Introduction

Riverbanks constantly change position in response to natural or anthropogenic influences (Lane and Richards 1997; Surian 1999; Li et al. 2007; Kummu et al. 2008). The rates and nature of channel change vary greatly both in space and time (Hooke 1980; Lawler et al. 1999) depending on numerous factors, including tectonic activity; the compositional, textural, and geotechnical properties of the bank materials; variations in the river stage; the discharge and flow regime; the channel configuration; the presence of bank vegetation; and anthropogenic activities (Simon and Collison 2002; Julian and Torres 2006; Bartley et al. 2008; Michalková et al. 2011; Janes et al. 2017). The phenomenon of channel migration is critical to several geomorphological and river management problems (Milton et al. 1995; Petts 1995). The dynamics of fluvial sediment transport, basin sediment yields, and the destruction and construction of floodplains or, in a holistic view, the evolution of fluvial channels, are largely determined by bank erosion and accretion (Mukherjee et al. 2016). Erosion of prevailing riparian, agricultural, rural, and urban lands causes subsequent migration of people through displacement and the destruction of hydraulic structures and transport systems (Lawler 1991; Larsen et al. 2006; Mukherjee et al. 2016). On the other hand, bank accretion leads to new alluvial land, which may then be used for agricultural and pastoral purposes.

The Ganges-Brahmaputra-Meghna (GBM) river system, carrying gluts of water and sediment (more than 1 billion tons annually), frequently changes its planform, leading to the development and displacement of sandbars, channel abandonment, and riverbank accretion and erosion (Coleman 1969; Biswas 2015). It has been estimated that the bankline shifting of the three major rivers (the Jamuna, Padma, and Lower Meghna) in Bangladesh during the last three decades of the twentieth century together has caused the destruction of a net 100,000 ha of floodplain land, leaving 800,000 people landless (Sarker 2009). Moreover, riverbank erosion in Bangladesh directly or indirectly causes suffering to around 1 million people annually (Elahi and Rogge 1990; Sarker 2009).

The Lower Meghna River (LMR), the downstream stretch of the GBM river system, is highly dynamic and experiences pronounced morphological changes. The seasonal abundance of water and remarkably fertile soils in the floodplain attract people to live along its banks and engage in intense agricultural activity. A large number of settlements, towns, ports, and industries have been built upon both banks. Hence, the LMR contributes greatly to the economy of Bangladesh, providing water for agriculture, fishing, navigation, and environmental sustainability. However, the long-term sustainability and economic development of its floodplain are threatened by bank instability. This calls for an extensive spatiotemporal investigation of the bank dynamics of the LMR.

Several studies, so far, had been conducted on the bank dynamics of major rivers in Bangladesh, but most of them have focused on the Ganges and Brahmaputra Rivers (e.g., Hossain 1991; Kalam and Hoque 1995; Rahman and Ashrafuzzaman 2004; Sarker 2004; Yeasmin et al. 2007; Thakur et al. 2012; Hossain et al. 2013; Dewan et al. 2017). Previous studies incorporating the LMR have investigated sediment transport and mineralogy (Barua 1990; Ahmed and Louters 1997; Datta and Subramanian 1997), its morphological response to seismic events (Sarker and Thorne 2006; Sarker 2009), bank protection measures (de Jong 2000), and erosion hazards (Sarker 2009; Hassan et al. 2017). Nonetheless, none of these studies intensively covered the LMR, and none used any particular models to calculate bank change rates nor determined the likely reasons for large changes in bank location.

Many methods and models have been adopted worldwide for analyzing riverbank dynamics, particularly in assessing riverbank erosion hazards (Graft 1984), bank shifting (Gilvear and Winterbottom 1992), bank erosion (Rutherfurd 2000; Prosser et al. 2001), the Photo Electronic Erosion Pin (PEEP) (Lawler 1991; Lawler and Leeks 1992), bank material strength (Thorne 1981), the Bank Erosion Hazard Index (BEHI) (Rosgen 1999), and near-bank stress (NBS) (Nieber et al. 2008). However, most of these methods required extensive fieldwork and highly advanced technology that made the analysis costly, time-consuming, and eventually very difficult to perform. Traditional ground survey techniques are not often feasible for a large area and offer no past history information such as previous bankline positions.

The recent developments in remote sensing (RS) and geographic information systems (GIS) provide an easy, economic, and time-saving platform for comprehensive spatiotemporal analysis of bank erosion and accretion (Winterbottom 2000; Mount et al. 2003; Mukherjee et al. 2016). GIS-based studies have been performed to evaluate the mechanism of riverbank shifting (e.g., Klaassen and Vermeer 1998; Gilvear et al. 1999; Winterbottom and Gilvear 2000; Bhakal et al. 2005; Das and Saraf 2007; Kummu et al. 2008; Janes et al. 2017). Utilizing Landsat imagery to calculate the mean annual bank shifting as well as erosion/accretion status of the Ganges-Padma river system (Dewan et al. 2017) and evaluating the morphological changes of the Jamuna-Padma-Lower Meghna river system in response to the disruption caused by the Assam earthquake in 1950 (Sarker 2009) are examples of such an approach in this region. Quantifying channel migration to account for morphometric changes in the Ganga River using manually drawn cross sections (Thakur et al. 2012) and analyzing the erosion and accretion dynamics of the Lower Ramganga River in Uttar Pradesh, India, by geospatial techniques (Mukherjee et al. 2016) are further examples. However, a statistically based automated calculation of riverbank change rates using multi-temporal satellite images in the GIS interface and the consequent zonation (into erosion and accretion) of the banks have not yet been carried out. Although Bandyopadhyay et al. (2014) did prepare a RS/GIS-based bank erosion vulnerability zonation map of the Haora River in Tripura, India, their method neither considered accretion nor generated erosion/accretion rates.

The Digital Shoreline Analysis System (DSAS), an extension to the ArcGIS software, has been successfully used for shoreline change assessment (Oyedotun 2014; Thi et al. 2014; Bheeroo et al. 2016; Stanchev et al. 2017) and cliff geometry modelling (Rio and Gracia 2009; Oyedotun 2014; Young et al. 2014) from multiple historic shoreline positions. However, the automated measurement of riverbank shifts along the user-defined transects for a number of statistical parameters could also be performed by this technique (Thieler et al. 2009), which facilitates riverbank zonation. The DSAS has therefore been used in this study to investigate the changes in bank position of the LMR between 1988 and 2017. This is a unique attempt to quantify bank shifting of a large river like the LMR using the DSAS. The overall aim of this study was to evaluate the bank dynamics of the LMR in Bangladesh over the study period, with specific objectives: (1) to calculate the erosion and accretion rates along the banks of the LMR, (2) to quantify changes in average river width and the braided index (BI), and (3) to investigate the factors responsible for the channel instability.

Study area

The LMR, the downstream part of the Meghna-Surma river system, originates in the Naga Hills of Manipur, India. Along its 946 km course, the Meghna-Surma River crosses the Manipur, Mizoram, and Assam states of India and lastly Bangladesh (containing the largest part, 669 km) before reaching the Bay of Bengal. The Meghna River from the confluence of the Surma and Kushiyara Rivers, just north of Bhairab Bazar, to Chandupura is hydrographically referred to as the Upper Meghna; when the Padma River joins it at Chandpur, it becomes the LMR.

This study covers 90 km stretch of the LMR from Chandpur to the top of Hatia Island (Fig. 1). The eastern boundary follows the left bank down to the coastline near Noakhali Sadar, whereas the western boundary follows the right bank down to Lalmohan, Bhola. The study area lies between longitudes 90°25′E and 91°10′E and latitudes 22°20′N and 23°18′N. Along its course to the Bay of Bengal, the LMR splits into a number of channels such as the Tetulia, Shahbazpur, and Hatia, while the main flow is through the Meghna Estuary.

Fig. 1
figure 1

Location map of the study area

Full size image

The LMR cuts across the Bengal Foredeep, a broad tectonic unit of the Bengal Basin (Fig. 2a). The basin has come into existence because of the subduction of the Indian plate beneath the Eurasian plate (Uddin and Lundberg 2004; Bilham 2004; Farazi et al. 2018). Bangladesh, which contains the major part of the Bengal Basin, lies at the juncture of three convergent plate boundaries, the Indian plate to the west and southwest, the Tibet sub-plate to the north, and the Burma sliver to the east. Moreover, the India-Burma plate boundary passes through Bangladesh—the eastern part belongs to the Burma sub-plate while the western part belongs to the Indian plate (Akhter 2010; Akhter et al. 2016). This conjunction of plates gave rise to the Himalayan Arc to the north and the Burma Arc to the east (Steckler et al. 2008; Akhter 2010). The sediment and water discharges from these two arcs through the GBM rivers jointly formed the vast GBM Delta. The India and Burma plates converging during the interseismic period brought changes to the landscape of the delta, particularly along the deformation front (Akhter et al. 2016) (Fig. 2b). During that time, the Indian plate dragged down the tip of the locked upper Burma plate triggering the westward tilting of the accretionary prism (i.e., the terrain east of the LMR). The deformation front is marked by the low elevation area of the LMR to the south and Sunamganj-Kishoreganj marshes to the north.

Fig. 2
figure 2

(a) Geodynamic setup of Bangladesh and surroundings showing the major tectonic elements and potential source of earthquakes, modified from Akhter et al. (2015). (b) Section through the Indo-Burma plate boundary, modified from Akhter and Steckler (2019). (c) Geological map of the study area modified from the Geological Map of Bangladesh published by the Geological Survey of Bangladesh

Full size image

The LMR exhibits high seasonal variations, with an average annual water discharge rate of around 34,600 m3/s, a mean annual flood discharge rate of about 97,000 m3/s (Sarker 2009), and a tidal range of up to 4 m (Barua 1990; Allison et al. 2003). The floodplain is characterized by low elevation, less than 5 m above mean sea level (Wilson and Goodbred jr. 2015), and a very gentle slope, i.e., 5 × 10−3 (Sarker 2009). Both banks contain Holocene alluvial, deltaic, and tidal deltaic sediments derived from the Himalaya and Burma Arcs (Fig. 2c).

The study area has a tropical climate, governed by the southwest monsoon. The average annual rainfall in the LMR system is about 1600 mm, most of which falls during the monsoon (Sarker et al. 2003). This region is known to be immensely prone to various coastal and estuarine hazards such as floods, tropical cyclones, storm surges, salinity, and riverbank erosion. Though the exact data on population displacement because of riverbank erosion along the LMR are not available, the number would be, by any measure, very high, and it is likely to accelerate in the coming years as augmented rainfall due to climate change enhances discharge and runoff to the river.

Materials and methods

Data used

This study used multi-temporal Landsat satellite images, each covering the entire study area, together with the relevant hydrological information to examine the channel dynamics of the LMR. Seven images, covering the years 1988, 1993, 1998, 2003, 2008, 2013, and 2017 at approximately 5-year intervals (Table 1), were obtained from GLOVIS (www.glovis.gov.us) depending on scene availability. The first three (1988–1998) were from the Landsat Thematic Mapper (TM) sensor, and the remaining four (2003–2017) from the Landsat Enhanced Thematic Mapper Plus (ETM+).

Table 1 Satellite data acquisition and corresponding water level information. Water level data acquired according to chart datum (CD) used by the Bangladesh Inland Water Transport Authority (BIWTA)
Full size table

The LMR is influenced by diurnal tides, with low water levels during dry periods, exposing bars and islands. The tidal effect is small in the dry season, so dry season (January and February) satellite images were used to minimize tidal effects. This approach hence reduced the effect of discrepancies in river configuration while extracting data from the satellite images (Dewan et al. 2017). To analyze tidal influence on the study area, although only small fluctuations were observed, data were collected from four tidal gauge stations of the Bangladesh Inland Water Transport Authority (BIWTA) (Fig. 1; Table 1).

Data preparation

To achieve spatial accuracy, all the satellite images were georeferenced to the Universal Transverse Mercator (UTM) system. The most recent Landsat image (2017) was used as the parent image for all other image registration. A number of well-distributed ground control points (GCP) were identified on the parent image: these were permanent features such as road intersections and landmarks. They were then used in conjunction with first-order polynomial transformations and nearest neighbor resampling techniques to transform the other images to yield a root mean square error (RMSE) of below 0.5 pixels. All the images were then subject to further modifications such as histogram equalization and haze and noise reduction for enhanced visibility of the image features. The geometric correction and radiometric enhancement techniques were completed with the help of Erdas Imagine (version 14) software. Finally, each image was masked using an area boundary polygon shapefile representing the extent of the study area.

Because of their non-cohesive materials, mostly clay-to-silt-size particles with little sand, the banks of the LMR have hardly any resistance to eroding forces. Hence, exact demarcation of their position is somewhat difficult and time-consuming. Difficulties also arise owing to the presence of attached or partially attached islands/bars in the channel and to periodic position changes in sub-channels; for these reasons, the criteria set out by Hasan et al. (1997) were kept in mind during bankline demarcation. We relied on a semiautomated approach for the bankline extraction, as several studies have found it to be superior to automated ones (Yang et al. 1999; Gupta et al. 2013; Dewan et al. 2017). Firstly, land and water were separated from the respective images by  normalized difference vegetation index (NDVI) , which is considered to be a good separator of waterbodies from land features (Sun et al. 2012). An onscreen digitization was then carried out, with a fixed representative factor (RF) scale (1:25,000) used to maintain consistency during digitization. The banklines were digitized from the satellite images as vector file (polylines) in ArcGIS software (version 10.2), using the water boundary to indicate the edge of the channel (Nicoll and Hickin 2010). A geo-database containing the banklines for all the years was created and stored for later use on the DSAS.

Analytical techniques

The majority of the analytical techniques adopted for this study were executed on the DSAS interface in a GIS environment. We used ArcGIS (version 10.2) to run the DSAS extension tool, which was developed conjointly by the United States Geological Survey (USGS) and the TPMC Environmental Services (Thieler et al. 2009). Three parameters, namely, bankline change, river width, and the BI, were measured in order to assess the bank dynamics of the LMR.

At the first step, we utilized the DSAS technique that calculates riverbank shifting automatically along the user-defined transects across the banklines. Prior to that, a baseline was drawn parallel to the general orientation of the bankline of the LMR. This baseline was drawn approximately 500 m landward so that it lay in land in which no erosion was evident. A total of 430 transects (225 and 205 transects for the right and left banks, respectively) was generated automatically perpendicular to the baseline at 500 m intervals, assumed to be small enough to quantify the riverbank changes of such a large river. The intersections of the transects and the historic shoreline positions were then used to generate the rate of change statistics. Five statistical measures from the DSAS were calculated: the end point rate (EPR), the linear regression rate (LRR), the weighted linear regression rate (WLR), the shoreline change envelope (SCE), and the net shoreline movement (NSM). The EPR, LRR, and WLR provide bankline change rates in meters per year; the other measures the bankline changes in meters.

The EPR is calculated by dividing the distance of the bankline movement along each transect by the total time period elapsed between the oldest and youngest banklines. The LLR is calculated by fitting a least squares regression line to all the intersection points of the bankline along a particular transect. The WLR method is similar to the LRR but gives reliable data a higher weight. The SCE is the distance between the farthest and the closest banklines to the baseline at each transect; the NSM is the distance between the oldest and youngest bankline positions. The standard error, coefficient of determination R, and confidence intervals for the LRR and WLR were also calculated. In this analysis, R values for the LRR and WLR with a 95% confidence interval were used to judge the accuracy of the prediction. Historical bankline positions were also analyzed visually in order to determine any inconsistencies in the DSAS results. A 1-pixel (30 m) uncertainty was assumed for the rate of bankline change. A landward shift of the bankline with respect to the baseline was considered as erosion, denoted negative, and a riverward shift as accretion, denoted positive (Anders and Byrnes 1991).

Each bank of the LMR was divided into several erosion/accretion zones, based on the bankline change statistics. These zones were visualized on a color grid map, based on the LRR values, using the natural break method in the ArcGIS environment. Only the LRR values were used for mapping because the bankline change rates obtained from the EPR, LRR, and WLR were found to be very similar (Table 2); the LRR uses all the available data and is a purely computational method based on accepted statistical concepts (Dolan et al. 1991; Crowell et al. 1997).

Table 2 Pearson correlation matrix for statistical parameters depicting bankline change rates in the LMR
Full size table

At the second step, the river width, the shortest distance between the right and left banks ignoring mid-channels and lateral bars (Sarker 2009), at a given location was measured. In determining the width of the LMR, crosslines at 500 m intervals were drawn perpendicular to the channel (Winterbottom 2000; Mount et al. 2003). Considering the objectives of the study, 500 m intervals were thought to be sufficient to measure the width of such a long reach. The width of the river was determined along each of these crosslines using the ArcGIS measurement tool. The arithmetic mean of these channel widths was then calculated (Winterbottom 2000; Mount et al. 2003). The determination of the river width using satellite images involves errors related to positional inaccuracies and spatial distortions in each image—residual error (Mount et al. 2003; Mount and Louis 2008). However, the rates of channel change are valid if the amount of change significantly exceeds the errors involved in the measurement process (Downward et al. 1994; Tiegs and Pohl 2005; Yao et al. 2011). The LMR experienced a high degree of channel change, significantly exceeding the total absolute error.

At the third step, we calculated the BI, which characterizes the changes in river planform morphology. Various approaches have been developed to measure the BI of a river (Brice 1964; Howard et al. 1970; Friend and Sinha 1993; Germanoski and Schumm 1993). Of these techniques, the two simplest indices, proposed by Brice (1964) and by Germanoski and Schumm (1993), respectively, were adopted. According to Brice (1964), the BI is twice the sum of the lengths of all the islands and/or bars Li in a reach divided by the length Lr of that reach, measured mid-channel:

$$ \mathrm{BI}=2\Sigma\ Li/ Lr $$
(1)

The BIs are sensitive to the river stage, and reach lengths of a minimum of ten times the average wetted width are needed to measure the BIs with a precision of about 20% of the mean (Brice 1964; Egozi and Ashmore 2008). Here, to maximize the accuracy, dry period satellite images were used, and the whole LMR was taken as a single reach. While Brice (1964) only considered the length of bars, later Germanoski and Schumm (1993) modified the equation by incorporating the total number of bars Ni per reach length Lr. This subtle modification decreased the possibility that a reach could have an equivalent or even greater BI with one large bar than with several small bars (Germanoski and Schumm 1993):

$$ \mathrm{BI}=2\Sigma\ Li/ Lr+ Ni/ Lr $$
(2)

The results obtained from the two methods were compared to observe whether there was a difference due to the existence of single large bars.

Field investigation

An extensive field investigation was carried out once the zonation map of the LMR had been created. This was intended to verify the bank change rates in these zones and also to try to identify the reasons for the bank instability. A social survey was conducted in households in selected zones along the riverbank, applying a purposive random sampling approach; residents were interviewed using a semi-structured questionnaire. As the temporal range of this study goes back to 1988, only people who had lived at least 40 years in that locality were selected for the survey. Five people in each zone were interviewed. They were asked about the status of the riverbank (whether there had been erosion and/or accretion), as well as its rate of change. People were also asked to point out the position of the bankline during the 1988 flood. Based on this information, we tried to identify the rate of bankline change over the last 30 years. The prime intention was to cross-check their actual experience of bankline change with the DSAS-based estimates. Bank failure types along the LMR were also investigated; these indicated variations in the geological and engineering properties of the bank materials.

Results

Right bank change rates

Figure 3 shows the distributions of the EPR, WLR, and LRR along transects across the right bank of the LMR. This explicitly demonstrates the high variability in bankline movement along the right bank and the strong dominance of erosion. Based on the fluctuations in bankline changes, the right bank was divided into eight distinct zones (Fig. 3); this zonation was verified by comparing it with the NSM and SCE derived statistics (Fig. 4). The NSM showed a similar trend to the EPR, WLR, and LRR, but the SCE displayed the opposite trend. The similar alignment of lines drawn from the NSM and SCE indicated accretion, and the gap between them erosion. Table 3 gives a summary of the bankline change statistics for the right bank zones. From the average bankline change rates, each of the eight zones was classified into one of six categories: low erosion (0 m/yr to −50 m/yr), moderate erosion (−51 m/yr to −100 m/yr), and high erosion (< −100 m/yr) and low accretion (+1 m/yr to +50 m/yr), moderate accretion (+51 m/yr to +100 m/yr), and high accretion (> +100 m/yr).

Fig. 3
figure 3

Bank change rates in EPR, WLR, and LRR along the right bank of the LMR

Full size image
Fig. 4
figure 4

Bank changes in NSM and SCE along the right bank of the LMR

Full size image
Table 3 Zone-wise statistics for the right bank of the LMR. The EPR, WLR, and LRR provide bank change rates in m/yr; the NSM and SCE provide the bank changes in m
Full size table

Six of the eight zones in the right bank were erosion dominant (Table 3). Among them, Zones R3 and R7 were classified as high erosion zones, R4 and R5 as moderate erosion zones, and R1 and R8 as low erosion zones. R2 and R6 were categorized as moderate accretion zones. Overall, by averaging the bank change rates (both positive and negative values) along all transects, the right bank of the LMR was found to be eroded at a mean rate of 81 m/yr during the study period (Table 4).

Table 4 Average bank change rates in the LMR for the last 30 years
Full size table

All zones on the right bank, except R1, R2, and R6, showed high R2 values, from 0.78 to 0.94, indicating good bankline change rate prediction (Table 3). The relatively low R2 values for R1 (0.26), R2 (0.43), and R6 (0.54) could be accounted for the high dynamic nature of these zones, fluctuating between erosion and accretion. For example, according to the historical analysis of bankline position changes, R1 gained a huge amount of land between 1988 and 1993. Erosion then began, and continued till 1998, followed by 5 further years of accretion. Erosion started again in 2003 and was sustained until the end of the study period. Similarly, R2 underwent both erosion and accretion stages. R6 consistently accumulated land until 2013 but underwent slight erosion between 2013 and 2017.

The field investigation and social survey results in Zones R2, R3, R5, and R6 confirmed the bankline change rates obtained from the DSAS. Interviews with local people found that R2 was, in general, undergoing erosion until 2008. It then gained a large amount of land within a very short span of time (until 2017). Moreover, the participants reported that the area is currently accreting land at a rate of nearly 100 m/yr. Locals in R3 stated that the area had been eroding at a rate of more than 100 m/yr. Similarly, the riverbank in R5 was reported to be eroding at a rate between 50 and 100 m/yr; it was also pointed out that the erosion rate was more than 100 m/yr in some places about 4 years ago. Recently, bank protection measures were carried out in many places in R6, which reduced the bank erosion. However, the failure of some of these structures was also observed in this zone. People were of the opinion that R6 started accreting around 20 years ago at a rate of 50–100 m/yr, but the area underwent slight erosion between 2013 and 2017.

Left bank change rates

The left bank of the LMR was divided into seven erosion/accretion zones (Fig. 5). As for the right bank, a line diagram based on the NSM and SCE values was used to verify the reliability of zonation based on the EPR, WLR, and LRR (Fig. 6). The zone statistics for the left bank are provided in Table 5. Similar to the righ bank, each zone in the left bank was classified into one of the six categories, based on bankline change rates.

Fig. 5
figure 5

Bank change rates in EPR, WLR, and LRR along the left bank of the LMR

Full size image
Fig. 6
figure 6

Bank changes in NSM and SCE along the left bank of the LMR

Full size image
Table 5 Zone-wise statistics for the left bank of the LMR. The EPR, WLR, and LRR provide bank change rates in m/yr; the NSM and SCE provide the bank changes in m
Full size table

Four of the seven left bank zones were erosion dominant: L5 was a high erosion zone, L2 a low erosion zone, and the other two (L3 and L6) moderate erosion zones (Table 5). In contrast, L1, L4, and L7 were accretion dominant (Table 5), with L1 displaying an overall stability, and L4 and L7 significant accretion. As a whole, the average bankline change rate in the left bank, taking all transects into account, was −28 m/yr (Table 4).

The coefficient of determination values from the LRR and WLR for the left bank were fairly high (greater than 0.5), except for L1, L2, and L7 (Table 5). Again, the analysis of historical bankline position changes revealed that these three zones switched between erosion and accretion several times during the study period; this may again explain the low R2 values for these zones.

Field investigations and social surveys in selected zones (L1, L2, and L3) of the left bank also showed very good agreement with the results from the DSAS. In most of L1, protection measures with concrete blocks, installed in the 1980s, stabilized the riverbank. According to the local people, erosion rates at L2 had been between 50 and 100 m/yr before the bank protection measures. Similarly, the riverbank in L3 was reported to be eroding at a rate of more than 150 m/y, but it too had recently been protected with concrete blocks.

For both the right and left banks, there were some differences between the bankline change rates obtained from the DSAS and from the social survey. In some cases, assumptions made by the local residents were slightly higher due to the fading of 30-years-old memory with regard to bankline positions. Furthermore, the negative effects of bank erosion on their life also could have influenced them to over-report the bankline change rates.

A color grid map of all zones from the right and left banks of the LMR provides a vulnerability zonation map of the study area (Fig. 7). This map shows the relative stability of the banks of the LMR.

Fig. 7
figure 7

Riverbank vulnerability zonation map of the LMR

Full size image

River width

Table 6 shows the average width of the LMR in five different years. From the calculated statistics, it is evident that the channel width of the LMR increased consistently during the study period. The average channel width of the river enhanced from 8.7 km in 1988 to 13 km in 2017, a substantial increase of about 49%. Over this period, the river widened at a very high average rate of 148 m/yr.

Table 6 Channel width and the braiding index (BI) statistics of the LMR for the seven years
Full size table

Braiding index

The BI values for the LMR for the seven selected years also exhibited a significant rise, from 2.2 in 1988 to 6.4 in 2017 (Table 6) because more bars and/or islands appeared in the river. There was no significant difference between the BIs calculated from the Brice (1964) and the Germanoski and Schumm (1993) methods. This implies that no single big island or bar affected the BI calculation using the Germanoski and Schumm (1993) method.

Discussion

The study revealed that the banks of the LMR are extremely dynamic, indeed so much so that accretion may switch to erosion within a very short period of time. Both the rate and nature of the bank movement exhibited extensive spatial and seasonal variability. Erosion usually occurred in the monsoon season (Wilson and Goodbred jr. 2015), when both banks experienced retreat. However, the erosion rate of the right bank was nearly 34% greater than that of the left bank, forcing the LMR to gradually migrate west. So, at least over the study period, erosion and accretion in the study area were not balanced.

The high dynamicity of the LMR, i.e., rapid bank erosion and accretion and channel migration, could be linked to several causes. Factors like tectonic setup, channel morphological changes (e.g., widening of the channel and formation of new islands and bars), engineering properties of the bank sediments, water and sediment discharges, climate change, and anthropogenic impacts could all play a role to different degrees.

The active tectonic setting of the study area, and of the Bengal Basin as a whole, could be one of the main causes of the higher erosion rates of the right bank compared with the left bank and the consequential westward migration of the LMR. The elevation profiles across the Meghna floodplain reveal that the eastern side of the LMR is slightly more elevated than the western side (Fig. 8, profiles BB′ and CC′); the westerly topography is flat, whereas the eastern side shows a gradual eastward rise. This rise is associated with the active subduction of the Burma Arc on the eastern side of the Bengal Basin, allowing the GBM Delta to form a huge accretionary prism and fold belt (Alam et al. 2003; Steckler et al. 2008; Steckler et al. 2016) (Fig. 2b). The deformation front of the active fold belt is blind, buried by the thick sediments of the delta (Nielsen et al. 2004; Steckler et al. 2016). Again, the fold belt is detached over a shallow dipping basal megathrust, where the eastern side is considered to be upthrown (Steckler et al. 2016). The LMR follows the buried thrust front of the fold belt (Small et al. 2009) (Fig. 8). A slight upliftment of the western part of the Meghna floodplain, the Tippera Surface (TS), relative to the adjacent areas was also noted by Morgan and McIntire (1959) (Fig. 8). They pointed out that the subsidence in the major deltaic basin to the southwest and the Sylhet Basin to the northeast is counterbalanced by the uplifting of the Barind to the northwest and the TS to the southeast. This uplifting of the eastern side (or TS) of the LMR perhaps deviated its course to move westward, making the right bank more vulnerable to erosion. The westward diversion of the distributaries of the LMR almost at right angles to its course, opposite the TS (in the Hijla and Mehendiganj areas), even though the general trend of rivers in the GBM Delta is north-south, further supports this observation (Fig. 8).

Fig. 8
figure 8

Topographic map around the LMR. Eastern side shows geomorphic subdivision after Bakr (1977) and a more detailed classification by Rashid (1991) is given in brackets. Tectonic boundaries are indicated by heavy black lines with ticks on the upthrown sides of the thrust (left). Elevation profiles along three sections (AA′, BB′, CC′) in the topographic map (right)

Full size image

Bakr (1977), who called the TS the Chandina Deltaic Plain (CDP), reiterated its small uplifting relative to the surrounding floodplain (Fig. 8). In addition to this, he pointed out that the westward convexity of the LMR is analogous to the trend of the Tripura Fold Belt, which, in turn, pushes the flow westward. According to him, the sediments in the CDP are of Early Recent Age, slightly oxidized and more compacted than the recent floodplain deposits characterizing the older Meghna River. This implies that, in the past, the LMR flowed to the Bay of Bengal more to the east, along the foothills of Lalmai, than its present course. The present-day Titas River might be a remnant of the old Meghna River. Most probably, a megathrust earthquake has avulsed the Meghna River by 20–40 km west. It is also documented that the avulsion of the Brahmaputra to the present Jamuna in the late eighteenth century induced a series of changes in the morphology of the LMR; the course on the eastern side of Bhola Island was abandoned, which shifted the flow to the Shahbajpur Channel (Sarker et al. 2011). This allowed a very large land mass to accrete to the east (south of Noakhali; Fig. 7, Zone L7), eventually migrating the LMR towards the west. All of these findings reinforce the hypothesis of tectonic influence on the LMR dynamics.

Widening of the river plays an important role in regulating bankline change patterns. Erosion along both banks of the LMR caused concurrent channel widening; this may arise from a number of environmental factors, such as the changes in annual runoff, the occurrence of high-magnitude floods (Dewan et al. 2017), and an increase in bar/island area (i.e., the BI) (Goswami et al. 1999; Sarker 2004). Table 6 shows that channel widening was accompanied by an increase in the BI of the LMR. Enlargement of the existing bars/islands or the development of new ones diverted the river flow and exposed new sites to erosion. For example, the development of bars in Zone L4 (the Raipur area) pushed the river flow westward, leading to a large land loss in Zone R4 (the Mehendiganj area) (Fig. 7); the presence of these new/enlarged bars/islands, reflected in the increased BI, led to excessive erosion along both banks.

Riverbank stability may also be linked to the composition of the bank materials (Ikeda 1989). The LMR banks consist mostly of stratified fine-grained particles with varying spatial cohesion (de Jong 2000). They are usually of recent age (Bakr 1977) and weakly consolidated. Field observations in the study area revealed that bank failure is a common phenomenon, being a function of the hydraulic characteristics of the flow as well as the engineering properties of the bank sediments. Phenomena such as shear failure, cantilever failure, shallow slides, wave-induced erosion, liquefaction, and flowage of bank materials are common mechanisms in bank failure in the study area (Fig. 9a–g). These diverse mechanisms of bank failure suggest that differences in the engineering properties of the bank materials influence the relative stability of the riverbank (Thorne and John 1979); for example, shear failure occurs due to different cohesion between alternating layers. However, a detailed analysis of the engineering properties of bank soils in the study area and their corresponding spatial variability is necessary to evaluate the role of bank materials in bank stability; this is beyond the scope of this study.

Fig. 9
figure 9

Bank conditions along the LMR. (a) Shallow slides at Mehendiganj. (b) Shear failure at Harinathpur, Hijla. (c) Slab failure at Burhanuddin. (d) Cantilever failure due to the erosion at Mehendiganj. (e) Flowage of bank materials due to liquefaction near Bhola Sadar. (f) Erosion due to wave action at Hanar Char, Chandpur. (g) Step-like erosion in different river discharge conditions near Bhola Sadar. (h) Bank protection measure at Haimchar, Chandpur (source: field survey, 2018)

Full size image

Fluctuation in water discharge is one of the major contributing factors to the hydraulic changes in river geometry (Merritt and Wohl 2003). The discharge into the Meghna Estuary varies from less than 10,000 m3/s during the dry season to 50,000–100,000 m3/s during the monsoon (Wilson and Goodbred jr. 2015). This large seasonal variation in river discharge may have influenced bank erosion in the LMR. Such a direct relationship between river discharge and bank erosion has also been observed in other large rivers worldwide (Kummu et al. 2008; Ahmed and Fawzi 2011; Yao et al. 2011; Dewan et al. 2017). Furthermore, the LMR has periodically experienced major floods in the past decades (such as in 1974, 1987, 1988, 2004, and 2007). Such flooding events can trigger dramatic and sudden changes in riverbank movement (Yao et al. 2011), as observed in many other rivers elsewhere (Winterbottom 2000; Sharma 2005). During field interviews, local people stated that the high flow velocities during the floods accelerated bank erosion in the study area.

About one-fifth to one-third of the annual sediment content of around 1 billion tons that retain in the Meghna Estuary potentially provide material for land formation in the central part of the coast of Bangladesh (de Wilde 2011; Hussain et al. 2013). The LMR is characterized by high fluctuations in suspended-sediment concentration, ranging from 0.5 g/l to 9 g/l; this varies both seasonally and over spring-neap tidal cycles (Sokolewicz et al. 2008; Wilson and Goodbred jr. 2015). In addition, historical records provide evidence of the occurrence of several high-magnitude earthquakes in Bangladesh and its surrounding region, such as the 1897 Great Indian earthquake (M = 8.6), the 1918 Srimangal earthquake (M = 7.6), and the 1950 Assam earthquake (M = 8.6) (Fig. 2a). It is also likely that a high-magnitude earthquake will ravage the area again (Steckler et al. 2016). The LMR is believed to have been affected by sediment pulses from these seismic events (Sarker and Thorne 2006; Brammer 2014). The high sediment supply, fluctuation in suspended-sediment load, and an excessive sudden sediment pulse from an earthquake may lead to the modification of the river morphology (Brandt 2000; Kummu et al. 2008). However, the lack of sediment load data precludes confirming the extent of its role in riverbank changes.

Due to its low elevation, flat terrain, and dense population, the GBM Delta is likely to be one of the worst sufferers from climate change (Ahmed 2006). The upstream water and sediment inputs of the delta are strongly influenced by climatic forces (Xu et al. 2007). Higher rainfall and temperatures as a result of climate change may significantly increase water and sediment fluxes in the delta (Sarker et al. 2011). As an aftermath of very wet monsoons during the early Holocene, the rate of sedimentation in the GBM Delta was two and a half times greater than that currently (Goodbred and Kuehl 2000). These variations in water and sediment inputs are likely to have had significant effects on the morphology of the LMR. However, the extent of the impact of climate change on the bank dynamics of the LMR is still uncertain, and we did not delve much into it in this research.

Human activity has greatly affected the morphology and flow dynamics of most of the large rivers in the world (Gregory 2006). The LMR is not an exception; a considerable length of both banks has been stabilized using bank protection structures (Fig. 9h). For instance, several kilometers of the left bank from Chandpur to Haimchar, the right bank from Bhola Sadar to Daulatkhan, and places in the Hijla region have been protected by engineering works. These bank protection measures have decreased erosion rates to some extent, although they have already been damaged in some places by the high flow velocity of the river. As well, people in erosion-prone areas usually cut down trees, afraid of losing them to the river, and do not replant. This weakens the strength of the riverbank, making it more vulnerable to erosion. Local people also told abouthow waves generated by naval ships accelerate bank erosion. However, we do not have sufficient data to confirm the extent of the human impacts on bank erosion.

This study has several strengths but also some notable limitations. Among the strengths is the application of the DSAS tool in bank change analysis, which allows reliable delineation of bank movement as well as determination of the accuracy of the measurements through the generation of a number of statistical parameters. Apart from this, field visits coupled with a questionnaire survey at several sites enabled validation of the rates of bankline shifting from the DSAS. A detailed visual analysis of bankline change in every zone along the LMR helped explain any apparent anomalies in the DSAS results. Among the limitations, the unavailability of tidal data at the exact time of image acquisition prevented us from applying tidal corrections, although all images were from the dry season to minimize tidal effects. The times at which bank protection structures were installed were not incorporated in the analysis; in some places, these might have influenced bank change measurements. However, this influence is probably not great, as not many locations have been structurally protected, and the installations did not extend very far. Data on the hydrological characteristics of the LMR and on the engineering properties of its banks were not available. Nonetheless, the precision of the measurements still appears to be reasonable, and the DSAS, in association with field investigation and visual interpretation, to be a useful approach in assessing bank dynamics of a large river like the LMR. Comparing the DSAS outcomes with other bank assessment methods will stimulate the acceptability of the DSAS-based analysis.

Conclusions

In the present study, we used the Digital Shoreline Analysis System (DSAS) to evaluate the bank dynamics of the Lower Meghna River (LMR) from multiple bankline positions between 1988 and 2017. Both banks of the LMR have changed their positions rapidly at different rates, showing the dominance of erosion, with the right bank eroding at a 34% higher rate than the left, forcing the river to migrate towards the west. Based on average bankline change rates, the right bank is divided into six erosion and two accretion zones, and the left bank into four erosion and three accretion zones.

The erosion rates along the banks of the LMR were correlated with the channel widening and an increase in the braiding index (BI). The average channel width increased by approximately 49% over the study period, and the BI from 2.2 to 6.4. Based on these outcomes and other findings in the literature, we hypothesized that the active tectonics and interseismic deformation that led to the westward tilting of the leading edge of the Burma plate is one of the main factors in the higher erosion rates of the right bank than of the left bank and the consequential westward migration of the LMR. Additionally, large water and sediment discharges, variations in the properties of bank material, climate change, and anthropogenic activities also likely contribute to the high dynamicity of the river to different extent.

Although there are some uncertainties in the DSAS-generated results, this method provides a useful tool for assessing bank dynamics of a large river like the LMR if supplemented by field investigation and visual assessment of bankline positions. The results obtained from this study should aid authorities in taking zone-wise bank protection measures and facilitate planning to mitigate environmental and socioeconomic problems. However, a detailed analysis evaluating the hydrologic properties of the river, the geological and geotechnical characteristics of its banks, and the role of tectonics and anthropogenic activities on bank dynamics could be considered in further research.