Where infrared.city fits
3,204,804 simulations run on Infrared Core (as of 29 Oct 2026).
infrared.city runs 12 models from one engine, Infrared Core, among them wind, wind comfort, sun, daylight, radiant heat and felt heat (UTCI). It works on real 3D building and terrain meshes, in a browser app, through an API and SDK, and in plugins for Grasshopper, QGIS and ArcGIS. Felt heat for 1 km² takes about 5 s once the geometry is uploaded, and 400 km² took 44 s as four 100 km² runs in parallel. Run times are not like for like across tools: area, grid, simulated hours and physics differ. Sun, sky and radiant heat come from ray tracing. Wind comes from an AI model trained on CFD results, not from a CFD run for each request.
- You need real 3D geometry: arcades, overhangs and sloped roofs stay as they are.
- You need wind, wind comfort, sun and daylight next to radiant heat, from the same engine.
- You need a hosted API, a browser app or an hourly city layer, with no local compute.
Use both
SOLWEIG is open code that researchers can read and change. infrared.city is a hosted service for project work at city scale.
Teams keep SOLWEIG for research and as a reference, and use infrared.city for studies with many options or large areas. We use SOLWEIG as the reference in our own field tests.
Facts side by side
| infrared.city | SOLWEIG | |
|---|---|---|
| Method | Ray tracing for sun, sky and radiant heat; wind from an AI model trained on CFD results (no CFD run per request) | 2.5D model: shadows, sky view and radiant heat for each pixel of a height raster |
| Run time | About 5 s for 1 km² of felt heat (UTCI, 08 to 18 h, geometry uploaded, measured); 400 km² in 44 s (four runs in parallel) | About 39 km² at 2 m, 24 h, in 254 s on one A100 GPU; 1,285 s on CPU (one published run) |
| Inputs | Real 3D building and terrain meshes, weather data; trees and ground materials for heat | Height rasters for buildings, vegetation and terrain, and weather data |
| Runs as | Cloud: browser app, API and SDK, plugins | Desktop: QGIS (UMEP) or Python, CPU or GPU |
| Licence and pricing | Tokens: free start with 900 tokens, plans from €240 a year | Free, GPL-3.0 |
The run times are not like for like: area, grid and simulated hours differ.
Our field test, same input for both models (same points, weather and 3D scene). Radiant heat (Tmrt) RMSE in kelvin, infrared.city / SOLWEIG: Gothenburg 4.5 / 5.0 (in-sample), Phoenix 11.1 / 14.3 (in-sample), Melbourne 11.8 / 12.5 (held-out), Guelph 9.7 / 12.3 (held-out). All sites and method.
The infrared.city values repeat the full comparison. Each value for SOLWEIG comes from the sources below.
Sources
- UMEP manual: SOLWEIG , accessed 24 Oct 2026.
- Zonato et al. (2026), Geosci. Model Dev. 19:7389 , accessed 24 Oct 2026.
- Kamath et al. (2026), SOLWEIG-GPU, JOSS 11(118):9535 , accessed 24 Oct 2026.