GIDPC.Green Infrastructure · Disaster Prevention
Solutions

Spatial Optimization

Spatial Optimization of Green Infrastructure

XGBoost surrogates and NSGA-II multi-objective optimization produce cost-effective Pareto fronts for green-infrastructure / LID layouts under budget and area constraints.

NSGA-IIpymooXGBoostSHAPOptunaGeoPandas

What this solves

"Where and how much green infrastructure?" is a multi-objective decision with no single right answer. More runoff reduction means more area and cost; reducing cost means reducing effect. The value lies in producing the trade-off curve (Pareto front) under your specific budget and area constraints.

How we analyze

| Step | Tools | Output | |---|---|---| | 1. Scenario definition | OSM, GeoPandas | 0–100% coverage grid | | 2. Physics simulation | SWMM / OpenFOAM | Per-scenario performance | | 3. Surrogate training | XGBoost / GATv2 | R² 0.79–0.96 | | 4. Multi-objective optimization | NSGA-II (pymoo) | Pareto front | | 5. Variable importance | SHAP / Sobol | Priority investment locations | | 6. Bayesian search | Optuna | Fine-tune continuous parameters |

Demonstrated case

For Seocho-gu permeable pavement:

  • 100% coverage delivers 24.8% reduction, but the first 30% delivers 73% of that — a clear cost-effective sweet spot
  • Pareto front: 14.7–22.7% runoff reduction vs 3.8–63.5% installation area
  • SHAP identifies Naegok-dong and other high-leverage sub-districts

Typical deliverables

  • Pareto-front graph + 3–5 ranked recommendation scenarios
  • Sub-district priority map
  • Cost / area / performance scenario table
  • Surrogate model package (for future scenario re-evaluation)