Cluster analysis-based hydrological similarity assessment in small watersheds of Shandong Province's hilly areas

ABSTRACT The clustering of small watersheds based on hydrological similarity serves as an effective technique for identifying similarities in watershed runoff generation and routing conditions. It also addresses the challenge of parameter transplantation in undocumented areas. In this research, 545 small watersheds in hilly areas within Shandong Province were studied using 22 selected indicators to represent their climate and underlying surface characteristics. The study employed a two-stage clustering method combining the self-organizing map (SOM) neural network with the K-means algorithm, facilitating the classification of these watersheds into various groups. Each group of small watersheds was then analyzed for its unique characteristics. To validate the reasonableness of the classification results, the flood peak modulus of each watershed was calculated using a hydrologic-hydraulic method, while a parameter transplantation study was carried out and generalized for the clustering results. The findings indicate that the SOM-K-means clustering method efficiently classified the watersheds into 12 similar groups, validating its effective application in small watershed classification. This classification assists in solving the problem of flood forecasting in the ungauged watersheds in Shandong Province and developing more effective flood risk management strategies.