Spatiotemporal analysis and entropy-based sandbar disorderliness with LSTM forecasting of the Koshi River shift

ABSTRACT The Koshi River embankment breach has impacted over a million people living along its basin. Changes in channel patterns and the formation or erosion of sandbars can significantly impact flood risks, navigation, and surrounding settlements. Hence, the need to understand the river’s dynamic behavior and to predict the future is urgent to guide effective river management strategies. This study investigates the spatiotemporal evolution of a selected reach of the Koshi River using satellite imagery. A deep learning model is employed to forecast future river channel line trajectories. Additionally, entropy-based analysis is utilized to examine the temporal patterns of sandbar formation and erosion. The results reveal distinct movement patterns across the river’s segments. The upstream section, located above the barrage, experiences significant alterations in sandbar morphology. The midstream section shows considerable centerline fluctuations within pre-existing paleochannels. In contrast, the downstream reach exhibits relatively minor lateral shifts but frequent sandbar changes. These complex movement patterns are influenced by a combination of factors, including sediment transport, riverbed dynamics, and tectonic activity. This study enhances our understanding of the Koshi River’s evolving nature and contributes valuable information toward river erosion and its spatial and temporal shifts for sustainable river engineering.