Recognizing Meanders to Reconstruct River Dynamics of the Ganges

The long-term discharge evolution of branches of rivers can potentially be derived from inactive and fossil river meanders using relations between discharge and channel width and meander dimensions. This evolution is relevant for the understanding of river dynamics and avulsion, and the resulting changes of downstream flooding risks and shipping conveyance. Unfortunately, digitalisation of infilled and overgrown meanders from images or maps is laborious and inaccurate. We developed and demonstrated an automated procedure for the recognition of river meanders from satellite-imagery of the river Ganges. For this automated recognition we used object-based image analysis, which allows including shape and neighbourhood information besides spectral information in the classification procedure. We needed information on both present, i.e. water carrying meanders, and fossil, i.e. partly vegetated meanders. This implies a wide range of spectral behaviour and hence the selection of meanders should be primarily based on shape. Therefore, we used an image showing the low-flow situation. Results demonstrate that the object-oriented image analysis approach offers detailed meandering metrics. The Gorai channel belt, once the main courses of the Ganges, has a fairly uniform meander size distribution and only a small number of old meanders, indicating a sudden emergence of the channel with limited dynamic meandering and a short period of activity. The Hooghly channel belt, another previous main course, has many more abandoned meanders and cut-off meanders including many small ones, suggesting a more gradual rise of discharge and a longer period of activity. The lack of small meanders in the present residual channels indicates that both were abandoned so suddenly that meander geometry could not adjust to the waning discharge. Given the importance of shape to recognize the meanders, this study was possible only thanks to the availability of object based image analysis. * Corresponding author.