Wind Field Estimation by Disparate Data Harmonization and Clustering: A Case Study in Split-Dalmatia County

Wind speed and direction are key meteorological variables that vary across space and time, influenced by topography, atmospheric conditions, and land features. Accurate wind field estimation is vital for applications such as wildfire management, renewable energy, weather forecasting, and environmental monitoring. Traditional interpolation methods often fall short in capturing complex wind behavior due to sparse station distribution and dynamic variability. This study investigates the use of clustering techniques—K-Means, DBSCAN, and Gaussian Mixture Models (GMM)—to segment harmonized wind data into distinct regimes. Data from the Weather Underground network in Split-Dalmatia County, Croatia, was preprocessed and vectorized using wind component decomposition (u, v), then reduced via PCA for clustering. The quality of resulting clusters was evaluated using the Silhouette Score. Results indicate that, despite its simplicity, K-Means provides the most consistent separation of wind regimes, while DBSCAN struggles due to overlapping atmospheric patterns. This analysis offers insight into the suitability of clustering methods for capturing structured wind variability in complex terrain.