An intercomparison of radar adjustment methods, with focus on radar–CML merging, using open datasets.
This repository contains the code for the paper:
Øydvin, E., Covi, E., Graf, M., and Chwala, C.: Improving weather radar rainfall estimates by merging with commercial microwave link data: a fully reproducible, large-scale method intercomparison, Hydrol. Earth Syst. Sci., accepted, 2026.
The discussion version (preprint) is available at https://doi.org/10.5194/egusphere-2025-6371.
Radar–CML merging methods (additive, multiplicative and kriging with external drift, using IDW, point kriging and block kriging) are compared on two open datasets: OpenMRG (Gothenburg, Sweden) and OpenRainER (Emilia-Romagna, Italy). Merging is done with mergeplg.
git clone --recursive https://github.com/OpenSenseAction/radar_adjustment_intercomparison
cd radar_adjustment_intercomparison
conda env create -f environment.yml
conda activate radar_adjustment_intercomparisonmergeplg, poligrain, pycomlink and OPENSENSE_sandbox are included as git
submodules, pinned to the versions used in the paper. The notebooks import them
from the submodule folders, so nothing needs to be pip-installed.
Run the notebooks in numerical order. The three 3_adjust_radar_*check notebooks
are independent of each other and of 3_adjust_radar.
| Notebook | Purpose | Output |
|---|---|---|
1_data_preparation |
Download and prepare OpenMRG and OpenRainER (CML, radar, gauges) | data/andersson_2022_OpenMRG/, data/covi_2024_OpenRainER/ |
2_cml_processing |
CML quality control, wet/dry classification and rain rate retrieval | data/processed_cml_*.nc |
3_adjust_radar |
Merge radar and CML data with all methods | data/adjusted_fields/ |
3_adjust_radar_nnearcheck |
Sensitivity analysis: number of neighbours | data/metrics/ |
3_adjust_radar_rangecheck |
Sensitivity analysis: range checks | data/metrics/ |
3_adjust_radar_variogramcheck |
Sensitivity analysis: variogram estimation | data/metrics/ |
4_analysis |
Evaluation of radar and merged fields against rain gauges | data/metrics/ |
5_paper_images |
Overview, performance and sensitivity figures | figures/ |
6_case_study |
Case-study figures | figures/ |
7_variogram_study |
Variogram parameter study | figures/ |
8_compress_adjusted_fields |
Lossless compression and metadata for archiving the merged fields on Zenodo (not needed to reproduce the results) | data/adjusted_fields_compressed/ |
data/metrics/ contains all evaluation results used in the paper, so most figures
can be reproduced without downloading data or running the merging. A full
reproduction runs all notebooks in order: notebook 1 downloads about 25 GB of
data, and merging (notebook 3 and its sensitivity checks) takes several hours
and about 15 GB of disk space.
The merged rainfall fields from 3_adjust_radar.ipynb are available on Zenodo:
https://doi.org/10.5281/zenodo.23367571. To skip the merging, download the files
into data/adjusted_fields/ instead of running 3_adjust_radar.ipynb.
The figures fall into the following groups:
Dataset overview (5_paper_images)
Map of the CML networks, gauges and radar rainfall, CML length and frequency
distributions, and data availability.
Requires notebooks 1–2.
Overall performance (5_paper_images)
Metrics of radar and merged fields against rain gauges: overall, by rainfall
intensity, and by distance from the nearest CML.
Runs from the included data/metrics/.
Sensitivity analyses (5_paper_images)
Effect of variogram estimation strategy, range checks and number of neighbours
on merging performance.
Runs from the included data/metrics/ (produced by the 3_adjust_radar_*check notebooks).
Variogram study (7_variogram_study)
Variogram parameters estimated from CMLs, gauges and radar over time.
Requires notebooks 1–2.
Case studies (6_case_study)
Merged rainfall fields and their differences from radar and gauges for selected
time steps, for both datasets.
Requires notebooks 1–3.
Both datasets are downloaded automatically by 1_data_preparation.
- OpenMRG: Andersson et al. (2022), version 1.1, https://zenodo.org/records/7107689 (CC BY-SA 4.0)
- OpenRainER: Covi and Roversi (2025), version 2.0.0, https://zenodo.org/records/14731404 (CC BY 4.0)