A Canberra distance-based complex network classification framework using lumped catchment characteristics

Abstract

Hydrological prediction in ungauged catchments remains a challenge despite numerous attempts in the past. The well-known solution to this challenge is transfer of information from gauged catchments to ‘hydrologically-similar’ ungauged catchments, an approach known as ‘regionalization.’ The basis of regionalization is, thus, classification of catchments into hydrologically-similar groups. A major limitation of the traditional classification methods, such as the K-means clustering algorithm, is that they are not very suitable when the classes are not well separated from each other. Additionally, they cannot determine the number of classes in a dataset automatically. To overcome these limitations, some recent studies have used complex networks-based classification algorithms, widely known as community structure algorithms, for catchment classification. However, such studies have applied the community structure algorithms only to time series of hydrological variables (e.g. streamflow) and have not so far used lumped information (e.g. mean rainfall and mean slope). In this short communication, we propose a Canberra distance-based metric that can enable a community structure algorithm to exploit lumped information. For demonstration, the proposed metric is used to compute link weights for the multilevel modularity optimization algorithm. The proposed classification method is applied to lumped data from 494 basins situated in the CONtiguous United States (CONUS) for their classification, and its performance is compared with that of the K-means clustering algorithm. By and large, the proposed classification framework opens up an alternative avenue towards prediction in ungauged catchments.