Identification of homogeneous regions in terms of flood seasonality using a complex network approach

Catchment classification aids in the identification of homogeneous regions in which catchments have similar flood timing and corresponding climatic drivers. As a new classification scheme from complex network theory, the community detection method is introduced to classify 242 catchments in the United States based on flood seasonality. The robustness of this method is tested by calculating the Adjusted Rank Index between the classification results from different random subsets of the 242 catchments and different flood sampling methods. In addition, three network metrics (network density, centrality, and k-core nucleus) are used to further unravel the hydrological connections within each community of catchments. Catchments with similar flood seasonality cluster into six large communities, while catchments in sparsely gauged areas or with unique physiographic properties are isolated from large communities. High values of the Adjusted Rank Index show the robustness of the community detection method for catchment classification. The results indicate that the complex network is valid for use in classifying catchments based on flood seasonality. Moreover, for each community, the network metrics are potential descriptors of: (a) the degree of homogeneity, (b) the representativeness of any catchment, and (c) the spatial scale of synchronized floods. Therefore, the complex network is able to characterize the interconnections between catchments within the community.