Regime Landscape Metrics Analysis

35 Years of Fragmentation, by City Sector

Class-level FRAGSTATS metrics computed directly from classified and CA-CNN-LSTM-predicted land-use rasters, 2000–2035, averaged by the same five city-sector regimes used across this repository's other Dortmund case studies — plus a formal structural-plausibility check on the model's own forecasts.

R (terra, landscapemetrics, sf) Python (pandas, GeoPandas) Leaflet

Source on GitHub

Setup

Raw Rasters to Regime-Level Trends

Eight classified/predicted land-use rasters (2000, 2005, 2010, 2015, 2020 observed; 2025, 2030, 2035 CA-CNN-LSTM predicted) were cropped to each of the 1,493 grid cells, run through landscapemetrics::calculate_lsm(), then averaged up to the five city-sector regimes (Core/North/East/South/West City).

Time span2000–2035 (5-yr steps) MetricsPD, ED, AI, COHESION, LPI, SHDI Regimes5 city sectors

Open interactive regime metrics monitor

01 — Fragmentation Trajectory

Rise Through 2020, Then a Predicted Reversal

Patch Density and Edge Density — both fragmentation indicators — climb steadily across every regime from 2000 through 2020, then the CA-CNN-LSTM forecast shows all five regimes consolidating from 2025 onward: fewer, larger patches by 2035 than at the 2020/2025 peak.

Patch Density (patches / 100 ha)

South City peaks highest (17.6, 2020); Core City stays lowest throughout

Line chart of Patch Density by city-sector regime from 2000 to 2035, observed solid and predicted dashed, showing a rise through 2020 then a decline — Patch Density by regime — observed (solid) vs. CA-CNN-LSTM predicted (dashed)

South City peak (2020)17.63 South City 203512.53 (−29%)

Edge Density (m / ha)

West City's predicted 2025→2030 drop is the single sharpest swing in the dataset

West City's Edge Density falls from 135.4 (2025) to 103.6 (2030) — a swing far outside its own historical rate envelope, and exactly the kind of transition the structural plausibility check below is designed to catch.

Line chart of Edge Density by city-sector regime from 2000 to 2035, observed solid and predicted dashed, showing a sharp predicted decline after 2025 — Edge Density by regime — the 2025→2030 transition is flagged as implausible in all 5 regimes (see below)

02 — Structural Plausibility Check

Not All Forecast Transitions Are Equally Trustworthy

Each regime/metric's predicted 5-year rate of change (2025→2030 and 2030→2035) is tested against that same regime/metric's own historical mean ± SD annual rate (2000–2020). A prediction outside that envelope is flagged — surfaced on the map and here, not hidden.

Heatmap grid of flagged versus consistent metric-regime combinations for the 2025 to 2030 and 2030 to 2035 prediction transitions — Structural plausibility flags by metric × regime × transition

Edge Density, 2025→2030Flagged in all 5 regimes Cohesion & LPINever flagged Most-flagged regimeNorth City (3 of 6 metrics)

This is the same caveat already documented qualitatively in the author's Dortmund Climate-Adaptation Landscape Monitor ("2030 forecasts found unreliable for a specific, identified reason") — here it's backed by a formal, per-regime, per-metric statistical test rather than a single documented failure case.

Methods & Tools

Landscape MetricsFRAGSTATS-style class metrics via R's landscapemetrics package (PD, ED, AI, COHESION, LPI, SHDI)

Raster Processingterra crop/mask per zone, CRS reprojection to EPSG:25832

Regime AggregationZone → Stadtbezirk → 5-sector regime crosswalk, area-weighted averaging

Trend AnalysisLinear trend slopes per regime × metric, historical-envelope forecast validation

OutputInteractive Leaflet monitor with plausibility-flag overlay