A GeoAI-based approach for long-term monitoring of urban fabric transformations

Abstract

Urban growth has accelerated land use transformations, underscoring the need for accurate and scalable methods to monitor changes over time. This study presents a GeoAI-based methodology to classify and quantify urban fabric transformations over a 24-year period (2000–2024). The methodological framework is applied to the medium-sized municipality of Ravenna, in northern Italy, to evaluate its effectiveness. Landsat 5 and Landsat 9 multispectral images were classified into six Land Use and Land Cover categories using Random Forest (RF) and Support Vector Machine (SVM) algorithms within the Google Earth Engine platform. RF consistently outperformed SVM in both reference years, achieving overall accuracies of 83.8 % (2000) and 86.2 % (2024), with F1-scores exceeding 0.90 for key classes including built-up areas. McNemar’s test confirmed the statistical significance of RF’s performance advantage. Geospatial analysis revealed a 21.6 % increase in built-up surfaces (+7.8 km2) a 28.6 % increase in grassland/shrubland (+50.4 km2) and a 66.3 % reduction in bareland (−35.0 km2). Urban Density (UD) increased from 4.49 % to 5.73 %, indicating a moderate shift toward more compact urban growth. The results demonstrate the methodology’s reliability and transferability, particularly in data-scarce contexts, and provide actionable insights for evidence-based urban planning and sustainable land management.