Zhao et al. (2026) A novel hybrid approach for enhancing precipitation data fusion: Bayesian and geographical regression integration for hydrological applications

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-03-09
• Authors: Zijian Zhao, Ke Zhang, Xuejun Yi, Xu Yang, Linxin Liu, Xi Li, Qinuo Zhang, Yuning Luo, Haijun Wang, Zheng Xiang, Wei Gao, Cuiying Chen
• DOI: 10.1016/j.ejrh.2026.103319

Research Groups

• State Key Laboratory of Water Disaster Prevention, Hohai University, Nanjing, Jiangsu 210024, China
• College of Hydrology and Water Resources, Hohai University, Nanjing, Jiangsu 210024, China
• Yangtze Institute for Conservation and Development, Nanjing, Jiangsu 210024, China
• China Meteorological Administration Hydro-Meteorology Key Laboratory, Hohai University, Nanjing, Jiangsu 210024, China
• Key Laboratory of Water Big Data Technology of Ministry of Water Resources, Hohai University, Nanjing, Jiangsu 210024, China
• Shandong Provincial Hydrology Center, Jinan, Shandong 250014, China

Short Summary

This study proposes and validates a novel three-stage hybrid precipitation fusion framework, integrating Mixed Geographically Weighted Regression (MGWR) and Bayesian Three-Cornered Hat (BTCH) methods, to generate high-quality, high-resolution precipitation data. The "Correct-then-Combine" (MGWR-BTCH) pathway significantly improved precipitation accuracy and hydrological utility in the data-sparse Shahe Basin.

Objective

• To design and validate an optimal "Correct-then-Combine" hybrid framework for multi-source precipitation data fusion.
• To test the hypothesis that the MGWR-BTCH pathway, which first corrects spatial biases using MGWR and then combines products using BTCH, produces the highest-quality precipitation product for hydrological applications.

Study Configuration

• Spatial Scale: Shahe River Basin, Hebei Province, China (2210 km²). Precipitation data were standardized to 1 km spatial resolution. Terrain elevation data (DEM) at 90 m resolution was used to derive 1 km covariates.
• Temporal Scale: March 1, 2018, to February 28, 2022 (4 years), with analysis conducted at both daily and hourly scales.

Methodology and Data

• Models used:
  • Mixed Geographically Weighted Regression (MGWR) for spatial bias correction and downscaling.
  • Bayesian Three-Cornered Hat (BTCH) for multi-product merging and random error minimization.
  • Ensemble Averaging (EA) as a baseline fusion method.
  • Gridded Xin'anjiang-Double Anthropogenic Regulation model (GXAJ-DAR) for hydrological utility assessment.
  • Standard Geographically Weighted Regression (GWR) for methodological benchmarking.
• Data sources:
  • Observation: Daily and hourly precipitation records from 8 rain stations within the Shahe River Basin (Haihe Water Resources Commission, Hydrologic Yearbook of the Haihe River Basin).
  • Satellite Precipitation Products (SPPs):
    • CMORPH (8 km / 30 min)
    • IMERG-F (0.1° / 30 min)
    • PERSIANN (0.25° / 1 h)
  • Auxiliary Data:
    • Terrain elevation data (DEM) at 90 m spatial resolution (Geospatial Data Cloud).
    • Derived covariates: watershed boundary, surface roughness, slope, aspect, distance to coastline (1 km resolution).
    • Meteorological data (China Regional Ground Meteorological Element Driven Dataset): daily and hourly, 1 km spatial resolution.
    • Hourly wind speed data (China Ground Meteorological Observation Data, China Meteorological Data Network): 1 km spatial resolution.

Main Results

• The "Correct-then-Combine" pathway, specifically the MGWR-BTCH product, was identified as the optimal fusion framework, significantly outperforming other tested pathways.
• Precipitation Accuracy:
  • MGWR-BTCH reduced RMSE by 54% compared to a single-stage BTCH product.
  • Achieved daily precipitation metrics: BIAS = 0.04 mm/day, R = 0.83, RMSE = 3.25 mm/day.
  • Achieved hourly precipitation metrics during flood events: BIAS = 0.01 mm/h, RMSE = 0.69 mm/h, R = 0.86.
• Hydrological Utility:
  • When driving the GXAJ-DAR model, the MGWR-BTCH product yielded excellent hourly streamflow simulation with a Nash-Sutcliffe Efficiency (NSE) of 0.91.
  • The relative error of peak discharge (DP) was drastically reduced from -29.7% (for BTCH-driven simulation) to -3.87% (for MGWR-BTCH-driven simulation).
• Benchmarking and Robustness:
  • Benchmarking against standard GWR-BTCH showed MGWR-BTCH reduced RMSE by approximately 12% (from 3.69 mm/day to 3.25 mm/day) and bias from 0.15 mm/day to 0.04 mm/day, with R increasing from 0.75 to 0.83.
  • Leave-One-Out Cross-Validation (LOOCV) confirmed the framework's robustness and high predictive accuracy at ungauged locations (R = 0.75, RMSE = 3.53 mm/day, BIAS = 0.07 mm/day).

Contributions

• Proposed and systematically validated a novel three-stage hybrid precipitation fusion framework, emphasizing the optimal sequence of error processing ("Correct-then-Combine").
• Demonstrated the synergistic integration of Mixed Geographically Weighted Regression (MGWR) for multi-scale spatial bias correction and Bayesian Three-Cornered Hat (BTCH) for random error minimization.
• Provided robust validation through a dual-level assessment (direct statistical metrics and hydrological simulations with the GXAJ-DAR model) and rigorous benchmarking (against standard GWR and Leave-One-Out Cross-Validation).
• Developed a high-precision, high-resolution (1 km) precipitation product for data-scarce mountainous regions, offering a practical solution for water resource management.
• Quantified the specific benefit of accounting for multi-scale spatial non-stationarity using MGWR over standard GWR in complex terrain.

Funding

• National Key Research and Development Program of China (2023YFC3006505)
• Provincial Key Research and Development Program of Guangxi (JF2503980041)
• Special Fund Project of Jiangsu Province Science and Technology Program (BZ2024035)
• Shandong Provincial Hydrology Center Project (37000000025001720240235)

Citation

@article{Zhao2026novel,
  author = {Zhao, Zijian and Zhang, Ke and Yi, Xuejun and Yang, Xu and Liu, Linxin and Li, Xi and Zhang, Qinuo and Luo, Yuning and Wang, Haijun and Xiang, Zheng and Gao, Wei and Chen, Cuiying},
  title = {A novel hybrid approach for enhancing precipitation data fusion: Bayesian and geographical regression integration for hydrological applications},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103319},
  url = {https://doi.org/10.1016/j.ejrh.2026.103319}
}