SynthRain

Synthetic CML network + synthetic wet/dry + synthetic rainfall intensity -> IDW interpolation -> PNGs.

This is a small standalone testbed to stress-test TelcoRain-like interpolation behaviour without InfluxDB/MariaDB.

Install

pip install numpy pandas scipy matplotlib

Quick start

python run_scenarios.py --out out_demo --seed 0 --city --n-sites 250 --mean-degree 3 \
  --wet-target 0.35 --wet-mode threshold --flip-dry-to-wet 0.03 --flip-wet-to-dry 0.15 \
  --grid-step-m 1000 --idw-dist-m 30000 \
  --idw-near 12 --idw-power 2.0 --dry-as-zero

Outputs (in out_demo/):

• true_field.png : underlying synthetic “ground truth” rainfall field
• links.png : link centers with wet/dry markers
• idw_field.png : interpolated rainfall field from links
• diff.png : interpolated minus true (useful for bias/artefacts)

Use --dry-as-zero to force dry links as 0 mm/h (strongly affects IDW output). Omit --dry-as-zero to ignore dry links (IDW uses only wet links).

Notes

• Bbox is lon_min,lon_max,lat_min,lat_max (degrees). Default is a rough Czechia bbox.
• Grid is built in meters using Web Mercator (like TelcoRain with use_mercator=True). Use --grid-step-m (e.g. 1000 for ~1 km) and --idw-dist-m (e.g. 30000 for 30 km). If you really want to do everything in degrees, pass --no-mercator (not recommended).
• Wet/dry labeling:
  • --wet-mode threshold: picks a threshold so ~wet_target links are wet (can create contiguous wet/dry regions).
  • --wet-mode random: wet labels are i.i.d. random.
  • --wet-mode stratified: random within coarse spatial strata (--wet-strata-nx/ny) to spread wet links evenly.

Using an INI config (TelcoRain-like)

Instead of passing many CLI flags, you can use:

python run_scenarios.py --config example_config.ini

CLI flags override config values, e.g.:

python run_scenarios.py --config example_config.ini --wet-target 0.15 --seed 3

Why both idw_near and idw_dist_m?

This matches TelcoRain’s behaviour:

• idw_dist_m sets a maximum radius (meters) for eligible neighbours.
• idw_near caps the maximum number of neighbours actually used per grid point (after filtering by radius).

In dense networks, without idw_near, a 30 km radius could include hundreds/thousands of links per pixel.

Wet/dry distribution modes

• wet_mode=threshold: wet links are chosen by rainfall intensity quantile (can create coherent wet regions).
• wet_mode=random: wet links are chosen uniformly at random (spatially “salt-and-pepper”).
• wet_mode=stratified: random wet selection within coarse spatial bins (more spatially balanced than pure random).

Acknowledgements

This output was financed through the project Precipitation Detection and Quantification System Based on Networks of Microwave Links (SS06020416) is co-funded with state support from the Technology Agency of the Czech Republic under the Environment for Life Programme. The project was further funded within the National Recovery Plan from the European Recovery and Resilience Facility.