Nguyen et al. (2025) Bias correction of precipitation from convection-permitting models at the point scale: a case study in Switzerland

Identification

• Journal: Climatic Change
• Year: 2025
• Date: 2025-12-01
• Authors: Trang Nguyen, Patricio Velasquez, Andreas Dietzel, Lauren M. Cook
• DOI: 10.1007/s10584-025-04079-z

Research Groups

• Department of Urban Water Management, Swiss Federal Institute for Aquatic Research (EAWAG), Dübendorf, Zurich, Switzerland
• Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland
• Climate and Environmental Physics, University of Bern, Bern, Switzerland
• Oeschger Centre for Climate Change Research (OCCR), University of Bern, Bern, Switzerland

Short Summary

This study evaluates five quantile mapping (QM) approaches to bias correct and downscale sub-hourly precipitation from convection-permitting model (CPM) simulations (2.2 km resolution) to the station scale in Switzerland. It finds that while QM reduces biases in annual precipitation indices, conventional QM often overcorrects, introducing dry biases in extreme quantiles and altering climate change signals, with a combination of spatial pooling and a moving window showing the most promise.

Objective

• To evaluate five quantile mapping (QM) approaches for bias correcting and downscaling sub-hourly COSMO-CLM simulations at 2.2 km resolution to the point scale using data from over 70 weather stations in Switzerland.
• To comprehensively assess biases in raw CPM output, conventional QM, and advanced QM techniques across the entire distribution of rainfall patterns, with a focus on overcoming limitations of limited CPM sample sizes.
• To analyze the impact of different QM methods on various rainfall features, including historical rainfall indices and the climate change signal, particularly for sub-hourly precipitation relevant to urban drainage applications.

Study Configuration

• Spatial Scale: Switzerland, focusing on 74 individual weather stations (point scale). COSMO-CLM simulations at 2.2 km horizontal resolution. Spatial pooling involves 3x3 grid cells.
• Temporal Scale: Sub-hourly (30-minute time step for QM application, COSMO-CLM output aggregated from 6-minute to 30-minute, observations aggregated from 10-minute to 30-minute). CPM simulation periods include an evaluation run (2000–2009), historical (1996–2005), and future (2090–2099) under RCP 8.5. Observational data for QM calibration typically spans 2000–2009, with an extended baseline approach using 1990–2020 (30 years). A 91-day moving window is used for one QM approach.

Methodology and Data

• Models used:
  • COSMO-CLM (Consortium for Small-Scale Modeling in Climate Mode): A non-hydrostatic convection-permitting regional climate model (CPM) from CORDEX, with TERRA-ML soil model.
  • Quantile Mapping (QM): Five approaches were tested: Conventional QM, Moving Window (MW), Spatial Pooling (SP), a combination of Moving Window and Spatial Pooling (MW-SP), and Extended Observational Baseline (ExOb).
  • DBSCAN (Density-Based Spatial Clustering of Applications with Noise): An unsupervised clustering algorithm used to group weather stations based on location, altitude, and rainfall characteristics.
• Data sources:
  • Observational rainfall data: Swiss automatic meteorological surface network (SwissMetNet, formerly ANETZ) at 10-minute resolution from 74 stations in Switzerland.
  • Reanalysis data: ERA-Interim (used to drive the COSMO-CLM evaluation run).
  • Global Climate Model (GCM): MPI-M-MPI-ESM-LR (used to drive historical and future COSMO-CLM runs).
  • Climate change scenario: Representative Concentration Pathway 8.5 (RCP 8.5).

Main Results

• Raw CPM output exhibits substantial wet biases, particularly for hourly extreme precipitation, with overestimations exceeding 90% (up to 30–35 mm/hour) in the highest quantiles.
• Conventional QM often overcorrects these wet biases, leading to significant dry biases in extreme quantiles (above the 99th percentile, approximately 10 mm/hour for wet regions and 7 mm/hour for dry regions).
• The combination of moving window and spatial pooling (MW-SP) provides the most effective bias correction, reducing wet biases in extreme quantiles to approximately 2 mm/hour in both wet and dry station clusters.
• QM successfully reduces biases in several annual precipitation indices (e.g., total annual rainfall, number of rain days exceeding 10 mm/day), but can introduce dry biases for hourly intensity and daily maximum rainfall.
• All QM approaches can significantly alter the climate change signal (CCS), inflating or reversing projected trends in key indices. For example, the projected increase in annual hourly mean intensity was inflated from ~15% to 45–60%, and the projected increase in daily 90th percentile rainfall was reversed to a 30% reduction.
• Challenges persist due to data scarcity for extreme events, significant precision disparities between observational (one-digit accuracy) and simulation data, and short calibration periods, which can lead to unstable transfer functions.

Contributions

• Provides a comprehensive assessment of five quantile mapping techniques for bias correcting and downscaling sub-hourly convection-permitting model precipitation data to the point scale.
• Highlights the critical need for bias correction in CPM outputs, especially for extreme precipitation, due to substantial wet biases.
• Identifies the combination of moving window and spatial pooling as the most promising QM approach for stabilizing transfer functions and effectively reducing extreme precipitation biases.
• Demonstrates that QM can significantly alter climate change signals, emphasizing the need for caution in impact studies relying on bias-corrected data.
• Underscores the challenges posed by limited CPM simulation periods, observational data precision, and the representation of extreme values for robust bias correction.

Funding

• SNSF (Swiss National Science Foundation) (BETTER project −200021_204790)
• Lib4RI – Library for the Research Institutes within the ETH Domain: Eawag, Empa, PSI & WSL (Open Access funding)

Citation

@article{Nguyen2025Bias,
  author = {Nguyen, Trang and Velasquez, Patricio and Dietzel, Andreas and Cook, Lauren M.},
  title = {Bias correction of precipitation from convection-permitting models at the point scale: a case study in Switzerland},
  journal = {Climatic Change},
  year = {2025},
  doi = {10.1007/s10584-025-04079-z},
  url = {https://doi.org/10.1007/s10584-025-04079-z}
}