Dzwonkowski et al. (2026) Evaluation of radar-based precipitation estimates during a flood event using rain gauge validation

Identification

• Journal: Scientific Reports
• Year: 2026
• Date: 2026-04-02
• Authors: Karol Dzwonkowski, Ireneusz Winnicki, Sławomir Pietrek, Krzysztof Kroszczyński
• DOI: 10.1038/s41598-026-40456-z

Research Groups

• Institute of Geospatial Engineering and Geodesy, Faculty of Civil Engineering and Geodesy, Military University of Technology, Warsaw, Poland

Short Summary

This study evaluates the accuracy of radar-based precipitation estimates using classical empirical and novel polarimetric Z-R relationships during an extreme flood event in Poland. It found that a locally calibrated polarimetric relationship (ZDR3) significantly improved rainfall estimation accuracy, particularly by reducing bias, compared to standard methods.

Objective

• To evaluate the quality of radar-derived precipitation data relative to ground-based measurements from selected meteorological stations, and to analyze the effectiveness of spatiotemporal matching methods.
• To assess rainfall totals using three empirical Z-R relationships (Marshall–Palmer, Wexler, Doumoulin–Cogombles) at seven height levels (1.0 km to 2.0 km AGL) and three polarimetric relationships based on ZDR at seven reference levels (1.0 km to 2.0 km AMSL).
• To determine the extent to which radar data can serve as a reliable source for meteorological, hydrological, aviation, and early warning applications for extreme weather events.

Study Configuration

• Spatial Scale: Central and Eastern Sudetes, Poland, including intermontane basins and mountain ranges (e.g., Kłodzko Basin, Śnieżnik Massif). Radar data spatial resolution: 1 km × 1 km. Validation window: 5 × 5 pixels (25 km²) centered on rain gauges. Altitude levels for Z-R: 1.0 km, 1.2 km, 1.4 km, 1.5 km, 1.6 km, 1.8 km, and 2.0 km above ground level (AGL). Altitude levels for ZDR: 1.0 km, 1.2 km, 1.4 km, 1.5 km, 1.6 km, 1.8 km, and 2.0 km above mean sea level (AMSL).
• Temporal Scale: Flood event from 13 September 2024 at 10:00 UTC to 14 September 2024 at 21:00 UTC (35 hours). Radar temporal resolution: 5 minutes. Rain gauge data: Hourly precipitation totals.

Methodology and Data

• Models used:
  • Classical empirical Z-R relationships: Marshall–Palmer, Wexler, Doumoulin–Cogombles.
  • Polarimetric Z-ZDR-R relationships: ZDR1 (literature-based), ZDR2 (locally calibrated), ZDR3 (locally calibrated).
  • RADVOL-QC algorithm for radar data quality control (ground clutter, non-meteorological echoes, beam blockage, attenuation correction).
  • Statistical indicators: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Bias, Pearson correlation coefficient (r).
  • Statistical tests: Kruskal–Wallis test (two-sided, α=0.05) and Dunn’s post-hoc test with Bonferroni correction.
• Data sources:
  • Remote sensing: Dual-polarization C-band weather radar data from the Polish POLRAD network, specifically from Pastewnik (PAS) and Góra Św. Anny (GSA) radars. Radar variables included reflectivity (Z), differential reflectivity (ZDR), specific differential phase shift (KDP), dual-polarization surface rainfall intensity (DPSRI), and precipitation accumulation (PAC).
  • In situ observations: Hourly precipitation totals from 21 rain gauges operated by the Institute of Meteorology and Water Management – National Research Institute (IMWM-NRI) within the study area.

Main Results

• For the analyzed extreme flood event, the locally calibrated polarimetric ZDR3 relationship provided the most reliable precipitation totals, with RMSE = 2.20 ± 0.90 mm, MAE = 1.84 ± 0.73 mm, and Bias = -0.67 ± 0.81 mm (GSA radar).
• The ZDR3 method exhibited approximately 69% lower Bias compared to the standard operational Marshall-Palmer method.
• Polarimetric relationships (ZDR2 and ZDR3) showed lower and more stable error metrics (median RMSE ~2.3 mm, MAE ~1.8 mm) compared to classical Z-R relationships (median RMSE ~3.8–4.2 mm, MAE ~3.2–3.6 mm).
• Radar-derived precipitation generally underestimated actual rainfall, particularly in mountainous areas, with errors increasing with station elevation (R² for RMSE = 0.25, MAE = 0.23, Bias = 0.23).
• The GSA radar, despite being further from the study area, provided higher accuracy than the PAS radar (RMSE ~0.17 mm lower, MAE ~0.20 mm lower, Bias ~0.31 mm higher (less underestimation) for GSA).
• Differences in RMSE values across the seven analyzed height levels were minimal (e.g., 0.01 mm for ZDR3), indicating that data height had no substantial impact on accuracy for the best methods within the study region.
• The Pearson correlation coefficient (r) ranged from 0.74 to 0.76 for all methods, confirming a strong linear relationship between radar estimates and rain gauge measurements.

Contributions

• Demonstrated the operational superiority of locally calibrated polarimetric Z-ZDR-R relationships (specifically ZDR3) over classical Z-R relationships for extreme flood events in complex terrain, achieving a 69% reduction in bias compared to the Marshall-Palmer method.
• Provided a novel approach by verifying the vertical structure of the precipitation field across seven distinct altitude levels for both empirical and polarimetric relationships.
• Highlighted that in complex mountainous regions, the radar beam's position relative to precipitation layers is more critical for accuracy than mere proximity to the radar.
• Confirmed the significant influence of terrain morphology on radar precipitation estimation errors, with errors systematically increasing with station elevation and in areas of complex orography.
• Emphasized the practical significance of improved accuracy for early warning and crisis management systems, despite the lack of statistical significance in pairwise post-hoc tests due to high natural variability in mountainous precipitation data.

Funding

• Military University of Technology in Warsaw, Poland, Faculty of Civil Engineering and Geodesy, Institute of Geospatial Engineering and Geodesy, Statutory Research Funds UGB 2026 WAT.

Citation

@article{Dzwonkowski2026Evaluation,
  author = {Dzwonkowski, Karol and Winnicki, Ireneusz and Pietrek, Sławomir and Kroszczyński, Krzysztof},
  title = {Evaluation of radar-based precipitation estimates during a flood event using rain gauge validation},
  journal = {Scientific Reports},
  year = {2026},
  doi = {10.1038/s41598-026-40456-z},
  url = {https://doi.org/10.1038/s41598-026-40456-z}
}