Dynamic channel network extraction from satellite imagery and dynamics of the Jamuna River

Abstract

Braiding rivers consist of many channels, separated by bars which split at bifurcations and join in confluences. Much is known about these basic elements and their dynamics. However, the nature, direction and effects of propagation of local changes through the entire channel network has barely been studied and is not fully understood. Logically, the evolution of the river planform is the result of development of the individual elements. The link between the two has not yet been made. In this study, a braided river is considered as a network of bifurcations, confluences and channels. The evolution of the braided river is analysed by considering the river as a network and analyse the development of individual elements in this network.

A novel methodology was developed to extract detailed channel networks of the river form satellite images (not to be confused with hierarchical stream order networks). Both conventional river properties (i.e. braiding index) as properties derived from network analyses were used to analyse the river network evolution. Furthermore, a weighted braiding index is proposed as a measure for the dominance of a channel, this measure is based on the importance of channels which is derived from the network. Besides changes in network properties, the evolution of the channel pattern is analysed by means of the development of individual bifurcations. Bifurcations in the channel networks at two dates are linked by evaluating the similarity in bifurcation configuration. The bifurcation asymmetry is calculated for every bifurcation, the development of bifurcation asymmetry is calculated for every bifurcation linked between two river networks. The developed methods were designed to be automated and repeatable for the purpose of application to a series of images. These methods were applied to a time series of satellite images of the Jamuna river (Bangladesh) from 1999 - 2004.

The network measure `betweenness centrality' maps the importance of all elements in the network. These values of importance are calculated from the local channel geometry, but does not directly relate to the local geometry because the whole up- and downstream configuration of the network is taken into account. Other network measures were not directly useful for river network analysis. An increase of bifurcation asymmetry and a decrease of braiding and weighed braiding index are observed in periods of persistent low discharge and an increase of bifurcation asymmetry and weighed braiding index are observed during high discharge conditions. Due to the large differences in the appearance of the channel pattern during high discharge conditions, the development of bifurcation asymmetry appeared chaotic possibly due to limitation of the current methods used. During the dry season, parts of the river had only one dominant channel, in other words there was no active braiding.

Development of bifurcation asymmetry during persistent low water stages caused the formation of a dominant channel and a decrease of braidingness of the river. During high discharges, abandoned channel are reactivated and discharge is spread more evenly between channels, increasing the braidingness of the river.

The developed methods, especially the network analysis, shows a big potential for river network analysis. The current methods have a few limitations which could be avoided by tailoring the methods more specifically to rivers.