Zhang et al. (2026) Multi-source precipitation fusion for hydrological models: Correction and metrics importance analysis

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-02-26
• Authors: Min Zhang, Yang Cheng, Shaowei Ning, Yuliang Zhou, Chengguo Wu, Yi Cui, Juliang Jin, Bhesh Raj Thapa
• DOI: 10.1016/j.ejrh.2026.103291

Research Groups

• College of Civil Engineering, Hefei University of Technology, Hefei 230009, China
• Universal Engineering and Science College, Pokhara University, Lalitpur 44700, Nepal
• Nepal Academy of Science and Technology, Lalitpur 44700, Nepal
• State Nuclear Electric Power Planning Design & Research Institute, Beijing 100080, China

Short Summary

This study developed a lightweight Absolute Distance Inverse Weighting (ADIW) framework to merge eight precipitation datasets, evaluating the merged product's performance and bias-corrected versions through hydrological simulations using HYPE and VIC models in the Ganjiang River Basin. The ADIW+Linear Regression (LR) approach demonstrated optimal hydrological performance, with Relative Bias (RB) and Mean Absolute Error (MAE) identified as key metrics controlling hydrological reliability.

Objective

• To develop a lightweight and interpretable precipitation merging framework based on Absolute Distance Inverse Weighting (ADIW) to integrate multiple mainstream precipitation datasets.
• To rigorously evaluate the performance of the merged product, corrected with four bias-correction methods (Linear Regression, Linear Scaling, Quantile Mapping, Quantile–Quantile), through hydrological simulation using the HYPE and VIC models.
• To analyze the importance of precipitation evaluation metrics in hydrological modeling.

Study Configuration

• Spatial Scale: Ganjiang River Basin, southeastern China (approximately 83,500 km²), with precipitation products resampled to 0.25° spatial resolution.
• Temporal Scale: 2008 to 2020, daily timescale for precipitation and runoff data.

Methodology and Data

• Models used:
  • Precipitation merging: Absolute Distance Inverse Weighting (ADIW)
  • Bias correction: Linear Regression (LR), Linear Scaling (LS), Quantile Mapping (QM), Quantile–Quantile (QQ)
  • Hydrological models: Hydrological Predictions for the Environment (HYPE, semi-distributed), Variable Infiltration Capacity (VIC, distributed, version 4.2d)
  • Importance analysis: Pearson's correlation coefficient (PCC), Random Forest (RF), Shapley value analysis (XAI techniques)
• Data sources:
  • Precipitation products (merged): CHIRPS, CMORPH, ERA5, GSMaP, IMERG, PERSIANN, SM2RAIN, TRMM (all resampled to 0.25° daily)
  • Reference precipitation: China Gridded Daily Precipitation Analysis (CGDPA) from China Meteorological Administration (CMA) (0.25° daily)
  • Evaluation reference: Multi-Source Weighted-Ensemble Precipitation (MSWEP version 2.8)
  • Ground-based observations: Daily maximum/minimum temperature, daily mean wind speed from CMA.
  • Runoff data: Daily observed runoff from Waizhou Station, Jiangxi Province (2008–2020).
  • Auxiliary data: Digital Elevation Model (DEM) (90 m), global land cover classification dataset (1 km), Harmonized World Soil Database (HWSD).

Main Results

• The ADIW merging framework effectively synthesized multiple datasets, producing a merged product with superior correlation and rainfall detection skill compared to individual inputs, despite a slight underestimation.
• All four bias-correction methods (LR, LS, QM, QQ) effectively mitigated systematic errors, with Linear Regression (LR) proving the most robust and consistent in enhancing both precipitation accuracy and subsequent runoff simulations.
• The combined ADIW+LR approach delivered optimal hydrological performance, achieving the highest Nash–Sutcliffe Efficiency (NSE) and Kling–Gupta Efficiency (KGE) values in both HYPE and VIC models.
• ERA5 and GSMaP were identified as the most influential datasets contributing to the merged product, consistently receiving the highest weights, with moderate seasonal shifts.
• Diagnostic analysis using Pearson's correlation and explainable artificial intelligence (XAI) techniques established that relative bias (RB) and mean absolute error (MAE) are the key metrics controlling hydrological reliability, exhibiting the strongest negative correlations with NSE and KGE.
• A dynamic ADIW scheme with a 7-day moving window (Merged-mv7) significantly improved NSE, RB, MAE, and KGE in the HYPE model, and notably reduced MAE in the VIC model, demonstrating improved temporal stability and occurrence reliability.

Contributions

• Development of a lightweight and interpretable Absolute Distance Inverse Weighting (ADIW) precipitation merging framework.
• Comprehensive evaluation of the merged product and its bias-corrected versions using two distinct hydrological models (HYPE and VIC), providing insights into their hydrological applicability.
• Identification of Linear Regression (LR) as the most robust bias-correction method for enhancing both precipitation accuracy and runoff simulations when combined with ADIW.
• Quantification of the relative importance of various precipitation evaluation metrics (RB, MAE, POD, CC, KGE, FAR) for hydrological model performance using Pearson's correlation and explainable AI techniques.
• Introduction and evaluation of a dynamic ADIW approach using a moving window to enhance merged precipitation products, demonstrating improved temporal stability and occurrence reliability.
• Provides an efficient precipitation merging strategy and more accurate precipitation inputs for hydrological modeling applications, particularly in humid subtropical basins.

Funding

• National Key Research and Development Program of China (Grant no. 2023YFC3206604–02)
• National Natural Science Foundation of China (Grant no. 52379006)
• Anhui Provincial Natural Science Foundation (Grant no. 2208085US15)

Citation

@article{Zhang2026Multisource,
  author = {Zhang, Min and Cheng, Yang and Ning, Shaowei and Zhou, Yuliang and Wu, Chengguo and Cui, Yi and Jin, Juliang and Thapa, Bhesh Raj},
  title = {Multi-source precipitation fusion for hydrological models: Correction and metrics importance analysis},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103291},
  url = {https://doi.org/10.1016/j.ejrh.2026.103291}
}