Application of k-means clustering on flood episodes in small catchments in Croatia

This study presents an application of the k-means clustering method to the analysis of historical flood episodes in small catchments across the Republic of Croatia for the period 1998–2018. A total of 845 flood events with reliable and complete information were selected and described using three key parameters: flood duration, the size of affected area, and the portion of arable land within the flooded area. The results indicate that flood events in small catchments of the Danube River basin can be classified into five distinct clusters, while those in the Adriatic Sea basin are more uniform and can be grouped into two clusters. The identified clusters reveal significant differences in flood characteristics, allowing for the identification of flood types that require priority attention in flood risk management. The study demonstrates that multivariate clustering is a valuable tool for characterizing and regionalizing floods in small catchments, thereby supporting decision-making processes in flood risk management in Croatia.