What this solves
Urban-heat danger is governed not by air temperature alone but by thermal-comfort indices — UTCI, PET, and SVF. Computing these at block scale requires resolving 3D momentum, heat, and radiation transport over realistic urban geometry.
OpenFOAM CFD is the credible answer, but each scenario runs in hours to days — too expensive to explore many green-infrastructure design alternatives.
How we analyze
| Step | Tools | Output | |---|---|---| | 1. Urban form modeling | CityGML, GeoPandas, OSMnx | 3D block mesh | | 2. Micro-meteorology CFD | OpenFOAM (Boussinesq) | Per-node wind, temperature, radiation | | 3. AI surrogate training | GATv2 + Transfer Learning | 0.01-second inference | | 4. Multi-objective optimization | Optuna · NSGA-II | Pareto front over tree spacing / LAI / budget | | 5. Indicator computation | UTCI / PET / SVF | Quantitative comparison report |
Demonstrated case
For Seoul Seocho/Gangnam, this solution:
- Trains on 1,500 CFD scenarios and achieves zero-shot R² = 0.845 on adjacent districts
- Identifies street-tree 5–8 m spacing with LAI 4–5 as the top cooling strategy
- Delivers 0.23 °C average and 1.98 °C maximum block-level cooling
Typical deliverables
- 3D urban model and UTCI/PET maps of your site
- Pareto-optimal street-tree and green-roof layout alternatives
- Quantitative comparison report for 3 ranked recommendations
- GNN model package (retrainable) handed off on request