Towards Automated Cadastral Map Improvement: A Clustering Approach for Error Pattern Recognition

Positional accuracy in cadastral data is fundamental for secure land tenure and efficient land administration. However, many land administration systems (LASs) experience difficulties to meet accuracy standards, particularly when data come from various sources or historical maps, leading to disruptions in land transactions. This study investigates the use of unsupervised clustering algorithms to identify and characterize systematic spatial error patterns in cadastral maps. We compare Fuzzy c-means (FCM), Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and Gaussian Mixture Models (GMMs) in clustering error vectors using two different case studies from Greece, each with different error origins. The analysis revealed distinctly different error structures: a systematic rotational pattern surrounding a central random-error zone in the first, versus localized gross errors alongside regions of different discrepancies in the second. Algorithm performance was context-dependent: GMMs excelled, providing the most interpretable partitioning of multiple error levels, including gross errors; DBSCAN succeeded at isolating the dominant systematic error from noise. However, FCM struggled to capture the complex spatial nature of errors in both cases. Through the automated identification of problematic regions with different error characteristics, the proposed approach provides actionable insights for targeted, cost-effective cadastral renewal. This aligns with fit-for-purpose land administration principles, supporting progressive improvements towards more reliable cadastral data and offering a novel methodology applicable to other LASs facing similar challenges.