QUANTIFYING RIVERBANK EROSION AND ASSOCIATED LAND COVER CHANGES ACROSS THE COASTAL UPAZILA OF HIZLA, BANGLADESH: AN INTEGRATED APPROACH USING GIS AND MACHINE LEARNING

ABSTRACT


INTRODUCTION
Bangladesh is one of the world's largest deltaic plains, with about 405 rivers spreading throughout the country.More than seven percent of its lands are occupied by river systems (Billah, 2018).The huge sediment loads carried by three major rivers, i.e., Ganges, Brahmaputra, and Meghna (GBM), have formed the floodplain of this country (Chen et al., 2005;Goodbred and Kuehl, 2000;Raff et al., 2023).Generally, due to rivers' lateral and vertical mobility, floodplains are subject to seasonal and cyclical changes, which also affect land-use patterns (Hazarika et al., 2015a).A fluvial system's two primary geomorphic processes are river bank erosion and accretion, driven by natural events such as flooding, sedimentation, high energy current and anthropogenic influences such as land use patterns (Lane and Richards, 1997;Surian, 1999).
In addition, the spatiotemporal changes in channel shape also depend on numerous factors, including physical, chemical, and geotechnical properties of bank materials, channel configuration, discharge and sediment load, river stage variation, and vegetation cover (Bartley et al. 2008;Julian and Torres 2006;Simon and Collison 2002).From a broader perspective, bank erosion, and accretion play the most crucial role in the dynamics of the overall fluvial system and influence the morphological changes of a floodplain (Mukherjee et al., 2017).Being a part of natural hazards, bank erosion causes the migration of riparian people and the destruction of hydraulic infrastructure and transportation networks (Islam et al., 2014).On the other hand, accretion reduces sediment transportation capacity and impedes navigation (Bizzi and Lerner 2015).In addition, new alluvial land will be created that can be used for agricultural purposes.
The existing LULC of a region is a cumulative output of the interactions between natural as well as anthropogenic variables and processes (Lambin and Geist 2006).Changes in LULC due to erosion and deposition are obvious phenomena in any floodplain, making it the most endangered ecosystem worldwide (Parveen et al., 2018).When the fertile topsoil is removed due to erosion activities, the land loses its agricultural production capacity and becomes barren.Additionally, periodic land cover change owing to bank erosion leaves negative imprints on people's livelihoods, which over time influences their economic, social, and psychological suffering.Hence, detecting temporal variations in LULC patterns is necessary to understand better the environmental dynamics in response to the morphological changes in the river systems.
Upazilas of Bangladesh are directly affected by extensive bank erosion (Rahman, 2010).The LMR downstream stretch of the GBM river system is highly dynamic and undergoes notable morphological changes.The LMR bisects Hizla upazila in Barishal district.People in this area tend to reside near its banks and engage in intensive agricultural activities due to the floodplain's remarkable fertility and seasonal abundance of water.Thus, this region's socioeconomic and environmental conditions are primarily threatened by bank instability.Regarding socioeconomic vulnerability, Hizla Upazila is the most susceptible coastal region of Bangladesh (Jahan et al., 2017).
Worldwide, numerous studies have been carried out to investigate the influence of bank dynamics on land cover change (Chakraborty and Saha, 2022;Hajra et al., 2017;Hazarika et al. 2015b;Masria et al., 2015;Rahman et al., 2022).In Bangladesh, these studies mainly focused on the Brahmaputra and Padma (Debnath et al., 2023;Alam et al., 2023;Arefin et al., 2021) Rivers, situated in the Northern and Northwestern parts.A few researchers have also addressed the phenomenon in the Meghna estuary along the Bay of Bengal (Islam et al., 2021;Kabir et al., 2020).Nonetheless, little attention has been paid to the lower part of the Meghna River.Though assessed bank dynamics and evaluated the future impact of land use change on the ecosystem across LMR, the impact of bank dynamics on land cover change is yet to be done (Mahmud et al., 2020;Hoque et al., 2020).In our focused area Hizla Upazila, issues regarding the impact of storm surges, water quality appraisal, and the biodiversity of fish species have been addressed (Jahan et al., 2017;Islam et al., 2023;Hossain et al., 2012).However, these investigations must incorporate the effects of riverbank erosion on land use patterns to identify the area's vulnerability and risk levels.Thus, monitoring the spatiotemporal dynamics of riverbanks and their influence on land use changes is essential for ensuring long-term sustainability and economic development in this region.
Remote sensing and Geographic Information System (GIS) technique is broadly used for detecting spatiotemporal changes in river morphology and land use (Hassan et al., 2017;Lovric and Tosic, 2016;Rahman and Tripathi 2013;Lo and Shipman 1990;Michalak 1993;Pijanowski et al., 2002).Remotely sensed satellite imageries like Landsat, sentinel 2, and Moderate Resolution Imaging Spectroradiometer (MODIS) data are frequently used as they have broad area coverage, synoptic perspective, and regular data.Time series analysis is helpful in identifying the rate and pattern of erosion accretion and its influence on land use change.Multi-temporal analysis of an area requires dealing with large amounts of data; a major concerning factor as large data size increases the possibility of inaccuracy and is tough to handle.In this regard, the GEE cloud-based platform is incredibly beneficial, considering the ability to analyse substantial spatial regions using high-performance computing resources without downloading (Ghosh et al., 2014).
Besides, advanced machine learning and deep learning algorithms are becoming increasingly popular because of their improved accuracy and automated real-time categorization processes that enable quick change detection (van Leeuwen et al., 2020;Roy, 2021).To address classification issues, trained supervised learning techniques, namely Artificial Neural Networks, Support Vector Regression, and Random Forest (RF), are mainly adopted.In this study RF classifier is used as it has high processing power for unwanted data.The present study estimates the rate and pattern of erosion and accretion of Hizla Upazila and its impact on land use.In addition to improving floodplain management, this study sheds light on the various riverine risks that locals in this area might encounter as a result of the fluvial dynamics of the LMR.

Study Area
The Hizla Upazila is located in the southern half of Bangladesh between 22°50' and 23°05' north latitudes along with 90°25' and 90°43' east longitudes (Figure 1).This Upazila occupies an area of 515.36 km 2 with a population of 146077.Most of the part of this region is delta type as the mighty Meghna River flows through it and is crisscrossed by several distributaries.The lower part of the Meghna River is formed by the confluence of the Ganges, Brahmaputra, and Upper Meghna River (known as the GBM river system) and falls into the Bay of Bengal.Six Unions (Harinathpur, Memania, Hizla Gaurabdi, Guabaria, Barajalia, and Dhulkhola) of Hizla Upazila reside on both sides of the Lower Meghna River.Because the high load of sediment carried by the GBM river system makes this region unstable and subject to frequent erosion and accretion, the area is also sensitive to several natural hazards, namely floods, cyclones, storm surges, and land use change.As a result, the economy of this region is greatly hampered.Close monitoring is needed to ensure sustainability in this region, which was the prime reason for selecting the study area.

Data Collection and Preprocessing
Six landsat images covering the period ranging from 1989 to 2020 were used in this present study, selected Landsat 5 Thematic Mapper (TM) and Landsat 8 OLI (Operational Land Manager)-TIRS (Thermal Infrared) sensor images based on satellite image quality and availability.For erosion and accretion assessment, Cloud-free Landsat imagery (Table 1) was downloaded from Global Visualization (Glovis) of the United States Geological Survey (USGS) in the years 1989, 1997, 2004, 2010, 2015, and 2020.The study area was delineated according to the Local Government Engineering Department (LGED) Upazila boundary map, followed by georeferencing and converting into vector data.However, Landsat TM and OLI data processing was conducted on the GEE platform for land use classification.A loud screening algorithm with <5% cloud cover was applied to ensure cloud-free optimal imagery for each year.The overall methodology of this research is summarized in a flow chart (Figure 2).

Total Erosion-Accretion Calculation
Several methods can be adopted for extracting water information from remote sensing imagery, like the single-band and multi-band methods, according to the number of bands used.However, the data from a single band may lead to an over or under-estimation due to mixing up with shadow and noise.Among the multi-band methods, the MNDWI is used to extract water body information more easily, quickly, and accurately than other classification methods (Li et al., 2013).The MNDWI (Xu 2007) for TM and OLI images is calculated as: MNDWI = (Green -SWIR) / (Green + SWIR) For Landsat 7 data, MNDWI = (Band 2 -Band 5) / (Band 2 + Band 5) For Landsat 8 data, MNDWI = (Band 3 -Band 6) / (Band 3 + Band 6) The Classified image was then converted into vector format, and the erosion and accretion were measured by comparing the vector data from different periods.

Land-use Classification and Change Detection
The LULC maps of the study area were prepared based on five major landuse classes: water bodies, bare land, vegetation, and settlement, considering our research objectives.Here, the term "water bodies" refers to both major rivers and other permanent bodies of water; "bare land" indicates sandy terrain and barren lands adjacent to rivers; "vegetation" refers to both natural forest and agricultural land; and "settlements" refers to places with built-up features.While preparing the map in GEE, the Random Forest (RF) algorithm was selected as the classifier because it conveys greater processing power for noise and overfitting.Furthermore, RF usually provides higher accuracy even though it handles complex data of large dimensions.It generates multiple decision trees and classifies a dataset based on the prediction modes for all decision trees.Training samples were collected through visual interpretation and field observations for classification purposes.Field data was collected randomly, maintaining at least 500m distance to minimize spatial homogeneity, and the other reference samples were derived from the highresolution image of the Google Earth map in different years.Annually, 500 training points were used for LULC classification in four classes.

Accuracy Assessment
The post-classification accuracy assessment is essential to quantitatively evaluate how well the pixels were sampled into the appropriate land cover classes.The accuracy of the classification algorithm has been evaluated in many recent studies using kappa coefficient (K) based indices, an articulate method of analyzing a single error matrix and comparing the differences between different error matrices.The degree of agreement between classified images and the ground truth is measured using five levels: bad or very poor, fair, good, very good, and excellent, which correspond to values less than 0.4, 0.4 to 0.55, 0.55 to 0.70, 0.70 to 0.85, and greater than 0.85, respectively.In this study, 100 random points were generated for each year in the classified image of the study area, and the points were then validated using the hydrological map, topographic map, Google Earth, and Google Map as reference sources.

RESULTS
The rate of erosion-accretion and changes of the LULC in the study region are summarized below.To clearly articulate the findings, and differences, this section provides an in-depth, detailed quantitative analysis of the observed changes.It is structured into two distinct sub-sections, each dedicated to a specific verdict and its corresponding visual representation.Firstly, the information related to fluctuations and trends observed in erosional and accretional activity for the past 31 years due to river dynamics.Secondly, the information related to changing, converting, and validating LULC for different study years.

Erosion-Accretion Pattern
Six satellite images (1989, 1997, 2004, 2010, 2015, and 2020) have been studied to investigate the morphological changes in the Meghna River for the targeted zone employing remote sensing techniques.Figure 3 illustrates the overall scenario and span of erosion and accretion in Hizla Upazila.Table 2 shows the summarized statistics where the average erosion and accretion rate was 5.03 km 2 and 5.72 km 2 over the period.The accretion rate of this region was even higher from 2004 to 2010, measured at 8.64 km 2 per year.During this period, the north-eastern and southeastern bars expanded significantly, forming 27.84 km 2 of land suitable for settlement and agriculture.In addition, a gain of 10.74 km 2 was observed between 2015 and 2020, meanwhile, the newly created land in previous years became stable.Contrarily, the years 1997 to 2004 saw substantial erosion, especially in the central region of Hizla Upazila, which comprises Harinathpur, Hizla Gaurabdi, and Memanina Union.The highest reported land loss from erosion occurred during this event (Figure 4), totaling about 8.73 km 2 , and it coincided with the total submergence of a newly formed bar in the Eastern part.Another noteworthy period of erosion occurred from 2010 to 2015 inducing a land loss of 6.87 km 2 , predominately damaging the eastern portion of Hizla Gaurabdi Union.A high degree of instability, mainly driven by the Lower Meghna River, is apparent in the studied area's erosion-accretion pattern (Figure 4).These fluctuations can be attributed to natural factors such as water discharge, velocity, sediment volume, and climatic variations.Over the 31-year study period, total river bank erosion was found to be less than the total amount of accretion, resulting in a net gain of approximately 24.93 km 2 of land (Table 2).    3 contains the quantitative analysis of all the classes.Over the past 31 years, bare land areas have expanded by about 20.31 km 2 at an average rate of 0.66 km 2 /year, while waterbodies have declined by 19.1 km 2 at a mean annual rate of 0.61 km 2 (Figure 6).This change can be attributed to the deposition of a vast sediment load originating from upstream sources and contributed to the formation of new land, mainly in the eastern part.In another view, it is evident that the extent of bare land has increased by approximately 53.13% from 1989 to 2020 (Table 3); however, the corresponding growth in vegetation cover was only around 0.78%.
The significant erosion and subsequent accretion processes occurring in this area (Figure 4) account for this disparity.A major portion of land consisting of vegetational cover and settlements was eroded from the central and western parts of the study area (Figure 3).Therefore, people residing in these areas were forced to relocate.As the newly formed land in the eastern part settled, people began to inhabit the area and engage in agricultural activities (Figure 5).The net change is relatively small (0.02 km 2 ), as the total gain and subsidence loss area are roughly equivalent (Table 3).Similarly, vegetation cover is the prominent land type and experienced a negligible net change, losing 1.19 square km.This approach highlights how the river's dynamic nature affects land-use change.

Validation of Classified LULC
Classified images should also be checked for accuracy before input to an application.The performance of the LULC model concerning a given category or class can be evaluated by considering individual accuracy assessment parameters, and in this study confusion matrix was utilized.
The results of the accuracy assessment, including the Kappa coefficient, User accuracy, and Producer accuracy are displayed in Table 4.The classified images displayed a kappa coefficient of 86%, 93%, 89%, 89%, 87%, and 90% in 1989, 1997, 2004, 2010, 2015, and 2020.Across all those years, the Kappa values exceeded the threshold of 0.85, demonstrating strong agreement between the observed and reference land cover classifications.User accuracy, or the percentage of correctly identified pixels according to the user's assessment, ranged from 96% to 100% for water, 92% to 96% for vegetation, 80% to 88% for settlement, and 84% to 100% for bare land.Moreover, the measurement of how many reference pixels were identified correctly by the classification process denoted as producer accuracy was between 89% to 100% for water, 86% to 96% for vegetation, 81% to 89% for bare land, and 100% for settlements.These findings exhibit the land cover classes' reliability and the methodology's strength and dependability, confirming the data's appropriateness for further research and decision-making.1989, 1997, 2004, 2010, 2015, and 2020)

Validation of Classified LULC
Classified images should also be checked for accuracy before input to an application.The performance of the LULC model concerning a given category or class can be evaluated by considering individual accuracy assessment parameters, and in this study confusion matrix was utilized.
The results of the accuracy assessment, including the Kappa coefficient, User accuracy, and Producer accuracy are displayed in Table 4.The classified images displayed a kappa coefficient of 86%, 93%, 89%, 89%, 87%, and 90% in 1989, 1997, 2004, 2010, 2015, and 2020.Across all those years, the Kappa values exceeded the threshold of 0.85, demonstrating strong agreement between the observed and reference land cover classifications.User accuracy, or the percentage of correctly identified pixels according to the user's assessment, ranged from 96% to 100% for water, 92% to 96% for vegetation, 80% to 88% for settlement, and 84% to 100% for bare land.Moreover, the measurement of how many reference pixels were identified correctly by the classification process denoted as producer accuracy was between 89% to 100% for water, 86% to 96% for vegetation, 81% to 89% for bare land, and 100% for settlements.These findings exhibit the land cover classes' reliability and the methodology's strength and dependability, confirming the data's appropriateness for further research and decision-making.

The Conversion of LULC Categories
Based on the post-classification methods, the change matrix was obtained for 1989 to 1997, 1997 to 2004, 2004 to 2010, 2010 to 2015, and 2015 to 2020, with a relative significance of the major LULC conversions shown in Figure 7.This calculation indicated that a significant portion of the vegetated area was acquired by converting mostly the existing bare lands and a minor part of the settlement area.In contrast, a considerable amount of the vegetative area also converted to bare land at a higher rate from 2015 to 2020 than from 1989 to 1997.The expansion of settlement areas has predominantly involved the replacement of vegetative areas and also a smaller proportion of bare land areas.In this model, the land area turning into waterbodies was regarded as land loss as opposed to land gain, conversion of water bodies into any other classes.The results showed a close correspondence between the observed land gain and loss patterns and the calculated erosion accretion along the river bank, though there were a few minor differences that could be attributed to factors like landfills in inland water areas, errors in the LULC mapping, and limitations in the erosion accretion modelling.A noticeable decline in lake-type wetlands was observed from 1989 to 1997 (Figure 5), suggesting that these areas have been converted to human settlements and agricultural activities.

DISCUSSION ON KEY FINDINGS BASED ON PREVIOUS LITERATURE
Over time, Hizla Upazila has continuously transformed due to the downstream dynamics of the Meghna River, leading to a substantial change in land use and land cover (LULC) within this river floodplain.The shifting and broadening of channels are closely connected to erosion and deposition processes along riverbanks.Throughout the study period, accretion consistently exceeded erosion.A notable occurrence was rapid erosion on the western side and the emergence of new land on the eastern side due to tectonic factors.Settlement and vegetational areas underwent abrupt changes due to catastrophic flooding from heavy rains and the downstream flow of excess river water from the upstream area.

Influence of Tectonic Activity on Erosion-Accretion Pattern:
The noteworthy shift of additional land formation in the north-eastern and south-eastern regions is a significant change that occurred throughout the entire period from 1989 to 2020 (Figure 5).This process began in 1989, leading to the establishment of two substantial deltas in these areas.The progression is evident on the land use change map (Figure 7-8), illustrating the stabilization of initially formed bare land followed by subsequent vegetation growth.Over time, communities migrated to these two deltas, capitalizing on favorable conditions for settlement and cultivation.In contrast, the central and western parts of the study area experienced erosion over time, causing resident displacement and agricultural land devastation.This phenomenon could be explained by the gradual westward migration of the LMR due to tectonic influences.Previous studies noted that the erosion rate of LMR is higher on the right bank than the left bank, initiating westward river movement (Mahmud et al., 2020).
Elevation profiles across the Meghna floodplain (Figure 9) reveal a slightly higher elevation on the eastern side of LMR than on the western side.This elevation is linked to the active subduction of the Burma Arc, resulting in the formation of a fold belt and significant accretionary prism in the GBM Delta (Small et al., 2009).The fold belt is situated above the underlying shallow-dipping basal megathrust, where the eastern side of the thrust belt is considered upthrown (Steckler et al., 2016).The LMR intersects the fault line, and the upliftment of the eastern side diverted its course westward, rendering the right bank more susceptible to erosion.

Impact of Extreme Geological Events
On the other hand, the total land area has notably increased over the years, possibly due to sedimentation in this region.An exception is observed between 1997 and 2004 when the land area decreased (Figure 10).Based on climate data, this decrease is assumed to result from rapid erosion caused by floods or high-intensity rainfall during that period.Severe earthquakes in the Himalayan region triggered significant landslides in Nepal, India, and the Kashmir region in 1950, 1963, 1999, 2005, and 2015(Parkash 2013;Valagussa et al., 2021).A considerable amount of avalanche debris flowed into GBM rivers through various tributaries.Since the fine sediment fraction primarily comprises clay and silt, it moves swiftly through the GBM basin rivers.The deposition of these sediments as they enter low-energy waters tends to alter river morphology.
Furthermore, severe floods occurred in 1998 (68% area of Bangladesh inundated), 2004 (38%), and 2007 (42%) (Hossain et al., 2014).Intense rainfall exceeding averages during monsoons triggers extensive landslides in the upstream region, delivering substantial sediment volumes (Tanvir et al., 2017).The combined impact of earthquakes and floods increased sediment accretion along the confluence and estuary of the Meghna River from 2004 to 2010 and after 2015.This depositional event led to an expansion of bare land area starting in 1997, peaking in 2004 and 2010 (Figure 10).Certain researchers characterized river responses through broadening, enhanced eroding capacity, and migrating sand waves following major depositional events.This explains the dominant role of erosion, contributing to the reduction in bare land area between 2010 and 2020 (Figure 10).

Factors Contributing Land Cover Change
The GIS-based spatial analysis also revealed patterns of land-type change during the study period.As shown in Figure 10, there is no significant overall change in settlements and vegetation.However, these two landcover classes displayed the most unstable conditions over the years, and their changes are quite similar.After a sharp decline from 1997 to 2004, both settlement and vegetation areas gradually expanded until 2015.
While the settlement area remained stable from 2004 to 2010, vegetational cover increased.Both areas started decreasing at a steady rate after 2015.Natural catastrophes, including cyclones, storm surges, and floods, frequently hit Bangladesh's coastal region, significantly impacting the local population, and often resulting in forced relocations.The central coastal region of Bangladesh was severely affected by notable cyclones in May 1997, November 1998, Sidr in November 2007(Aila, 2009;Agrawala and Van Aalst 2008;Saha, 2015).These cyclones played a vital role in the fluctuation of settlement areas.According to census data, in 1991, the central coastal region had a population of 166,265.By 2004, the population had slightly increased to 174,508, but then declined to 146,077 in 2011.The population size variation aligns consistently with the quantification of settlement areas and shows a similar trend.The powerful cyclones mentioned earlier, along with their effects on the area, and the destruction of agricultural land and loss of livelihood due to frequent flooding, could account for this population decline in this region.
Hizla Upazila is a vulnerable area, and extensive bank dynamics, significantly impacting the socio-economic situation of the region, are believed to be the primary reason for this vulnerability.The newly accreted land (sand char), due to its low fertility and limited use, has poor economic value and falls short in compensating for the loss of native land.Riverbank erosion damages large trees and adversely affects the ecosystem, as it takes years for the char region to develop the necessary soil layers for tree planting.The uncertain nature of farmers' livelihoods, causing wealthier and middle-income farmers to become marginalized or landless, leads to a social crisis.Due to erosion, this area needs more developmental and industrial facilities.Given its erosion-prone nature, population displacement is common as people prefer settling in areas with better access, improved educational and healthcare facilities, career prospects, and enhanced quality of life.Riverbank erosion is one of the most hazardous processes in the study area, with lasting impacts on land use patterns.Thus, this study offers insight into implementing appropriate policies and programs to enhance the livelihood situation in these vulnerable areas.

CONCLUSION
This study comprehensively evaluates spatial and chronological changes in the land area of Hizla Upazila, along with the adaptation of land-use and land-cover (LULC) types to land dynamics.Through detailed analysis of multi-temporal data, we observe that persistent erosion and accretion processes in the lower Meghna River consistently reshape this area, leading to substantial alterations in LULC.Findings reveal a dynamic pattern of land gain and loss across different time frames, with an average accretion rate of 5.72 km²/year exceeding the average erosion rate of 5.03 km²/year, resulting in a net land gain.The spatial distribution of erosional activity highlights predominant effects in the central and western parts of Hizla Upazila, while the eastern part witnesses the formation of new deltas over time.This is attributed to the westward movement of the lower Meghna River, rendering the central and western regions more susceptible to erosion.From 2004 to 2010, significant sediment deposition occurred due to extensive landslides upstream triggered by a powerful earthquake.This transported a substantial sediment load downstream.Dominant vegetation cover experienced a decline between 1997 and 2004, potentially due to notable flooding during that period.Conversely, the settlement area reached its minimum extent from 1997 to 2010, indicating that recurrent floods, cyclones, and heavy rainfall may necessitate population relocation.The study's accuracy was verified using User accuracy, Producer accuracy, and Kappa Coefficient.However, it lacks geotechnical analysis to assess land susceptibility to erosion, and it does not account for factors beyond riverbank erosion causing land-use changes.Therefore, additional research is needed before these findings can be applied to policy and planning.Despite its limitations, the study's methodology and analysis provide a basis for future research.Overall, this study enhances our understanding of how frequent erosion and accretion in the lower Meghna River impact land cover types, laying a foundation for informed decision-making and proactive measures in response to environmental changes in the region.

Figure 1 :
Figure 1: Location of the study area (a) The flow path of three major rivers (Padma, Meghna, and Jamuna) of Bangladesh; (b) Position of Hizla Upazila in relation to the Lower Meghna River (LMR) with surrounding districts; (c) All the unions of this Upazila

Figure 3 :
Figure 3: Spatial distribution and pattern of erosion and accretion for a timeline of 1989 to 2020

Figure 4 :
Figure 4: The bar chart represents total riverbank erosion and accretion with resultant gain/loss in different time periods ranging from 1989 to 20203.2Land-Use and Land-Cover ChangeRegarding land use classes, the target area is divided into water bodies, settlements, vegetation, and bare land.Land use is a dynamic phenomenon that changes due to the geo-natural process, variation in cropping patterns, seasonal changes, etc.The change in land use pattern is illustrated in Figure5, and Table3contains the quantitative analysis of all the classes.Over the past 31 years, bare land areas have expanded by about 20.31 km 2 at an average rate of 0.66 km 2 /year, while waterbodies have declined by 19.1 km 2 at a mean annual rate of 0.61 km 2 (Figure6).This change can be attributed to the deposition of a vast sediment load originating from upstream sources and contributed to the formation of new land, mainly in the eastern part.In another view, it is evident that the extent of bare land has increased by approximately 53.13% from 1989 to 2020 (Table3); however, the corresponding growth in vegetation cover was only around 0.78%.

Figure 5 :
Figure 5: Land use pattern of the study area for the years 1989, 1997, 2004, 2010, 2015, and 2020 calculated using Random Forest (RF) classifier for TM and OLI Landsat imagery in Google Earth Engine (GEE)

Figure 7 :
Figure 7: Interchange between major land cover classes in five periods where land converted into waterbodies was considered as erosion, water area converted into any of the classes marked as accretion, and the area remained into same classes is symbolized as no change

Figure 8 :
Figure 8: The relative percentage of land type conversion in different periods embodied by the grey color sheds

Figure 9 :
Figure9: Topography along the bank of Lower Meghna River (LMR) with lithologic boundary after(Rashid, 1991).The tectonic boundary is represented by a heavy black line with tick marks on the upthrown side and the position of Hizla Upazila in red color.Modified after(Mahmud et al., 2020)

Figure 10 :
Figure 10: Temporal variation of different land type areas (km 2 ) from 1989 to 2020

Table 1 :
Description of the Collected Landsat Imagery with Resolution and Cloud Cover https://glovis.usgs.gov/ Figure 2: Workflow chart of this research

Table 2 :
Total area (km 2 ) and rate of erosion-accretion between 1989 to 2020

Table 4 :
Overall Accuracy of The Land Use Classes Using Confusion Matrix and Kappa Coefficient