🟢 Green = Low Priority → 🔴 Red = High Priority Interactive ranking of Montréal municipal buildings by operational carbon intensity and flood exposure, with optional social/greening dimensions.
| View | Screenshot |
|---|---|
| Global view |  |
| Building details popup |  |
| Heat distribution |  |
Municipalities need to target retrofits where they deliver the greatest combined benefit: reducing building GHG emissions and reducing exposure to extreme rainfall and flooding. UrbanIQ provides a reproducible, auditable pipeline that turns disparate data into a multi-criteria priority score, and an interactive map for spatial triage.
- Mitigation: prioritize assets with high operational carbon intensity (tCO₂e/m²).
- Adaptation: factor hydrological exposure via flood proximity/risk.
- Equity & nature-based adaptation: integrate social vulnerability and urban greening indicators when available.
- Unified ingest from a single “all-in-one” CSV (e.g.,
finaloutputUpdate1.csv). - Transparent MCDA scoring:
priority_score = wE·energy_norm + wF·flood_norm + wS·social_norm + wG·green_norm, with non-negative weights summing to 1. - Robust normalization (1st–99th percentile) for comparability across the portfolio.
- Optional metaheuristics (GA / PSO / SA / multi-algorithm) to calibrate weights for maximal discriminative power.
- Analyst diagnostics: isolation-forest outliers, DBSCAN clusters, bootstrap p10/p90 score bands.
- Kepler.gl web viewer: color encodes priority; size can encode energy burden; sidebar for top-N and details.
-
Primary input:
./data/finaloutputUpdate1.csv(building attributes, coordinates, year, area, emissions, flood proximity, climate context; latest run ~2,070 rows). -
Produced artifacts:
./outputs/buildings_ranked.csv— normalized drivers, composite scores, diagnostics../outputs/buildings_resilience.geojson— WGS84 points for the web map.
A sample combined CSV is included: outputs data/finaloutputUpdate1.csv.
-
Install dependencies Create and activate a virtual environment, then install requirements:
python -m venv .venvsource .venv/bin/activate(Windows:.venv\Scripts\activate)pip install -r requirements.txt -
Run the ranking pipeline (combined CSV mode)
python src/run_pipeline.py --data_dir ./data --outputs_dir ./outputs --combined_csv finaloutputUpdate1.csv
Optional: enable metaheuristic weight search
python src/run_pipeline.py --data_dir ./data --outputs_dir ./outputs --combined_csv finaloutputUpdate1.csv --optimize_weights --optimization_algorithm multi --max_time_seconds 300 --comparison_metric fitness
- View the interactive map
Open
index.htmlin your browser. • Local file mode: double-clickindex.html. Iffetch()is blocked underfile://, click Load GeoJSON and select./outputs/buildings_resilience.geojson. • Local server mode (recommended):python -m http.server 8000then visithttp://localhost:8000/index.html.
-
Drivers Energy: operational carbon intensity (tCO₂e / m²) =
Emissions_GES / Superficie. Flood: proximity-based exposure (higher when closer to flood zones). Social & Greening: incorporated when present, standardized like other drivers. -
Normalization Robust scaling (1st–99th percentile) to [0, 1], preserving rank order while dampening outliers.
-
Composite score Convex combination of normalized drivers with non-negative weights summing to one. Default weights reflect a mitigation–risk balance and can be tuned to policy.
-
Calibration (optional) Metaheuristics search weight vectors that maximize discriminative power (higher variance and range, balanced correlation with drivers, regularization to avoid one-metric dominance). Multi-algorithm mode orchestrates SA → PSO → GA with time caps and selects the best outcome by fitness, stability, or efficiency.
-
Diagnostics Outlier flags, clustering, and bootstrap uncertainty bands (p10/p90) quantify stability and highlight assets sensitive to assumptions.
- Edit default paths and weights in
src/config.py. - Provide YAML overrides via
--optimization_config config/optimization_config.yml. - Logging clarifies resolved paths, year filters, and modeling decisions (for example: flood exposure derived from
distance_inondables).
src/run_pipeline.py— CLI entry pointsrc/pipeline.py— end-to-end flow (ingest → features → score → export)src/scoring.py— normalization + MCDA compositesrc/preprocess.py— numeric/string cleaning utilitiessrc/ml_extras.py— outliers, clusters, bootstrap bandssrc/weight_optimizer.py— GA, PSO, SA, multi-algorithm orchestratorsrc/optimization_analyzer.py— optional visual analysis/reportingsrc/io_utils.py,src/logging_utils.py,src/joiners.py,src/config.pyindex.html— kepler.gl web viewer (loads GeoJSON dynamically)outputs/— artifacts: CSV + GeoJSONdata/— input combined CSV
- Reproducibility: deterministic feature construction, explicit logging, and a self-contained GeoJSON + HTML viewer enable turnkey replication and CKAN packaging.
- Scenario analysis: adjust weights to explore policy stances; use p10/p90 score bands to communicate sensitivity.
- Extensibility: plug calibrated flood layers (e.g., return-period polygons), social indices, and greening metrics into the same decision scaffold.
Timothy Roch • Kevin Mock • Gamaliel Kalefe • Justin Joel Takodjou