EuroFab Progress Meeting

Eurofab team

2024-04-04

Agenda

  1. Progress overview
  2. Technical note D4 - Test and Verification Results
  3. Technical note D6 - Software and Example datasets generated during Verification Exercises
  4. Technical note D5 - Stakeholder Impact and Utility Assessment
  5. Discussion

Where we are

Technical note D4 - Test and Verification Results

Morphometric model

Extensions & changes

Results

Agg scores

Class scores

Confusion Matrix

Progress on imagery components

Approach

pipeline.spatial_sig_prediction(
    geo_path="../data/london_grid.geojson",  # Area to analyze
    vrt_file="../data/2017_satellite.vrt",   # Satellite composite
    xgb_weights="../models/xgb_model.bin",   # XGBoost model
    model_weights="../models/satlas.pth",    # Satlas embedding model
    output_path="../output/test_london.parquet",  # Prediction results
    h3_resolution=5
)

Data augmentation

Sliding Window Augmentation (50m)

Data augmentation

Classifier

Classes (K) Spatial Context Accuracy Macro Accuracy Macro F1 Score
7 None 0.4924 0.3856 0.3389
7 H3 (res 5) 0.6959 0.5713 0.5221
12 None 0.4617 0.2666 0.2127
12 H3 (res 5) 0.6654 0.4328 0.3654

Without data augmentation

Classes (K) Spatial Context Accuracy Macro Accuracy Macro F1 Score
7 None 0.4 0.2 0.14

Number of observations vs accuracy

Examples

Datacube

  • predictions for all of GB
  • 2016-2021
  • K=7 and K=12

Technical note D6 - Software and Example datasets generated during Verification Exercises

Software and datasets

  • Interactive morphometric web application
  • Morphometric characterisation pipeline for Microsoft Building footprints
  • Morphometric characterisation pipeline for Overture Maps Building footprints
  • AI Method for Urban Fabric classification and morphometric characterization
  • AI temporal data cube of Urban Fabric classifications

Technical note D5 - Stakeholder Impact and Utility Assessment

Major stakeholder engagements

  • European Covenant of Mayors
  • SSVA (Construction Sector Development Agency of the Ministry of Environment, Lithuania)
  • Prague Institute of Planning and Development (IPR)
  • 4ct

Takeaways

  • Data Integration
  • Taxonomy Description and Naming
  • Comparative Urban Analysis
  • Expanding Data Coverage
  • Reducing Manual Effort in Land Use Analysis
  • Geographical Scale of Results
  • Taxonomic Tree & Evaluation
  • Input data quality

Discussion