Øydvin et al. (2026) Improving weather radar rainfall estimates by merging with commercial microwave link data: a fully reproducible, large-scale method intercomparison
Identification
- Journal: Hydrology and earth system sciences
- Year: 2026
- Date: 2026-11-13
- Authors: Erlend Øydvin, Elia Covi, Maximilian Graf, Christian Chwala
- DOI: 10.5194/hess-30-7007-2026
Research Groups
- Faculty of Science and Technology, Norwegian University of Life Sciences, Ås, Norway
- Hydro-Meteo-Climate Structure, Arpae Emilia-Romagna, Bologna, Italy
- Department of Hydrometeorology, Deutscher Wetterdienst, Offenbach, Germany
- Institute of Meteorology and Climate Research (IMK-IFU), Karlsruhe Institute of Technology, Garmisch-Partenkirchen, Germany
Short Summary
This study assesses the impact of merging commercial microwave link (CML) data with weather radar for quantitative precipitation estimation (QPE) using two openly available datasets with contrasting observational densities. It finds that merging CML data consistently improves radar QPE, with reductions in mean absolute error (MAE) of up to 41 % and root mean square error (RMSE) up to 19 %, depending on network density and rainfall intensity.
Objective
- To quantify the improvement in QPE accuracy by merging CML data with weather radar compared to raw, unadjusted radar, evaluated against independent rain gauge observations.
- To determine what the relative performances of additive, multiplicative, and external drift paradigms reveal about radar error structures.
- To evaluate how spatial interpolation geometries (point vs. block kriging) and geostatistical choices influence merging performance across contrasting sparse and dense CML networks.
Study Configuration
- Spatial Scale:
- OpenMRG dataset: Gothenburg region, Sweden, covering approximately 2800 km², with 364 CMLs (density of 1 CML per 7.7 km²) and 11 rain gauges. Radar data at 2 km resolution.
- OpenRainER dataset: Emilia-Romagna region, Italy, covering approximately 21,000 km², with 153 CMLs (density of 1 CML per 137 km²) and 280 rain gauges.
- Local interpolation used a neighborhood of 70 radar–CML pairs.
- Temporal Scale:
- OpenMRG: June to August 2015 (3 months). CML data recorded every 10 s, resampled to 1 min. Radar data at 5 min resolution. Rain gauge data at 15 min resolution.
- OpenRainER: June to August 2022 (3 months). CML data recorded instantaneously at 1 min resolution. Radar data at 5 min reflectivity and 15 min accumulated rainfall maps. Rain gauge data at 15 min resolution.
- Merging process applied independently for each time step, with both radar and CML data aggregated to an hourly resolution.
Methodology and Data
- Models used:
- Merging paradigms: Additive merging, Multiplicative merging, Kriging with External Drift (KED).
- Interpolation techniques: Inverse Distance Weighting (IDW), Ordinary Kriging (OK), Block Kriging.
- CML processing: Total signal loss calculation, quality control (e.g., diurnal cycle, noisy signals), wet–dry classification (using radar), baseline estimation, Wet Antenna Attenuation (WAA) correction (KR-alt model), k–R relation (ITU (2005) power-law).
- Radar processing: Marshall–Palmer Z–R relation for reflectivity to rainfall conversion.
- Software:
mergeplg(Python package for merging),pycomlink(CML data processing),poligrain(geospatial operations and metrics).
- Data sources:
- OpenMRG dataset: CMLs, weather radar composite (from 3 SMHI radars), and rain gauges (1 SMHI, 10 municipality gauges).
- OpenRainER dataset: CMLs, weather radar composite (from 2 radars), and rain gauges (280 distributed gauges).
- Both datasets are openly available and feature contrasting observational densities.
Main Results
- Merging CML data with weather radar consistently improved QPE accuracy compared to unadjusted radar fields across both datasets.
- Dense Network (OpenMRG): All merging methods showed similar performance, reducing MAE by 20 % (from 0.84 mm to 0.67 mm), RMSE by 11 % (from 1.44 mm to 1.28 mm), and increasing PCC from 0.55 to 0.69. Additive methods were more effective for capturing peak rainfall intensities.
- Sparse Network (OpenRainER): Multiplicative merging achieved the best overall performance, reducing MAE by 41 % (from 2.0 mm to 1.18 mm) and RMSE by 19 % (from 3.6 mm to 2.9 mm), and reducing BIAS from 107 % to -1 %. KED also performed well (38 % MAE reduction, 17 % RMSE reduction). Additive methods performed best during heavy and extreme rainfall events.
- Radar Error Structures: Multiplicative merging's superior performance in sparse networks for low to moderate rainfall suggests that radar errors are largely intensity-dependent (e.g., Z–R parameter mismatch). Additive methods' strength in extreme events indicates localized, offset-type errors.
- Geostatistical Parameters: Merging performance was robust to variogram parameter choices, provided a non-zero nugget was used. A zero nugget led to worse performance, particularly for multiplicative methods in sparse networks. Block kriging showed no consistent advantage over point kriging.
- Range Checks: Robust range checks were essential for stability, especially for multiplicative methods and in the sparse OpenRainER dataset, preventing merged fields from performing worse than raw radar.
Contributions
- Development and open-source release of
mergeplg, a Python package for reproducible CML-radar rainfall data merging. - A comprehensive, fully reproducible, large-scale intercomparison of CML-radar merging methods using two openly available datasets with contrasting network densities.
- Detailed analysis of the relative performances of additive, multiplicative, and Kriging with External Drift (KED) merging paradigms, and the impact of spatial interpolation geometries (point vs. block kriging) and geostatistical choices.
- Insights into the nature of radar error structures and the influence of network density, rainfall intensity, and data quality on merging effectiveness.
Funding
- COST Action CA20136 (OPENSENSE – Opportunistic Precipitation Sensing Network)
- Project CLIMAXPO (grant no. LIFE21-IPC-IT-LIFE CLIMAX PO-101069928)
- MODMET agreement (2024/25) between the Italian National Civil Protection Department and Arpae-SIMC (funding for E. Covi)
- Karlsruhe Institute of Technology (KIT) covered the article processing charges.
Citation
@article{Øydvin2026Improving,
author = {Øydvin, Erlend and Covi, Elia and Graf, Maximilian and Chwala, Christian},
title = {Improving weather radar rainfall estimates by merging with commercial microwave link data: a fully reproducible, large-scale method intercomparison},
journal = {Hydrology and earth system sciences},
year = {2026},
doi = {10.5194/hess-30-7007-2026},
url = {https://doi.org/10.5194/hess-30-7007-2026}
}
Original Source: https://doi.org/10.5194/hess-30-7007-2026