Incorporating lumped catchment characteristics into complex networks for catchment classification

Abstract:

Hydrological processes are difficult to model and our limited understanding in the case of ungauged basins where observed data is not available makes it even more challenging. With most of the world's basins being ungauged, the traditional solution was to transfer information from gauged catchments to hydrologically similar ungauged catchments, known as regionalisation. However, the success of these methods usually require the prior knowledge of the number and separability of hydrologically similar catchment groups, into which the ungauged catchments are appended. To overcome these limitations, recent studies have proposed complex networks based community structure algorithms for catchment classification. However, these studies can be applied only when time series of hydrological variable is available (eg. streamflow) and limits their usage upon availability of lumped catchment characteristics We propose a Canberra distance-based link metric which uses lumped catchment information including mean rainfall, mean relative humidity, mean potential evapotranspiration, mean air temperature, stream density, standard deviation of elevation and mean watershed slope for classifying catchments. Multilevel Modularity Optimization (MMO) algorithm was executed on the thresholded link weights using lumped catchment data from 494 basins situated in the CONtiguous United States (CONUS) region, and its performance was compared with K-means clustering algorithm.This method results in apriori similar catchment groups in contrast to the requirements of supervised classification methods. By and large, the proposed classification framework opens up an alternative avenue towards prediction in ungauged basins.