Szturc et al. (2025) Can we reliably estimate precipitation with high resolution during disastrously large floods?

Identification

• Journal: Hydrology and earth system sciences
• Year: 2025
• Date: 2025-10-21
• Authors: Jan Szturc, Anna Jurczyk, Katarzyna Ośródka, Agnieszka Kurcz, Magdalena Pasierb, Mariusz Figurski, Robert Pyrc
• DOI: 10.5194/hess-29-5405-2025

Research Groups

• Institute of Meteorology and Water Management – National Research Institute, Centre of the Weather Forecasting Service, Warszawa, Poland
• Gdansk University of Technology, Faculty of Civil and Environmental Engineering, Department of Geodesy, Gda´nsk, Poland
• Institute of Meteorology and Water Management – National Research Institute, Hydrological and Meteorological Measurement and Observation Network Centre, Warszawa, Poland

Short Summary

This study evaluates the reliability of various real-time and offline precipitation estimation techniques during a disastrous flood in the upper and middle Odra River basin in September 2024. It finds that rain gauge measurements, radar data adjusted to rain gauges, and multi-source estimates (RainGRS) provide the most reliable high-resolution precipitation fields for flood protection, especially for extreme events in mountainous areas.

Objective

• To examine the real possibilities of precise estimation of a precipitation field with a high spatial resolution of about 1 km and a high temporal resolution of at least 10 min or 1 h during intense precipitation events that caused floods in the upper Odra River basin area in September 2024.
• To verify all available real-time and offline measurements and estimates to determine their applicability and quantify their reliability.

Study Configuration

• Spatial Scale: Upper and middle Odra River basin in south-western Poland, including mountainous areas (approximately 44,000 km²). Precipitation fields were analyzed at a resolution of 1 km.
• Temporal Scale: Flood event occurred from 12 to 15 September 2024. Precipitation accumulations were analyzed for daily (24 h) and hourly (1 h) periods.

Methodology and Data

• Models used:
  • RainGRS system (multi-source merging model)
  • WRF (Weather Research and Forecasting) model (with initial conditions from ICON-EU model)
  • Noah land surface model (within WRF)
• Data sources:
  • Real-time/Near real-time:
    • Telemetric rain gauges (GAU, 10 min, 1.0 km)
    • Weather radars (RAD, RAD Adj; POLRAD network and neighboring countries, 5/10 min, 0.5/1.0 km)
    • Satellite-based precipitation (SAT, H61B; EUMETSAT NWC SAF and H SAF products, 5/10 min, roughly 3.5 km × 6.0 km)
    • Commercial Microwave Links (CMLs, 15 min, 1.0 km)
    • Multi-source estimates from RainGRS system (GRS, 10 min, 1.0 km)
  • Offline (not in real-time):
    • Manual rain gauges (GAU Manual, Hellmann type, 24 h, point-wise) - used as primary reference data.
    • Satellite-based reanalyses (IMERG Final from NASA, 30 min, roughly 7 km × 11 km; PDIR-Now from University of California, Irvine, 1 h, roughly 2.8 km × 4.5 km)
    • NWP-based reanalyses (ERA5 from ECMWF, 1 h, roughly 18 km × 28 km; WRF reanalyses, 1 h, 1.0 km)
• Verification Metrics: Pearson correlation coefficient (CC), Root Mean Square Error (RMSE), Root Relative Square Error (RRSE), and Statistical Bias.
• Verification Strategy: Daily accumulations verified against manual rain gauges. Hourly accumulations verified against RainGRS estimates. Additional verification for extreme precipitation events (daily > 50 mm, hourly > 5 mm).

Main Results

• Most Reliable Real-time Data: Spatially interpolated telemetric rain gauge data (GAU), radar data adjusted to rain gauges (RAD Adj), and multi-source RainGRS estimates (GRS) showed the highest reliability. For daily accumulations, they achieved Pearson correlation coefficients (CC) > 0.95 and Root Mean Square Errors (RMSE) < 15 mm. For hourly accumulations, RAD Adj and GAU had CC > 0.9 and RMSE < 0.6 mm.
• Raw Radar Data: Raw radar estimates (RAD) accurately captured spatial patterns (CC > 0.78 daily, > 0.9 hourly) but significantly underestimated precipitation (Bias -25.71 mm daily, -0.56 mm hourly).
• Commercial Microwave Links (CMLs): CML-based estimates showed fairly good reliability, outperforming satellite data but being less accurate than rain gauge or radar-adjusted data. For daily accumulations, CC = 0.721, RMSE = 32.74 mm, Bias = -20.65 mm. For hourly, CC = 0.673, RMSE = 1.11 mm, Bias = -0.45 mm.
• Satellite-based Estimates: Real-time satellite products (SAT, H61B) and most offline satellite reanalyses (PDIR-Now) performed poorly, with low correlations (CC < 0.5 daily, < 0.26 hourly) and significant underestimation. The IMERG reanalysis was comparatively better but still had limitations (CC = 0.552, RMSE = 33.40 mm daily).
• NWP-based Reanalyses: Mesoscale model simulations (ERA5, WRF) showed higher reliability than satellite data but lower than radar and rain gauge measurements. For daily accumulations, they had CC > 0.7 and RMSE ~26 mm. Their performance for hourly and extreme precipitation was limited. WRF, with its higher spatial resolution, performed slightly better for high daily accumulations.
• Extreme Precipitation: For extreme daily accumulations (> 50 mm), GAU, RAD Adj, and GRS maintained high reliability (CC > 0.85, RMSE < 25 mm). For extreme hourly accumulations (> 5 mm), only RAD Adj showed high reliability (CC = 0.907, RMSE = 0.76 mm). Other methods, including CMLs and NWP models, showed significantly reduced reliability for extreme events.

Contributions

• Provides a comprehensive, quantitative intercomparison of a wide array of real-time and offline precipitation measurement and estimation techniques (rain gauges, radar, satellite, CMLs, NWP models, multi-source) during a major flood event in a complex, mountainous region.
• Quantifies the reliability of these techniques at high spatial (1 km) and temporal (10 min/1 h) resolutions, specifically addressing their performance during extreme precipitation crucial for flood protection.
• Highlights the superior performance of adjusted radar and multi-source merged products (RainGRS) for operational flood monitoring in challenging environments, offering practical guidance for National Meteorological and Hydrological Services.
• Demonstrates the limitations of satellite-only products and mesoscale models for very high-resolution, extreme precipitation events, while also showing the potential of CMLs despite their current underestimation issues.

Funding

No specific funding projects, programs, or reference codes were explicitly listed as funding the research. Simulations using the WRF model were performed at the Academic Computer Centre in Gdansk (CI TASK).

Citation

@article{Szturc2025Can,
  author = {Szturc, Jan and Jurczyk, Anna and Ośródka, Katarzyna and Kurcz, Agnieszka and Pasierb, Magdalena and Figurski, Mariusz and Pyrc, Robert},
  title = {Can we reliably estimate precipitation with high resolution during disastrously large floods?},
  journal = {Hydrology and earth system sciences},
  year = {2025},
  doi = {10.5194/hess-29-5405-2025},
  url = {https://doi.org/10.5194/hess-29-5405-2025}
}