Debrie et al. (2025) Hourly precipitation fields at 1 km resolution over Belgium from 1940 to 2016 based on the analog technique

Identification

• Journal: Earth system science data
• Year: 2025
• Date: 2025-11-24
• Authors: Elke Debrie, Jonathan Demaeyer, Stéphane Vannitsem
• DOI: 10.5194/essd-17-6405-2025

Research Groups

Meteorological and Climatological Information Service, Royal Meteorological Institute of Belgium, Brussels, Belgium

Short Summary

This study develops and validates a high-resolution (1000 m, hourly) gridded precipitation dataset for Belgium from 1940 to 2016 using an analog technique, demonstrating that a median-based ensemble of 25 analogs provides optimal performance for precipitation estimation. The resulting dataset is publicly available for various hydrological and climate applications.

Objective

• To develop a high-resolution (1000 m, hourly) gridded precipitation dataset for Belgium spanning 1940–2016 using an analog technique, validated against observed data.

Study Configuration

• Spatial Scale: Belgium, covering a rectangular region from approximately 0.3° W to 9.7° E and 47.4° N to 53.7° N, with a spatial resolution of 1000 m. The analog search region for atmospheric parameters extended from 1.5° E to 6.5° E and 49.4° N to 52.4° N (approximately 1.17 x 10^11 m^2).
• Temporal Scale: Hourly precipitation fields for the period 1940–2016. Analog days were sampled from the period 2017–2022.

Methodology and Data

• Models used:
  • Analog technique: Identifies past days with similar large-scale atmospheric conditions to a target day.
  • Teweles–Wobus score (TWS): Used to quantify similarity between geopotential fields.
  • Root Mean Square Error (RMSE): Used to quantify differences in precipitation fields for validation and for assessing similarity of relative humidity and temperature.
  • Receiver Operating Characteristic (ROC) curve and Area Under the Curve (AUC): Used for evaluating the discrimination ability of precipitation forecasts.
• Data sources:
  • ERA5 reanalysis (ECMWF): Atmospheric parameters (geopotential, temperature, relative humidity) at 0.25° spatial resolution, from 1940 to present. Used for identifying analog days.
  • RADCLIM product (RMI): High-resolution (1000 m) radar-based quantitative precipitation estimation for Belgium and surroundings, from 2017 to 2022. This product is derived from a network of four C-band dual-polarization weather radars and merged with rain gauge measurements. Used as the high-resolution proxy for analog days.
  • Rain gauge data (AWS from RMI): Hourly precipitation measurements from 11 Automatic Weather Stations across Belgium, available since 2003/2004. Used for validation of the analog technique.

Main Results

• A gridded hourly precipitation dataset for Belgium at 1000 m resolution, spanning 1940 to 2016, was successfully generated using the analog technique.
• The most effective predictor set for identifying analog days was based on four geopotential height pressure levels (TWS4) at varying timestamps (85 kPa, 50 kPa, 70 kPa, 100 kPa), outperforming sets based solely on relative humidity or temperature.
• The median of a 25-member analog ensemble provided the best overall performance (highest AUC) for 24-hour precipitation accumulation estimation, with performance plateauing around 23 analogs.
• Model performance was stronger for low precipitation thresholds (1 x 10^-3 m/day) but decreased for higher thresholds (5 x 10^-3 m/day and 15 x 10^-3 m/day), though still maintaining good performance. Higher thresholds benefited more from larger ensemble sizes.
• Geopotential height patterns and autocorrelation of hourly precipitation in analog days closely resembled those of real days, indicating the method reliably captures key dynamical features and temporal consistency.
• The dataset, containing median hourly precipitation fields, is available on Zenodo (https://doi.org/10.5281/zenodo.14965710) in Zarr format.

Contributions

• Development of the first long-term (77 years) high-resolution (1000 m, hourly) gridded precipitation dataset for Belgium using an analog technique.
• Demonstrates the effectiveness of the analog method for downscaling precipitation over a long historical period, providing valuable insights for hydrological applications and climate studies.
• Provides a validated, publicly available dataset (RADCLIM-Analogs) that can be used for diverse applications such as flash flood forecasting, studying precipitation extremes and trends, validating weather and climate models, and training machine learning models for forecasting or downscaling.
• Identifies and validates optimal predictor sets and ensemble sizes for the analog technique in the context of high-resolution precipitation reconstruction.

Funding

Not explicitly stated in the provided text.

Citation

@article{Debrie2025Hourly,
  author = {Debrie, Elke and Demaeyer, Jonathan and Vannitsem, Stéphane},
  title = {Hourly precipitation fields at 1 km resolution over Belgium from 1940 to 2016 based on the analog technique},
  journal = {Earth system science data},
  year = {2025},
  doi = {10.5194/essd-17-6405-2025},
  url = {https://doi.org/10.5194/essd-17-6405-2025}
}