Comparative Analysis of Clustering Algorithms for Meteorological Regime Characterization Over Türkiye Using ERA5 Reanalysis Data

Abstract

Precise analysis and region-specific classification of meteorological parameters constitute a critical prerequisite for advanced scientific processes, ranging from artificial intelligence modeling to complex atmospheric research. The multidimensional and inherently volatile nature of these parameters often hinders the extraction of latent patterns through conventional statistical methods. At this juncture, clustering analysis facilitates the derivation of optimized analytical domains by automatically classifying data points with shared climatological signatures. This study adopts a comparative perspective by evaluating data-driven (K-means and K-medoids), spatial-based (Spatially Constrained Multivariate Clustering: SCMC), and density-based (DBSCAN) clustering approaches to reveal regional variability and spatial dependency across Türkiye. The primary motivation of this research is to determine which approach reflects the heterogeneous structure with high representational power across diverse topographical settings, thereby establishing an optimal clustering strategy for atmospheric modeling. This study aims to characterize and regionally decompose Türkiye’s meteorological profile by leveraging high-resolution European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis parameters, including temperature, pressure, relative humidity, instantaneous moisture flux, geopotential, 2m temperature and dewpoint temperature. Methodologically, the ERA5 were downscaled to the geographical coordinates of meteorological stations via point-based extraction and validated against in-situ observations provided by the Turkish State Meteorological Service. Performance analyses indicate that the SCMC approach achieved the highest precision in delineating meteorological similarities and disparities. The comparative success ranking was established as SCMC>DBSCAN>K-medoids>K-means. Findings demonstrate that while K-means and K-medoids are effective in identifying broad climatic zones, the SCMC algorithm generates significantly more consistent and spatially continuous clusters. Furthermore, DBSCAN successfully isolated microclimatic regions with extreme values, such as the Eastern Black Sea, from the primary clusters. Consequently, this research demonstrates that integrating high-resolution reanalysis data with advanced spatial clustering techniques provides a more precise and dynamic decision-support mechanism for regional meteorological analysis and environmental planning compared to conventional methodologies.