary

This study evaluates multi-source evapotranspiration (ET) products, quantifies their uncertainty using the Three-Cornered Hat (TCH) method, and fuses them via the Bayesian TCH (BTCH) method to create a more accurate ET dataset. It then demonstrates that using this fused ET significantly enhances the calibration and runoff simulation performance of the VIC-glacier hydrological model in semi-arid, data-scarce basins.

Objective

• To evaluate the accuracy of six ET products at multiple scales.
• To quantify the uncertainty in each ET product at the basin scale and within individual grid cells.
• To integrate multiple ET datasets to develop a more accurate basin-scale ET product.
• To explore the potential of integrating multisource ET data to improve the calibration of the VIC-glacier model.

Study Configuration

• Spatial Scale: Ili River Basin, Xinjiang, China (approximately 5.53 × 10^4 km^2, 42°15′−44°50′N, 80°10′−84°56′E). ET products were resampled to a 0.1° resolution, and the VIC-glacier model was executed at a 0.05° spatial resolution.
• Temporal Scale: The study period for ET product evaluation and fusion was 2003–2019. The VIC-glacier model was calibrated for 2003–2011 and validated for 2012–2019. Monthly ET datasets were developed.

Methodology and Data

• Models used:
  • Uncertainty quantification: Three-Cornered Hat (TCH) method.
  • Data fusion: Bayesian Three-Cornered Hat (BTCH) method.
  • Hydrological model: VIC-glacier model (VIC model coupled with a Degree-Day Factor (DDF) algorithm for glacier melt).
  • Model calibration: Genetic Algorithm (GA) using Nash–Sutcliffe efficiency (NSE) as the objective function.
• Data sources:
  • Evapotranspiration (ET) products: ERA5, ETMonitor, FLDAS, GLEAM, SSEBop, MOD16.
  • ET validation data: Station-observed ET (pan evaporation), water balance-based ET (ET_WB) derived from precipitation, runoff, and GRACE satellite data for terrestrial water storage change.
  • Hydrological model inputs: Meteorological forcing data (interpolated daily observations from multiple stations), topography, soil, vegetation information, and glacier data.
  • Runoff validation: Observed streamflow data from the Jilintai gauging station.

Main Results

• At the point scale, ETMonitor showed the best performance (average Pearson correlation coefficient (r) = 0.85, Root Mean Square Error (RMSE) = 36.55 mm/month), while FLDAS performed the worst.
• At the basin scale, GLEAM ET was the most accurate compared to water balance-based ET (r = 0.76, bias = 2.56%, Mean Absolute Error (MAE) = 36.01 mm/year, Root Mean Square Deviation (RMSD) = 44.56 mm/year), outperforming other products across all metrics.
• Uncertainty in individual ET products ranged from 8.24 to 19.43 mm/month, with GLEAM having the lowest and SSEBop the highest uncertainty. Uncertainty was notably higher during summer (April–August) due to increased ET variability.
• The Bayesian Three-Cornered Hat (BTCH) method significantly reduced ET uncertainty. BTCH3 (fusion of ETMonitor, FLDAS, and GLEAM) achieved the highest accuracy with the smallest uncertainty (5.83 mm/month), outperforming individual products at both point (average r = 0.84, RMSE = 31.99 mm/month) and basin scales.
• Integrating fused ET products (BTCH3) for VIC-glacier model calibration (Strategy III) significantly improved streamflow simulation reliability compared to using single ET products. The validation period showed an NSE of 0.78 and R² of 0.86, which was comparable to calibration using observed runoff (Strategy IV: NSE = 0.83, R² = 0.86). Strategy III also more accurately simulated peak flow timings and magnitudes.

Contributions

• Developed a multiscale evaluation and fusion framework for multisource ET products, systematically assessing accuracy and quantifying uncertainties, and generating high-precision, continuous spatial ET estimates using the BTCH method.
• Proposed a novel hydrological model parameter calibration strategy based on fused multisource ET data, offering a practical approach for calibrating hydrological models in data-scarce semi-arid basins.
• Provided a high-accuracy ET dataset and a transferable calibration framework for ecohydrological modeling and water resource management in ungauged or poorly monitored regions.

Funding

• The Third Xinjiang Scientific Expedition [2021xjkk1001]
• The Excellent Doctoral Innovation Project of Xinjiang University [XJU2024BS076]

Citation

@article{Han2025Fusion,
  author = {Han, Chuqiao and Zheng, Jianghua and Han, Wanqiang and Liu, Liang and Li, Congren and Yu, Wenjie and Yang, Juan and Wu, Jiale},
  title = {Fusion of Multi-Source Evapotranspiration Products Via the Bayesian Three-Cornered Hat Method and its Application in Runoff Simulation for Semi-Arid Basins},
  journal = {Water Resources Management},
  year = {2025},
  doi = {10.1007/s11269-025-04348-7},
  url = {https://doi.org/10.1007/s11269-025-04348-7}
}