Dynamic Channel Network Extraction from Satellite Imagery of the Jamuna River

Evolution of the largest rivers on Earth is poorly understood whiletheir response to global change is dramatic, such as severe drought andflooding problems. Rivers with high annual dynamics, like the Jamuna,allow us to study their response to changing conditions. Mostremote-sensing work so far focused only on pixel-based analysis ofchannels and change detection or manual digitisation of channels, whichis far from urgently needed quantifiers of pattern and pattern change.Using a series of Landsat TM images taken at irregular intervals showinginter- and intra-annual variation, we demonstrate that braided riverscan be represented as nearly chain-like directional networks. These canbe studied with novel methods gleaned from neurology. These networksprovide an integral spatial description of the network and should not beconfused with hierarchical hydrological stream network descriptionsdeveloped in the ’60s to describe drainage basins. The images werefirst classified into water, bare sediment and vegetation. Thecontiguous water body of the river was then selected and translated intoa network description with bifurcations and confluences at the nodes,and interconnecting channels. Along the entire river the well-knownbraiding indices were derived from the network. The channel width is acrucial attribute of the channel network as this allows the calculationof bifurcation asymmetry. The width was also used with channel length asweights to all the elements in the network in the calculation of moreadvanced measures for the nature and evolution of the channel network.The key step here is to describe river network evolution by identifyingthe same node in multiple subsequent images as well as new and abandonednodes, in order to distinguish migration of bifurcations from avulsionprocesses. Once identified through time, the changes in node positionand the changes in the connected channels can be quantified. Thesechanges can potentially be linked to channel migration and vegetationcover along the channels. A network evolves in time by adding orremoving channels and their bifurcation- and confluence couples. Usingthe network topology, we quantified network properties such as`centrality’, which provides a measure for the overall importanceof individual channels in a network. This is a novel and robustindicator to assess the effect of a change or engineering measure in achannel on the entire network. The physical basis for downstreampropagation of information through a fluvial network is the floodconveyance and sediment transport, and for upstream propagation it isthe backwater effect. Using the dynamic network description we can startquantifying the effects of local changes in the network on the entireupstream and downstream network. We conclude that the developed workflowallows the use of novel and useful measures borrowed from other sciencesin river network analysis, and provides, e.g., the assessment of theimportance of individual branches in a large complicated network.