GRANULAR Rural Typology Classifier: K-Means-based sub-classification of European rural grid cells (DEGURBA Level 2)

This dataset and accompanying Jupyter notebook classify European 1 km² grid cells into six granular rural typology categories, refining the three rural classes of the DEGURBA (Degree of Urbanisation) Level 2 framework — 311 (very low density), 312 (low density), and 313 (rural cluster) — using unsupervised machine learning. Method For each of the three DEGURBA rural classes, grid cells are clustered independently using MiniBatch K-Means on seven engineered features spanning three dimensions: Demographic: log-transformed population change, 2011–2021 Physical: log building volume (GHS-BUILT-V), hemeroby index (degree of human landscape modification), log elevation Functional: composite accessibility measures (travel time and town counts) to towns of ≥5k, ≥10k, and ≥50k population Features are scaled with a RobustScaler prior to clustering. The number of clusters (k = 2–5) is selected automatically per DEGURBA class via silhouette analysis. Within each class, the resulting clusters are collapsed into a two-level hierarchy: the largest cluster by cell count is assigned rank 1 (majority type) and all remaining clusters are assigned rank 2 (minority type), producing a stable two-digit suffix regardless of how many clusters silhouette analysis selects. Output typology Combining each DEGURBA class with its rank produces six four-digit GRANULAR codes: Code Label 3111 Open rural cells 3112 Open rural cells in peri-rural areas 3121 Rural settlements in peri-rural areas 3122 Rural settlements in peri-urban areas 3131 Rural towns in peri-rural areas 3132 Rural towns in peri-urban areas Contents of this repository GRANULAR_Rural_Typology_Classifier.ipynb — the full classification pipeline (data loading, spatial join with the DEGURBA grid, feature engineering, K-Means clustering, GRANULAR code assignment, export) README.md — setup instructions, input/output schema, and runtime notes Input data: main grid cell attributes (population, building volume, hemeroby index, elevation, geometry), travel-time and town-count accessibility indicators, and the DEGURBA Level 2 classification grid Output data: GRANULAR_Rural_Typology_2026.parquet, a consolidated GeoParquet file (EPSG:3035) with GRD_ID, GRANULAR_code, GRANULAR_name, GRANULAR_short, DEGURBA_class, and geometry Coverage European territory, 1 km² grid resolution, ~4.4M grid cells. Data sources Population (Eurostat / GHS Population Grid, 2011 and 2021), building volume (GHS-BUILT-V, EU Joint Research Centre), land modification (hemeroby index from CORINE Land Cover), elevation (EU DEM), accessibility indicators (project-specific, imputed), and the DEGURBA Level 2 grid (2021 edition, EU Joint Research Centre).