Data & methodology ▾
Grid: 1,493 landscape-metric analysis cells across Dortmund's twelve Stadtbezirke, each assigned to one of five spatial regimes (Core, North, South, East, West City) used by the underlying regime-switching spatial Durbin / spatial-IV model.
2015 & 2025 are observed landscape metrics computed from classified land-use rasters (CNN-LSTM / CA-CNN-LSTM classification pipeline).
2035 is a one-way scenario forecast: the 2025-fitted regime-specific spatial-IV coefficients (log/logit-transformed outcome, bounded reformulation) applied to CNN-LSTM-predicted 2035 land-use. It is illustrative, not a precise prediction — see caveats below.
2030 is intentionally omitted. The methodology report found the 2030 forecast unreliable for a specific, identified reason (a 5-year predicted change standardized against the model's 10-year training distribution), concluding it "should not be used for any purpose beyond illustrating this failure mode."
Forecast reliability by metric: Vegetation cohesion's bounded 2035 forecast is the cleanest (zero bound violations, plausible distribution). Built-up fragmentation's 2035 forecast is usable as illustrative but runs on the highest spatial multiplier (ρ = 0.80) of the three. Vegetation fragmentation's 2035 forecast has a known residual issue: a small share of cells (concentrated in South City) exponentiate to implausibly large values — these are flagged on the map with a dashed outline rather than hidden.
Grey cells have no value for the selected metric/year — excluded by the model for that specific dependent variable (differs slightly by metric).
Grid cells and boundaries reprojected to WGS84; geometry simplified (3 m tolerance) for web delivery.