Studying eco-geomorphological changes in a braided river : A case study of Koshi River

Braided rivers are morphologically complex and challenging to study. They flow in multiple channels covering a wide span. While they flow in multiple channels forming a large floodplain, they give life to a large variety of flora and fauna. This forms a complex riparian ecosystem, which is highly dependent on the morphological behaviour of the river. Moreover, river habitats and species are affected by the changes in morphology and ecology of a braided river. Such interaction between the river morphology and ecology may further be complicated and affected by various anthropogenic interventions such as embankments, bridges, dams, barrages, dredging, etc. This reveals necessity of studying the interrelation between geomorphology and ecology, referred to as eco-geomorphology, considering river interventions and their impacts on aquatic environment and habitats. In this research, eco-geomorphological changes of braided reach of the Koshi River, which emerges from mountains and flows through lowlands of Nepal and India, is studied using freely available satellite images and cloud computing platform - Google Earth Engine (GEE). An attempt is made to study the impacts of human interventions in the river on river morphology and vegetation dynamics, and in turn, on habitat of diverse faunal species in the Koshi Tappu Wildlife Reserve (KTWR), located within the study reach. This interlinking between river morphology, habitat and ecology is one of the novel aspects of this study. The study reach is about 123 km long starting from Chatara of Nepal to Mahisi in India. Within the study area, the structural interventions in the river are the Koshi Barrage, embankments and the Koshi Bridge (Koshi Mahasetu). The annual morphological changes of the river are studied using an automated image processing tool, Morphology Monitor (MoMo), built on the GEE platform. All the freely available satellite images in GEE from Landsat and Sentinel series for each year (during 1987 to 2020) were accessed and NDWI of each image was calculated. The images were stacked based on weighted average (based on goodness of pixel) to form a single composite image. Composite images of were used to calculate the area of erosion and accretion in each year interval by subtracting image of one year from the subsequent year and calculating the change in the value of NDWI of each pixel. The positive changes were recorded as erosion and the negative changes as accretion. Only a few such studies have been carried out so far, and most of them quantify only the maximum changes, i.e. the difference between two dry periods. While, in this study, the changes within the area between any selected ranges of water occurrence can be quantified. In this way, the changes in low-amplitude sandbars can also be quantified separately. This is a novelty, although there are uncertainties and limitations of the method. The sources of such uncertainties may be the noise in the composite image as well as the noticeable difference in water levels between two compared years given that one year is wet while another is dry. The analysis of annual erosion and accretion was performed for whole study reach as well as for some selected critical reaches. The analysis showed that the whole study reach was accreting more frequently despite the large erosion occurring occasionally. The upstream reach of the Koshi Barrage was found to be accreting significantly in comparison to the downstream. This could be attributed to the barrage that blocks the flow (reducing the velocity due to backwater effect) and sediment transport, as well as to the embankments which confine the flow and sediment causing the increased sedimentation. The observation of the planform showed that lateral movement of the deep-channels is highly governed by the operation of the barrage. The operation of the barrage has increased the morphological activities both in the upstream and downstream. The upstream reach has been significantly narrowed during 1987-2020 due to large deposition and abandoned channels in western half of the river. This resulted in increasing deep-channel dynamics in eastern half leading to the breach of embankment in 2008. The deep-channel dynamics of the river in the upstream has increased further since 2008 caused by the operation of the barrage (and may be other interventions) to move the deep channel away from the eastern embankment. Similarly, the downstream experienced channelization of the flow along the eastern or western embankment. In a reach further downstream, the construction of the Koshi Bridge in 2011 constricted the flow area and further channelized the river along the embankments in the downstream increasing the vulnerability of the embankments to breach. To assess the interaction between morphology and ecology, also a study of vegetation dynamics was carried out using satellite images. The composite images are also useful in monitoring the vegetation cover of the area. The NDVI calculated for each composite image represented average vegetation health round the year in each pixels. The pixels with high to low values of NDVI are identified as perennial vegetation to more seasonal vegetation with the lowest NDVI (0 and below) interpreted as water. This approach was useful in monitoring the growth of vegetation in the newly formed islands (accreted area) which later evolved as stable islands due to growth of perennial vegetation (increasing trend of NDVI through 1987-2020) leading to more stable channels around them. The growth of vegetation was also observed along the embankments in the areas protected by the spurs. ML algorithms, Random Forest (RF) and Support Vector Machine (SVM) were used for supervised classification to identify and quantify the LULC in four classes as water, grassland, forest and barren land within the KTWR. The classifier was build using the training data from composite image of one year (2018/19) and used to classify the images of other nine years (for which population of Bubalus arnee (Arna) were available so that it could be used in ecological analysis). The RF showed higher accuracy of classification (more than 85%) for most of the years whereas the SVM showed highest accuracy of 83%. The RF was used to generate the LULC maps of the KTWR. RF very accurately identified the water and barren land but distinguishing between forest and grassland was a challenge. This may be due to inaccurate training samples that were prepared from visual interpretation of the composite image with a spatial resolution of 30 m, whereby the forests were present in very small patches (even up to size of a pixel) in the area. The inaccuracy may also be due to the tall grasses or bushes that are present in the flood plain and wetlands which are interpreted as trees. The percentage of area covered with water showed consistent trend over the year. The largest area was occupied with grassland. This would be helpful in identifying the habitats for faunal species. An attempt was made to assess the impact of changes in morphology and vegetation cover on KTWR habitats. Various literatures were studied for identifying the key species of KTWR and collecting their population and behavioural data. Population data for Arna was available and only the habitat behaviour was obtained for other key species (Swamp francolin, Ganges Dolphin and Marsh mugger crocodile). The analysis of the vegetation and morphology was used for determining the impact on wildlife of KTWR. The average NDVI showed a positive correlation of 0.66 with the population of the Arna. The population of Arna showed an increasing trend with increasing area of grassland (obtained from supervised classification). However, the trend was negative for barren land. The habitat analysis for the Swamp francolin showed their possibility of occurrence near the swampy area with tall grasses and sparse trees and that of Crocodiles was near the sandbanks. This showed the possible habitat estimation using the LULC images. Similarly, the Dolphins liked dwelling in deep channels and near eddies of confluence and bends. Hence it can be inferred that the probability of finding Dolphins when the river is flowing in shallow area with multiple channels is low. In conclusion, the study demonstrated the potential of advanced satellite image processing technique to address eco-geomorphological processes and their interactions and impacts. Furthermore, the study considered the human interventions that have made significant impact on the morphology and thus on the riparian ecology (vegetation and wildlife) of the Koshi River. In this regard, the MoMo seems to be a promising tool for monitoring and assessing the morphological changes of the river. Moreover, it was found to be adaptable to the study area and was able to address the effect of seasonal variability in river morphology to some extent. The tool was further expanded to detect the vegetation dynamics and to identify the correlation of morphological changes of Koshi River to the riparian ecology. The further development of this tool into an application would be an aided advantage in the field of river, floodplain and wildlife management