Ding et al. (2026) An ensemble method for integrating rainfall forecast products based on average mutual information decomposition

Identification

• Journal: Environmental Modelling & Software
• Year: 2026
• Date: 2026-01-06
• Authors: Wei Ding, Minglei Ren, Yawei Ning, Xuan Li, Jinnan Zhang, Haibin Zhou
• DOI: 10.1016/j.envsoft.2026.106864

Research Groups

• School of Hydraulic Engineering, Dalian University of Technology, Dalian, China
• China Institute of Water Resources and Hydropower Research, Beijing, China
• Research Center on Flood and Drought Disaster Reduction of the Ministry of Water Resources, Beijing, China

Short Summary

This study proposes MME-MID, a novel multi-model ensemble method that dynamically weights rainfall forecasts based on categorized rainfall intensities using Normalized Average Mutual Information Decomposition, significantly reducing miss ratios during flood season and false alarm ratios post-flood compared to conventional MME.

Objective

• To develop a novel multi-model ensemble (MME) framework that dynamically weights rainfall forecasts based on categorized rainfall intensities to optimize for specific error types (missed detections in flood season, false alarms post-flood) and improve decision reliability for reservoir operation.

Study Configuration

• Spatial Scale: Dahuofang and Huanren basins
• Temporal Scale: Flood season and post-flood season

Methodology and Data

• Models used: MME-MID (Multi-Model Ensemble based on Mutual Information Decomposition) framework, utilizing Normalized Average Mutual Information Decomposition with two decomposition approaches (Uncertainty decomposition and Information decomposition). Compared against conventional MME methods (e.g., simple ensemble mean (EMN), bias-removed ensemble mean (BREM), super-ensemble forecast (SUP), Kalman filter, machine learning).
• Data sources: Rainfall forecast products (from multiple institutions, e.g., CMA, ECMWF, NCEP, facilitated by TIGGE).

Main Results

• MME-MID reduced the miss ratio by 5.5 % during the flood season.
• MME-MID reduced the false alarm ratio by 11 % post-flood.
• These improvements were achieved without increasing the overall error.
• Significantly improved decision reliability for reservoir operation.

Contributions

• Introduces MME-MID, a novel multi-model ensemble framework that dynamically weights forecasts based on categorized rainfall intensities using Normalized Average Mutual Information Decomposition.
• Addresses the limitation of conventional MME methods that use static weights, which ignore forecast skill variations across rainfall intensities and seasonal operational priorities.
• Provides two distinct decomposition approaches (Uncertainty and Information decomposition) tailored to specific error reduction goals (false alarms post-flood, missed detections in flood season).

Funding

Not specified in the provided text.

Citation

@article{Ding2026ensemble,
  author = {Ding, Wei and Ren, Minglei and Ning, Yawei and Li, Xuan and Zhang, Jinnan and Zhou, Haibin},
  title = {An ensemble method for integrating rainfall forecast products based on average mutual information decomposition},
  journal = {Environmental Modelling & Software},
  year = {2026},
  doi = {10.1016/j.envsoft.2026.106864},
  url = {https://doi.org/10.1016/j.envsoft.2026.106864}
}