bluegreencity-ki

Subway Tunnel Criterion

Overview

The subway_tunnel criterion evaluates drought risk from proximity to underground Stadtbahn/subway tunnels. Tunnels limit the available rooting volume, can act as local drainage (lowering the groundwater table) and warm the surrounding soil — all factors that increase drought stress for nearby street trees (original motivation: issue #139, based on a suggestion by LHH).

The criterion is currently configured with weight 0.0: it contributes extras and the ML feature subway_tunnel_dist_m to the AutoGluon training dataset, but not to the weighted total score (pending weight recalibration).

Implementation

Location

• Module: bluegreencity_ki/criteria/subway_tunnel.py
• Loader: bluegreencity_ki/io/overture_loader.py (load_subway_tunnels)
• Test: tests/test_subway_construction.py

Data Source

• Overture Maps transportation theme (GeoParquet on S3, fetched via the overturemaps Python package)
• Segments with subtype=rail, class=subway, filtered to sections whose rail_flags contain is_tunnel
• For Hannover (verified 2026-07): 39 segments, all flagged is_tunnel — the Stadtbahn tunnel network (A-, B-, C-Tunnel), total ~29.7 km
• Tunnel alignments are static infrastructure → cached as GPKG (data/derived/transport/subway_tunnels.gpkg, refresh_days: 365, stale-cache fallback)

The issue originally referenced the Basis-DLM (ver03_l, BKT=1200), which contains the whole Stadtbahn network including surface sections. Overture was chosen instead because its is_tunnel flag directly identifies the underground sections.

How It Works

1. Load tunnels: GPKG cache or Overture fetch (bbox from config).
2. Nearest distance: sjoin_nearest computes the distance from each tree to the nearest tunnel segment.
3. Distance breaks: The distance is mapped to a risk score via configurable breaks.

Scoring Logic

distance ≤ 15 m  → 0.8   (tree directly above/beside the tunnel)
distance ≤ 30 m  → 0.5   (within the influence zone)
distance > 30 m  → 0.1   (no tunnel influence)

Trees receive the neutral score when no tunnel data is available.

Configuration

inputs:
  overture_transportation:
    bbox_wgs84: [9.60, 52.30, 9.92, 52.46]
    rail_classes: ["subway"]
    tunnel_only: true
    cache_path: "./data/derived/transport/subway_tunnels.gpkg"
    refresh_days: 365

criteria:
  - name: "subway_tunnel"
    weight: 0.0
    params:
      distance_breaks: [15, 30]
      distance_scores: [0.8, 0.5, 0.1]

Parameters

Parameter	Type	Default	Description
distance_breaks	list[float]	[15, 30]	Distance thresholds in metres
distance_scores	list[float]	[0.8, 0.5, 0.1]	Risk score per class (one more than breaks)

Output Fields

1. subway_tunnel_risk — risk score (0–1)
2. subway_tunnel_dist_m — distance to the nearest tunnel segment in metres (ML feature)

Known Limitations

1. Weight 0.0 until weights are recalibrated.
2. 2D distance only: tunnel depth is not considered.
3. Overture completeness: relies on OSM-derived tunnel mapping quality.

Changelog

2026-07-22: Initial Implementation (#139)

• Overture loader with is_tunnel filtering and GPKG cache
• Distance-break scoring, subway_tunnel_dist_m as AutoGluon feature

---

Author: BlueGreenCity-KI Team Last Updated: 2026-07-22