Complex networks, community structure, and catchment classification in a large-scale river basin

This study introduces the concepts of complex networks, especially community structure, to classify catchments in large-scale river basins. The Mississippi River basin (MRB) is considered as a representative large-scale basin, and daily streamflow from a network of 1663 stations are analyzed. Six community structure methods are employed: edge betweenness, greedy algorithm, multilevel modularity optimization, leading eigenvector, label propagation, and wallctrap. The influence of correlation threshold (i.e. spatial correlation in flow between stations) on classification (i.e. community formation) is examined. The consistency among the methods in classifying catchments is assessed, using a normalized mutual information (NMI) index. An attempt is also made to explain the community formation in terms of river network/branching and some important catchment/flow properties. The results indicate that the correlation threshold has a notable influence on the number and size of communities identified and that there is a high level of consistency in the performance among the methods (except for the leading eigenvector method at lower thresholds). The results also reveal that only a few communities combine to represent a majority of the catchments, with the 10 largest communities (roughly 4% of the total number of communities) representing almost two-thirds of the catchments. Community formation is found to be influenced not only by geographic proximity but also, more importantly, by the organization of the river network (i.e. main stem and subsequent branching). Some communities are found to exhibit a greater variability in catchment/flow properties within themselves when compared to that of the whole network, thus indicating that such characteristics are unlikely to be a significant influence on community grouping. (C) 2016 Elsevier B.V. All rights reserved.