Pengxin et al. (2026) Multi-model integrated error correction for extreme precipitation: method and application

Identification

• Journal: Climate Dynamics
• Year: 2026
• Date: 2026-03-26
• Authors: Deng Pengxin, Bing Jianping, Zhang Dongdong, Li Wei
• DOI: 10.1007/s00382-026-08138-8

Research Groups

• Bureau of Hydrology, Changjiang Water Resources Commission, Wuhan, Hubei Province, China

Short Summary

This study developed a novel multi-model integrated error correction framework for CMIP6 extreme precipitation projections, significantly improving simulation accuracy in the Hanjiang River Basin (HRB). The corrected data reveal pronounced upward trends in extreme precipitation in the HRB, particularly in its southwestern and downstream areas, under future moderate to high radiative forcing scenarios.

Objective

• To develop and apply an advanced multi-model integrated error correction framework to mitigate estimation biases of CMIP6 climate models in simulating extreme precipitation.
• To enhance the predictive capability of climate models for extreme precipitation events and analyze the spatiotemporal characteristics of corrected extreme precipitation in the Hanjiang River Basin under future climate change scenarios.

Study Configuration

• Spatial Scale: Hanjiang River Basin (HRB), China. CMIP6 models (100–300 km resolution), NEX-GDDP-CMIP6 (25 km resolution). Data extracted for 96 rainfall stations.
• Temporal Scale: Historical period: 1956–2014. Future period: 2015–2100. Extreme precipitation indices calculated on an annual basis.

Methodology and Data

• Models used:
  • CMIP6 climate models: MRI-ESM2-0, EC-Earth3, EC-Earth3-Veg, MPI-ESM1-2-HR (from NEX-GDDP-CMIP6 and CMIP6 datasets).
  • Bias correction framework (QSVMC): Polynomial-based empirical relationships, Support Vector Machine (SVM) for multi-model integration, Parametric Quantile Mapping (PQM) with Pearson Type III (P-III) distribution.
  • Trend analysis: Trend-Free Pre-Whitening Mann–Kendall (TFPW-MK) test.
  • Spatial interpolation: Kriging method.
  • Extreme precipitation indices: RX1day (annual maximum 1-day precipitation), RX5day (annual maximum 5-day precipitation), R95p (annual total precipitation from days exceeding the 95th percentile), R99p (annual total precipitation from days exceeding the 99th percentile).
• Data sources:
  • Ground-Based Rainfall Gauge Data: Daily precipitation data from 96 rainfall stations in the HRB for 1956–2014, obtained from hydrological yearbooks.
  • CMIP6 Model Data: Downloaded from the Earth System Grid Federation (ESGF) portal.
  • NEX-GDDP-CMIP6 Data: Sourced from NASA’s official repository.
  • Experimental scenarios: Historical (1956–2014) and future Shared Socioeconomic Pathways (SSP126, SSP245, SSP370, SSP585).

Main Results

• Polynomial-based quantitative transformation significantly improved extreme precipitation predictions, reducing absolute relative errors (compared to gauges) from a range of 4.73–51.6% to 1.21–2.33%.
• SVM-based multi-model fusion substantially enhanced CMIP6 model accuracy for extreme precipitation simulation: Correlation Coefficient (CC) increased from −0.11–0.60 to 0.72–0.79; Nash–Sutcliffe Efficiency (NSE) improved from −8.17–−0.10 to 0.50–0.62; Relative Bias (RB) decreased from −1.33–3.22% to 0.93–2.47%.
• Parametric Quantile Mapping (PQM) further reduced RB to 0.22–0.80% without compromising CC or NSE metrics, effectively addressing total volume estimation errors.
• Under future moderate to high radiative forcing scenarios (SSP245, SSP370, SSP585), the HRB is projected to experience significant upward trends in extreme precipitation, with RX1day and R95p indices showing the most pronounced changes.
• High-incidence zones of extreme precipitation are identified in the southwestern sector of the basin and the downstream areas of HRB (e.g., Jingmen-Shayang, Tianmen, Wuhan, Hanzhong, Ziyang, Zhashui), which exhibit heightened sensitivity to climate change.
• As radiative forcing intensifies, the upward trends in extreme precipitation within these regions become increasingly pronounced, suggesting a growing risk of flood-inducing torrential rainfall. For instance, under SSP370 and SSP585, RX1day linear trend rates reach 1.3 mm/decade and 2.4 mm/decade, respectively.

Contributions

• Developed a novel "quantification + simulation + correction" error correction framework (QSVMC) that integrates multi-model advantages for extreme precipitation, significantly enhancing estimation accuracy.
• Pioneered the integration of polynomial-based empirical quantitative relationships with Support Vector Machine (SVM) algorithms for multi-model fusion in extreme precipitation simulation.
• Introduced parametric quantile mapping with Pearson Type III (P-III) distribution for refined bias correction, particularly effective in reducing bulk errors in extreme precipitation predictions.
• Enhanced the predictive capability of climate models for extreme precipitation events, addressing limitations in current CMIP6 outputs.
• Provided a scientific foundation and a "climate scenario-data correction-trend diagnosis" technical chain for basin-scale hydrological forecasting and flood risk mitigation, revealing the spatial heterogeneity of extreme precipitation under climate change.

Funding

• National Natural Science Foundation of China (Grant No. 52309004)
• National Key R&D Program of China (Item Nos. 2022YFC3002701, 2023YFC3206001)
• Key R&D Program of Hubei Province (Item Nos. 2023BCB115)

Citation

@article{Pengxin2026Multimodel,
  author = {Pengxin, Deng and Jianping, Bing and Dongdong, Zhang and Wei, Li},
  title = {Multi-model integrated error correction for extreme precipitation: method and application},
  journal = {Climate Dynamics},
  year = {2026},
  doi = {10.1007/s00382-026-08138-8},
  url = {https://doi.org/10.1007/s00382-026-08138-8}
}